Automatic colouring book creation
Contents
1. CZECH TECHNICAL UNIVERSITY IN PRAGUE Faculty of Electrical Engineering Department of Cybernetics Bachelor Thesis AUTOMATIC COLOURING BOOK CREATION ON TOUCH DEVICES WITH OS ANDROID Stanislav Steidl Thesis advisor RNDr Daniel Prusa Ph D Study Programme Open Informatics Specialisation Computer and Information Science Prague 21 May 2015 Czech Technical University in Prague Faculty of Electrical Engineering Depa tme BACHELOR PROJECT ASSIGNMENT Student Stanislav Steidl Open Informatics Automatic Coloring Book Creation on Touch Devices with OS Android Guidelines The task is to design and implement an application for tablets supporting work with coloring books The application will support an automatic creation of a coloring book based on a chosen image The target user is a child of age 3 6 years s Analyze accuracy and relabilty of the method for various types of input images Utilize possibilities of touch devices Implement the application for OS Android Create a user manual and programmer documentation Bibliography Sources 1 Sonka M Hlav W Boyle R Image Process Thomson Learning Toronto Canada 2007 2 Mudelman G Android Design Patterns 1st edition Wiley 7 Analysis and Machine vision 3rd edition s USA 2013 Bachelor Project 5 Valid until the end of the summer semester of academic year 2015 2016 LS doc Dr Jan prof Ing Pavel Ripka CSc2. On the other hand the implementation of detection functions used will not be described here The description of implementation can be found in implementation chapter Application life cycle In the diagram bellow the grey colour represent actions which are automatically performed Yellow colour marks a centre of application Blue is for managing colouring pages Green is for simple tasks which are designed to be handled by children Red actions are meant for experienced users only inititation create colouring select source image e new colouring automatic creation save new colouring export page Figure 45 Life cycle graph 41 Project overview The project is divided into several packages for better readability Activities package This package contains all activities used during the application life cycle Each activity is responsible for event handling and layout when it is active Detection package This package contains classes which are the only ones that actually use OpenCV library functions They provide more suitable API or interface for this work and therefore guarantee better readability of code for programmers who are not familiar with OpenCV entities Manager package This package comprises managers which often use tools from detection or utility tools package to perform the required action An example of this is saving or loading colouring pages Utility tools packa
3. partme Dean Prague January 14 2015 Katedra kybernetiky ZAD N BAKAL SK PR CE Student Stanislav Steidl spesnost metody podle typu vstupn ch obr zku V n vrhu vlastn aplikace pln vyu ijte mo nost dotykov ho zarizeni Vse implementujte pro OS Android Vytvorte u ivatelskou can d Seznam odborn literatury 1 onka M Hlav W Boyle R Image Processing Analysis and Machine vision 3rd edition Thomson Learning Toronto Canada 2007 2 Mudelman G Android Design Patterns 1st edition Wiley Indianapolis USA 2013 Platnost zadani do konce letniho semestru 2015 2016 LS doc Dr Ing Jan Kybic prof Ing Pavel Ripka CSc vedouci katedry d kan V Praze dne 14 1 2015 Prohl en autora pr ce Prohla uji e jsem p edlo enou pr ci vypracoval samostatn a e jsem uvedl ve ker pou it informa n zdroje v souladu s Metodick m pokynem o dodr ov n etick ch princip p i p prav vysoko kolsk ch z v re n ch prac VPE MM LLL Podpis autora pr ce Abstract Main goal of this thesis is to study possibilities of implementation image processing methods namely edge detection and segmentation Main purpose is to automatically create colourings for children between 3 and 6 years The result implementation is presented as an OS Android based application for tablets The resulting application Colour it automatically generates colouring based
4. s EM vera a 21 Advariced cartoons detection usa adeps ode o a cai vua iea fola paa Vau aso TUR Pr adds 21 Photos Of cartoons detection uio tid repe ee NOE HD SERRE dh IER ERA M DEM YARN ERAT deed EUM EU ENS 21 Customized creation of coloUrings viii epo Corb eas Cer vp dos ae aS ap Aa ii eo Ep decas 22 Getune Palette P ada Dudes aa nte oO esu Me IND RK NE D EIE 22 Result rr av enitn AERA haut uomo tana dad s 23 Classic arar OL 23 Detection EVA ON aaa Aaa 24 SIMPle CaFLOONS ui ads 24 AOVaticed CAMION SN 26 PISTOL 29 Bore UB eire eee O A 31 COMPAlISON Ol classes ei 33 Comparison With Scrap COIN 35 DISCUS TO Meio sete S 38 Result evaluation tii 38 SPIE CI OON S ane m cu tu cm erne 38 Advanced eal tools dikes tmt hh dtl ae ence t ant ee e or c eRe 38 PROTOS OF CATO ONS e TEE 38 leo SARS SRP A T 38 Creation recommenda oN adeo e E E a 39 CreationsOr a simple co ou aeniea a a E A 39 Creation of an advanced colouring eese 39 Creation On expert ColoHlligS usina d eat auta a Drac be dean De den UE deem 39 CORSO dos daos im bersatuka czenas E 40 PUCE DEOISCE zz base sa aan noes ce dI Mb ta D MM EE LS SUE 40 Vii Possible ImtDOVelTielibs rl 40 POSSIDIe T ataca 40 ADDeHdICOS Gossip eruere i TEIL quae Ru b OPI INI T ees 41 DOS MERE HON eat RUN I S 41 PROP Ta Te TTT 41 PRO ECLOVEVIS W T 42 Programming DNNOSODIAY PRETI OE OE 43 Usermanual Colourit xcensie siete lr 44 COLOQUES SCh COIN A O O O 45 MOVANCEO ChE ATION o P t
5. to finish this thesis would also like to thank to Itseez team which leads the development of open source OpenCV library for all their hard work which enabled me to work on this thesis Last but not least would like to thank my family for all their support during my studies and work on this thesis Table of contents it OPCW se RR RO O 1 MOUV ee 1 GO sucess 2 THEO fe Heal back TOUN NUT Tm 3 BST ee p O 3 Scd Pu 5 means IU SERRA EB jara dois ie 6 TRAFICO aa iio 7 Canny e Te T e le TEE EE E 8 OES SN SS plo PPP 8 Slate OT LME ET SEE o E E EA TE E OE 9 Toddler Coloring book free Pr0 ccccccccccccccccccceecseceeesssesseeeecccceccsceseeeeeeeuuuueeeseseeeeeeeeeeeseeeess 9 CONE DOOK ABR acacia actos E m 10 BE FID COLO RR O V OT ETE 11 COMO p TR EE 12 MOC CICK CO OGING DOOR NR 12 Ha ego AE cat coterie E O 13 SOD A O O O EN 13 RECOMIMEn da PRO O ORT 13 aa VEIT lE ela MTS E S E 15 DetectionmgmplemelHtalloliue essetis tovt SEER REDE ata Ui dU EU 15 he SCARS SO So c buda MMC EE UEM iE MEME M MEI LU 15 SOUP CS FSCO RINNE Og e ES SEERE 18 PP mM en II ere 19 PRO 8 0 CAS bas 19 vi Colo rin class detectlolasxidexi edt ER ERR cada 20 Simple cartoons detect loli ziskat okudk
6. 7290 MN 137099 N 3 LA un gt o N e o cl A x LO 00 m o 3 8 NN Oo lt o E Ss e So N EN lt o D lt En lt M o m N Don AR ki E i l N nA A H ME IMAGE ANALYSIS DETECTION COLOURING SAVING FILE TOTAL CREATION m advanced cartoons m photos of cartoons m general photos m simple cartoons Figure 41 Time consumption comparison SIMPLE CARTOON ADVANCED CARTOONS Bi image analysis m detection E image analysis detection Bi creating colouring M saving file m creating colouring W saving file Figure 42 a b Time consumption of separate classes 34 PHOTOS OF CARTOONS GENERAL PHOTOS B image analysis Bi detection B image analysis Bi detection m creating colouring M saving file m creating colouring M saving file 2 5 Figure 43 a b Time consumption of separate classes Comparison with Scrap Coloring Examples from previous chapter were used to create colourings at Scrap Coloring as well And in this chapter the received results are compared There is no objective metric how to compare colourings and therefore comparison of results will be presented but not commented Scrap Coloring images were left in their original shape and size in respect to Colour it results P AA Y 7 PE A F L N e EZ L7 X d K d Y X l J A wa Ne SE m S VI CV Mol Nf i L RE ko S x 4 k 1 N bu e 1 N X a M X k VA A x R
7. Otsu Thresholding online 2010 accessed 13 5 2015 available at lt http www labbookpages co uk software imgProc otsuThreshold html gt 10 Braxmeier Steinberger GbR Pixabay online 2015 accessed 15 5 2015 available at lt http pixabay com gt 48 11 Elsa and Stephane Gigandet Scrap Coloring online 2010 accessed 8 5 2015 available at lt http scrapcoloring com gt 12 Google Inc Google Play online 2015 accessed 14 5 2015 available at lt https play google com store gt 13 O Sekora Kn ka Ferdy mravence 6 edition Albatros Prague Czech Republic 1962 ISBN 13 864 78 14 Z Miler H Dosko ilov Krtek a parapl ko 4 edition Albatros Prague Czech Republic 2002 ISBN 13 712 002 15 Arthur D and Vassilvitskii S 2006 How slow is the k means method ACM New York NY USA 144 153 16 Digital trends Inc Android claims 81 596 of the global smartphone OS market in 2014 iOS dips to 14 8 online 2015 accessed 15 5 2015 http www digitaltrends com mobile worldwide domination android and ios claim 96 of the smartphone os market in 2014 17 Wikipedia contributors Wikipedia The Free Encyclopaedia Determining the number of clusters in a data set online 5 5 2015 accessed 18 5 2015 lt http en wikipedia org w index php title Determining the number of clusters in a da ta set8 oldid 661006527 gt 18 Testing device ASUS MeMO Pad 8 ME181CX 16G
8. caused by the similarity of object and background colour in some places Photos of cartoons The results are highly dependent on the source quality It should be mentioned that even low resolution cameras can take good enough photos if the photos are taken with potential usage in mind In this case it means that the object should be a simple cartoon otherwise it will be most probably too noisy Real photos Nowadays the quality of average photos is far beyond the necessary level Therefore in this class the composition of the photo is more important than real density If the object in the photograph is well distinguished from the background and is not too complicated itself then the results should be reasonably useful It is important to mention that this is not the primary source for automatic creation It is also possible that an experienced user can achieve better results using custom creation 38 Creation recommendation Usually the difficulty of the source image corresponds to the results obtained However in many cases the image can be simplified during the process It should be mentioned that significantly simplified images can be gained using custom creation where the user can determine what is truly important for the particular image and his goals Creation of a simple colouring Our results demonstrate that images from the simple cartoon class or photos of simple cartoons should be recommended as a source for simple colour
9. changed its colour to yours In the middle is your colouring Just click to some area to fill it with selected colour Yes it is that simple rr uud i BAD S LT L 5 MM S Figure 52 colouring is really simple 45 Would you like to remind how the original looked like No problem just click the miniature of original image You can find it on the right side in the upper part Then click the miniature again to return to colouring You can find a menu which contains additional woe features just above the miniature From the menu you EN can Save current state Clear Colouring Open new Colouring 1 2 3 4 save current progress for later clear the whole colouring to start again open new colouring Or export the image it will save the colouring as a picture Then you can print it and colour with pencils Export as picture or share it with your dearest Figure 53 colouring menu Advanced creation Advanced creation screen allows you to unleash full potential of detection methods to enhance the detection of more sophisticated images There is a powerful toolbox prepared containing variety of methods and packed into simple control panel Figure 54 advanced creation example select source thick edges prolong edges set initial bounds so 8 v TI 100 Y T2 set colour count a E gt SED U t v K colours detect initial reduce Colo
10. hd gt y Figure 38 detection example Preikestolen rock 2 5 of 5 stars The landscape is detected considerably well The main issue of this image is the detection of tiny people and their shadows which make the rock look like being composed of noise only iip x Y d kad 6 LO O IS Ge EU A M Ng ow E TEE T A Figure 39 detection example 32 Flowers 2 of 5 stars The original image consists of lots of petals and so does the result However it is hard to divide which petal belongs to which flower The yellow centres of the flowers are not well detected KT UA E KI BESS Lo y AS na Na t f S Y N Es Figure 40 detection example Comparison of classes Quality comparison For each class a set of 20 representative images had been chosen and their colourings created Created colourings had been rated with same metric as the examples shown above Final mark for each class is based on average rating across its set As been expected simple cartoon class has the highest average rating There is a considerably large gap between cartoons and photos However the rating of real photos is better than being expected 33 Time comparison The detection of colouring is a very complex task and therefore it requires a lot of computation time In the table below is a comparison of average times for each of classes on testing device 18 TIME CONSUMPTION COMPARISON MS ME 260356 A 62824 B
11. on provided image The implemented detection methods are focusing on cartoon images or their photos However the source image is not bounded just to these types and therefore the application does generate a colouring for any type of provided image Colour it application also provides colouring environment for subsequent colouring Thesis also contains evaluation of generated colourings documentation and user manual Abstrakt Cilem t to prace je prozkoumat moznosti implementace metod zpracovavajicich obrazky a to jmenovit detekci hran a segmentaci elem je automatick vytv en omalov nek pro d ti ve v ku od 3 do 6 let V sledn implementace je prezentov na formou aplikace pro tablety s opera n m syst mem Android V sledn aplikace Colour it tak automaticky generuje omalov nky pro zadan obr zek Detek n metody se zam uj p edev m na komiksov obr zky i jejich fotografie Nicm n vstupn obr zek nen nijak omezen pouze na tento typ a tud je mo n vytvo it omalov nku pro libovoln vstup Colour it aplikace tak podporuje n sledn vybarvov n Sou st bakal sk pr ce je i hodnocen takto vygenerovan ch omalov nek dokumentace a u ivatelsk p ru ka Acknowledgements would like to express my gratitude to my bachelor thesis supervisor RNDr Daniel Pr a Ph D for introducing me to computer vision for his valuable hints and overall aid during my work which allowed me
12. or simple photos such as a photo or image from a fairy tale book Since the idea is to use a click to fill method of colouring a gap in an edge could mean a complete failure of detection Therefore in the thesis will not try to solve complicated photos which pose problems for detection and would require a lot of computational time which is still limited on such devices The program should also provide an environment for experienced users to manually set variables to further enhance the results of detection Another very important part is to provide a colouring environment for comfortable colouring which will provide template display click to fill colouring and a unique colour set for each colouring page corresponding to the original image but also the export of prepared colourings for possible print and colouring with real pencils afterwards Theoretical Background This chapter describes the theoretical background which is necessary for this work However it is expected that the reader is familiar with the basics of Java the least square metric or gradients Android overview Android is a Linux based operation system released by Google in 2009 Based on Google Statistics Every day more than 1 million new Android devices are activated worldwide 7 It is estimated that nowadays there are more than 85 of Android devices with OS version 4 0 or newer Based on Digital Trends statistics 16 from the end of year 2014 it is also
13. the most used operating system in mobile devices and its market share is still growing Operating system share of global OS market iOS 14 8 Programing for Android is supported in the Java language Android library which contains all necessary API to create all kinds of applications for Android devices The programs for Android devices are therefore written in the Java language Since it is expected that the reader is familiar with Java basics the main differences between Android and Java programming will be explained here In Android programming there is no Main function Instead of Main function there is a Main Activity which will launch after starting the application Every Android application is composed as a set of Activities with only one active Activity which is displayed at the device screen and with the rest in idle or sleep mode The active Activity can be switched to another while the previous one is put to sleep The graphical layout of each Activity is described by an XML file which is the same for many other constant resources needed for the application Event handling is similar to Java it is performed by adding event listeners For the purpose of detecting gestures on the display there is a library prepared which invokes one of events whenever it detects one of standard gestures This is useful for zooming an image There is also another library prepared for getting coordinates of clicking the display which will be u
14. B Tablet Intel Atom Quad Core Z3745 1 86GHz Bay Trail 8 1280x800 IPS 1GB RAM internal storage 16GB 2x camera 2MPx 0 3MPx Google Android 4 4 lt http www asus com us Tablets ASUS MeMO Pad 8 ME181C gt 49
15. H ER 46 Hotad ete m 47 DOIIE L 48 viii Table of figures Figure 1 gradient based colour MIU arias 6 Figur 2 3 colour quantization ample aussi rca dic 7 Figure cele lel ee leis Tee ICON summaries 9 Figure 4 toddler coloring book example Cat sse eee eee eee 9 Fjetire 5 Colonne OOK ICOM narco 10 FUSS AMOS LW AIRS ono 10 Figure 7 Coloring Dook example Sal nansisiiin did 10 USS cra COLO AOS ICO O it E NEEE 11 Figure 9 Scrap Coloring dette Le sms rias 11 Flere T0 ource DU 11 HBUFe DI Source NOUS Brea 12 Figure 12 Scrap Coloring detection NOUSC ccccssccccssececesecccsesecccesececaeceeeeneceseecesegecessueeesseecesegecetsgeeetsges 12 Figure 13 an example of simple cartoon Cass eee eee eee 15 Figure 14 an example of advanced cartoon class esses eene nennen eene enne snnt nnns 16 Figure 15 an example of class containing photos of Cartoons see eee ee eee eee 16 Figure Tog example of photo Class noe cto actos 17 Figure 17 an example OF colouring ClASS siii di di ii 17 A ASA dd e H ee ee 19 Figure 19 dilatation example the cross element has been used sss esse eee eee eee eee 19 Figure 20 6xamblie Or i GION clica cta tan 19 SIA de EO Tel roto 24 Figure 22 detection example usce es osi riada 24 Fietire 23 mete OTT example 25 E 8 te EON o O E MEME PIU PESE IM dri E MR EE ird 25 Pie 529 de Re e lel BXSITIDIB arado icieral 26 euro 26 Detecuon example coda du IA cus esee seb prre cun edema duze cU out ee ene M
16. IRI uU ee 26 FURS 27 derechon e kampl ase estu noes Rm MM eru EE UM MM M UEM EM 27 Ligure 25 BetectloNmeXdiTibleie esee EET METTRE OMIT ERI T E MERIT EIE MUI MI MM 27 Feeder HON BXdTTI DIG cnn 28 Figure S0 Kole lel 29179 id PU qiSIMINIM AUI ES 28 Figure 31 detection sam 00 eicicta ileso O DE SM USE 29 Heure 32 de Tl E enan E E E E EEEE EEE E 29 Heure 33 detection example ra 30 Flete o4 dette EO ERIE Omer eerie udi ents ia 30 Ligure 35 detection example it dis 30 Figure 36 detection SX nel ii A A A AAA ESO eie uade o C tue Des 31 PIC UVES 7 detecuoH exa le aiii 31 FISUre 38 detection example sss alii 32 FIGURE 39 detection example A AAA di 32 Heute dO detecHon example tentando tintos 33 Figure 41 TimesconsumptiormcormpariSORTssve s bin seo er yet SR REM ROE ias 34 Figure 42 a b Time consumption of separate classes sse eee eee eee 34 Figure 43 a b Time consumption of separate Classes sees eee eee 35 Figure 44 a b c d e f g h i j k Scrap Coloring and Colour it comparison sees 37 FIEUFe Ao BITe CVCle ADI cashes cise eisai lard dics EEN mode cas va reu seule se nf E etd ddd vadne da 41 Figure 46 credte TIOW ocio Ud 44 Feure 7 Colour it ICON m 58 44 Fieure d5 contine ColOUF Ing see iste init Quid up dpPiced idad br edd 44 Figure do advance Sd crea ON aa edad nu RO OS PO O A e aio oca eN oec tal Bosne hd cM cM ASER 44 FIgure 50 colouring screen example tei eire be Ee phe A ee
17. able and the toddler is likely to find colouring the same page over and over rather boring 12 Conclusion This solution aims at similar users as my work However the functions it provides are very simple and it has a very limited pool of pages Coloring book The database of pages is large and there are pages for everyone Page selection is well organized into difficulty groups which are divided into smaller groups on the basis of the themes they present Page selection is simple and does not require an experienced user On the other hand the main menu has too many items and performing actions which are repeated many times during colouring such as picking a different colour requires too much clicking and the actions are harder to find than they should be Conclusion This solution is nice with plenty of pages and advanced functions but it aims at older users At the same time some key functions such as the creation of new pages are missing Scrap coloring This browser based implementation has a lot of disadvantages Most importantly the user needs internet connection and colouring from Android devices is slower than it should be On the other hand the creation of unique colouring pages based on user provided images provides an inexhaustible source of new fresh colouring pages Some may not be perfect though Conclusion This solution offers suitable colouring environment and all key functions even though the creation of
18. al with them There are methods which can detect and slightly revert the motion blur but there is not sufficient computational power for such processes Resources of Android devices are often limited and the user will not wait minutes for a single colouring In order to normalize noise the Gaussian blur is applied Afterwards the noise normalized image is colour quantized This simplified image is less noisy while core contours are still left strong or at least as strong as they originally were The edge prolonging method is applied to these simplified images During the application of the prolonging method a dilatation method is injected several times The number of injections is based on the source image difficulty and varies from 1 to 3 The Injection of the dilatation method into the process of prolonging aims at connecting some edges which might be moved by motion blur and may consequently be left unused since they are not actually connected to the core parts of edges and are considered as noise Simplification methods may not be strong enough in some cases and noise can crawl into real edges Therefore a breaker of prolonging has been set which stops the iteration if over 25 of image is covered with edges 21 Customized creation of colourings Since human perception is stronger in many ways than the computer it would be wise to allow the user to use his advantage of additional knowledge to improve the colouring creation Therefore an interfa
19. are k a C d P lt gt lo fon L N 20 2 we v KE P E ge a W M A AN EJ ats Neu 1 m Y C d RT A x 4 m A 0 SS a pce n 2 CAE 4 gt U wo Y L N de k s A e p ee EV Sec MT OS T 4 mo LT PRA A a TA E j A A V 2 e y k e Y ES by A k i d B t d oh x Te Sy D v gt S NE oso m You Clicked Return to Coloring zl Figure 26 detection example Flowers 3 of 5 stars The purple flower is not well detected The rest of the image is well detected but the gradient colour mixture was simply too big for smoothing methods Figure 27 detection example House 5 of 5 stars All surfaces are well recognized and details of the drawing are carried as thin lines without making the colouring too messy 4 7 un E m FAME Fl q BZ j TT P E pa om l eC o Z meee m d a y a AA m Figure 28 detection example 27 Animals 4 5 of 5 stars Hard source image with lots of details probably containing too many details for the target group From the detection point of view there is a problem with the rock in the left bottom corner of the image There are too many edges present which do not correspond to the source image However this problematic spot is marginal since the important parts of the image are the animals rather than the rock e P T Figure 29
20. are unknown all assumptions are based on the current state of the image Therefore the only source of information is a matrix of pixels where each pixel has an integer value which describes the colour it presents First step is to compute the Otsu s threshold OT The OT is used in two ways First it initializes Canny edge detection and second it determines if the source image is a prepared colouring or not by the method described in chapter Colouring class detection below After this sub step it is clear whether to continue or to skip the rest of source recognition The most straightforward metrics is counting the size of the set which contains all unique colours and their shades in the source image This metrics Colour count is good to determine the difficulty of the source image Based on observation photos usually contain approximately over 80 000 different shades due to noise and on the other hand simple cartoon images usually have less than 10 000 shades It is a large number especially for simple cartoons which optically consist of just a few colours Most of those colour shades are found at edges This is caused by anti aliasing techniques which are applied for example during image reshaping Second useful metric is counting of solid colours A Solid colour is defined as RGB colour such is contained at least 100 times in analysed image Solid colour metric helps to distinguish the correct class It gives a different point o
21. ce is implemented which allows the user to employ detection techniques but also set the parameters manually The toolset consists of Canny with settable thresholds K means with settable number of clusters applying single iteration of edge prolonging and edge thickening performed by dilatation Results achieved by an experienced user may differ in the more difficult classes but in the simple cartoon class it should not be possible to achieve a better result Getting palette The idea of getting a colour palette based on the source image is fine however we should not forget that there is not only one correct way of colouring Therefore it is wise to add a static part of the palette which will provide all missing colours that might not be present in the original image so the user could make his own choice on the basis of his feelings rather than being limited by the application Getting the static part is simple because it comprises a predefined list of colours Getting the customized part is more complicated It is implemented by getting cluster centres from colour quantization method K means This method has been chosen because this set describes the image best from a mathematical point of view It is also important to note that colours contained in the palette may not even be present in the original image 22 Results This chapter describes the results of the implementation of the methods which were dealt with in the previous chapter Th
22. d teras Del eai ues dt 45 Figure 5T colour anything Los ui perte quts honda D v SAC test R beca vau CU BLA VO EIAUE tud teas DU rare reale 45 Kigure 52 COLOURING ds red SIND IO su v in o cumst deese sepu ke ume mo iue ub Um D diae MR du M DIM Id 45 Heure 5 5 COLO te TCI T T mms 46 Figure 54 advanced creation eXaMmPple uses euim ire b AAA at de E 46 Introduction This chapter explains the motivation for the thesis and its aims Motivation Nowadays tablets are involved in children s lives more than ever Therefore it would be desirable to use them for activities that can improve children s abilities and skills have decided to implement a colouring book application Colouring books help children improve their creativity and mainly their fine motor skills There may be disagreement on whether colouring via the tablet is as efficient as classical colouring on paper or not but think tablets are an alternative that is worth exploring since tablets surround us wherever we go while classical colouring books might not At present there are colouring books for Android devices available However they all lack what is in my opinion the most critical feature namely the creation of new colourings by the users themselves This opens the possibility of generating a large variety of new colouring pages which can be chosen to correspond accurately to children s needs and interests There can be found a project which implements creation of c
23. detection example Duck 5 of 5 stars The detection is flawless Grass is detected as thin lines which describes the image even better Figure 30 detection example 28 Photos of cartoons Photos were taken by the test device s camera 18 The camera has a 2Mpix resolution For purpose of demonstration the illustrations from children s books 13 14 were chosen as objects of photos Flower 5 of 5 stars Flawless detection of a simple image Figure 31 detection example Lizard 4 5 of 5 stars Well detected complicated colouring Figure 32 detection example 29 Hound with an umbrella 3 5 of 5 stars Water drops were not detected The bounds should be thinner However the overall impression is great Figure 33 detection example Snail carriage 4 of 5 stars The wheel of the carriage is not separated from the carriage itself Figure 34 detection example Bugs on a mushroom 2 of 5 stars Missing white discs the right shard of the bug on the left is not separated from the mushroom Figure 35 detection example 30 Real photos Unlike previous classes this one is not a target source Therefore these results cannot be taken as seriously as the results from previous tests A typical successful result from this class is divided into more surfaces which are smaller The typical result colouring is often too complicated for target users however colouring is not impossible Norwe
24. dges but stand alone Since the method aims at detecting edges which are basically curves there is no point in keeping single dots After this step only thin lines or curves will remain The last step of Canny is Hysteresis In this step only those edges which are between the lower and upper threshold are accepted These thresholds are given to the Canny detector as an input alongside the inspected image Otsu s threshold Otsu s threshold is the best binary threshold which divides the target image with the least possible variance into two sets State of the Art This chapter describes the solutions currently available for the colouring task There are many other colouring applications at Google Play store For a comparison have chosen 2 applications Toddler coloring book free pro and Coloring book Considering the number of total downloads and user ratings both are interesting from different points of view For the purposes of comparison have also included a web application Scrap coloring Since it is not an Android application full comparison is not possible however it is the only solution have been able to find which implements automatic creation of colouring pages from images Toddler Coloring book free pro Toddler coloring book TCB has a very lightweight approach which is based on very simple colouring of a few included colouring pages TCB is available in a free or paid version The paid version is an extend
25. e results bellow are divided into classes which are described in previous chapter It is very important to note that the results of detection are very much dependent on the source image If a user chooses an image which would be difficult for him to transform even manually then it cannot be expected that the result of this detection would do any better All the results were made as screen captures of the testing device 18 The example testing sets of images consists of 5 images from each class which had been chosen to cover all possible issues of the specific classes Classification The results were evaluated based on the accuracy of detection Evaluation must also consider the purpose of detection That means that areas must be well bounded without gaps Evaluation may not be precise since the target group of users is not capable of rating properly Evaluation is performed by rating from 0 to 5 stars which is same system as the one used for rating applications in Google Play Store Star rating is accompanied by a brief explanation of rating Stars are gained for fulfilling the following criteria 23 Detection Evaluation Simple cartoons Simple cartoons are easiest for detection and therefore evaluation can be stricter than in more difficult classes What is expected is a gapless detection with possible issues of unnecessarily thick bounds Bouquet 4 of 5 stars Main edges are thick Inner edges are thinner and descr
26. e well made and one not so much well made colouring from a provided image is below Som E Figure 9 Scrap Coloring detection Figure 10 source butterfly 11 NS Ge ng N f Coloring Pages First Name Coloring Color your photos and drawings News Patterns Friends of ScrapColoring Convert your Drawings and Photos to Coloring Pages New You can turn into a coloring page any drawing image or photo that you found on the Internet or that you have drawn and scanned The coloring page you creste can then be colored online with the colorful gradients and patterns of ScrapColoring All you need to do is to upload your image with the form below and it will be automatically and almost instantly into shapes that can be colored Notes Converting t e picture into shapes is a process called vectorisation Results are generally not perfect They vary greatly from the images used as input Shapes with sharp strokes filled with flat colors or black and white work best Photographs with s lot of details many colors or fuzzy strokes sre more difficult You will often get surprising but always interesting results You can also try with already colored drawings posters etc The optimal size of the input image is 600x600 pixels If you send an image with a different size it will be automatically cropped and resized We hope you will have much fun using our coloring page generator as wed had creating it and testing it Please tell us what you think and sh
27. ed here because it guarantees upper bound of optimality as log k competitive to optimal solution Since OpenCV supports K means no additional implementation is needed Figure 1 gradient based colour mixture In this work the K means clustering has been used as the colour quantization method which transforms an image into an image containing only K different colours These K different colours are cluster centres gained from the K means method This simplification of an image is useful in edge detection in areas where colour is slowly transforming from one colour to another Figure 2 3 colour quantization example Whenever K means is used there is one potential problem And it is a question how to select a correct number of clusters This is very subjective question and therefore there is no general answer There are some state of art approaches like the elbow method and others 17 However they are all too computationally complex and this work requires a quicker approach Therefore the K is determined as k max k c Where k 15 has been set based on observation of testing images and Csoiia computation is based on analysed image and is further described in Implementation chapter Another use of this method is to generate a corresponding colour palette for future colouring which makes the colouring much more similar to the original Thresholding Thresholding is a simple method used on grayscale images which perfor
28. ed version of the free one providing a larger pool of Figure 3 toddler coloring book icon predefined pages The free version has 8 pages and the paid one 58 It is a well rated application with 4 6 stars of 5 based on rating of more than 43 thousand people It has over 1 million of total downloads This application as the name hints is targeted at 1 to 3 year old children Therefore the interface consists only of a colour picker and the colouring itself There are no menus or advanced settings The buttons of colour picking are big enough for i even a toddler to use without the support of any adult E M Figure 4 toddler coloring book example cat Coloring book Coloring book CB is an application with over 14 million of total downloads It is rated as 3 8 stars of 5 based on over 38 thousand votes Figure 5 Coloring book icon CB consists of more than 400 prepared colouring pages sorted into groups based on colouring difficulty and theme It offers a large scale of degrees of difficulty which makes the target customer category wider STEEL ESBE RB hokej RB hokej P Coogle play P Google pia Fandi nasim z tribuny Fandi nasim z tribuny Figure 7 Coloring book example snail Figure 6 example dwarfs The colouring interface is rather complicated Functions are available in a menu which is positioned at the top of the screen and is composed of simple images Howeve
29. f view about the image Simple cartoon class simply does not contain that much colours On the other hand blurred images which usually comes from photo classes does contain fewer solid colours and more colour shades Therefore combination of those two metric can tell whenever to use noise reduction techniques and when it is probably pointless Solid colour count is also used in computing how many clusters will be used in Colour quantization Image class Average colour shades Average solid colours Simple cartoon Less than 20 000 Less than 100 Less than 60 000 More than 100 Photos of cartoon 6 000 to 150 000 Less than 350 General photos 60 000 and more 350 and more 18 Dilatation Dilatation is a method of thickening edges It is useful in connecting edges which are close to be connected but there is still a gap between them It can be also used to underline core edges In this work the dilatation is used with a cross element The cross element is iteratively placed to every pixel of the edge and every pixel the cross element covers is set to be an edge after dilatation Figure 18 cross element Figure 19 dilatation example the cross element has been used Prolong edges e AC Figure 20 example of prolonging From the left side 1 initial image 2 initial detection 3 prolonging method had been applied 4 dilatation to close gaps Prolonging of core edges is a method which allows to detect edges in nois
30. ge Here all methods needed are implemented Methods from this package are divided into separated classes based on their functions for better readability Classes here are the real hard workers of this application However they should never been called directly In this program they are always approached trough managers since methods here have not implemented any exception handling which is not mandatory This is designed this way for better overall performance And it also provides better readability Since there are only functions which provide results to some inputs without a bound to any object they are all static Entity package This package contains entities used for this application namely the entity class Colouring Page which is the fundamental cornerstone of this application 42 Colouring Page consists of several datasets which together describe a colouring page It consist of e bitmap of original image for the purpose of showing a template e bitmap of detected colouring e bitmap of current state of colouring e bitmap of masks with IDs of surfaces to which target pixel belongs e set of colours which are detected from the original image Programming philosophy As a device application it performs actions based on user inputs Whenever an event is started and the input is recognized the active Activity calls one of managers to perform a corresponding action and when manager returns the result the activity will make i
31. gian church 3 5 of 5 stars Even though there is a lot of needless edges the overall appearance is good The main surfaces are well bounded even though they may be composed of multiple smaller pieces This might be confusing for the target group but a more experienced user should be able to deal with it E aes 54 9 5 ma AED Merl we 7 dam p A 3 L c vt ad A a a uL VE x aar eo em Cu EO m ORE D c4 R v e e U R pd z md s c rA IT eet IA Y s gt Ah H s w n al 1 9 TAO ww g ae 42 y e E T i NC MP uu E Raaz ET N 4 AA EU S la gt LA MEL Figure 36 detection example Water Lilly 3 5 of 5 stars The colouring contains too much noise which makes the colouring very difficult However the main object is clear enough to be coloured wv Figure 37 detection example 31 Cow 1 of 5 stars A typical example of failure Snoothing methods were not able to clean the grass and therefore so many unwanted edges were found that the real objects in the image are completely lost On the other hand the cow has been detected well enough to be coloured a 252 3 f ca IO a E vest E i c TUE However this is not sufficient for the colouring to be marked as successful A2 A ft kon e d 4 KE d Y das D xs rear i ES D i de OE cue NS 4 fm om v mm l2 x e 45 T 7 Y SX SS F i DY A io gt gt 225 pa P pa 1 e NO A k d v U gt ME
32. hapter we discuss the implementation of more sophisticated methods which are employed in creating the colourings and their colouring afterwards Further details of implementation can be found in Documentation chapter bellow Detection implementation The image format covers a large field of possible source inputs It can be a high definition photo or a very simple image drawn for example in Microsoft Paint in Windows Realizing this fact leads to the conclusion that it would be wise to use different approaches for different kinds of source images Therefore the implementation can be divided into two main steps In the first step Source recognition the metrics are applied to determine what kind of image the algorithm is dealing with In the second step the knowledge from the previous step is used to modify detection according to the source image Image classes Basic classes of source images can be determined based on the behaviour of detection algorithms These classes react similarly and yield similar results based on same settings It is important to mention that human perception does not work the same way as computer based edge detection Therefore images are not divided into those classes by human perception but by mathematical metrics Simple cartoons This is the most common class for this work target group Such an image consists of blocks of solid colours The class includes for example templates of colourings or simple pictures w
33. hy the whole interface has more in common with C than with Java code structure This might be quite confusing especially for those who are not familiar with C OpenCV provides a full API description on their website 6 However it is often very brief and sometimes rather confusing For Android devices it is recommended to install OpenCV library as a self standing application rather than include the library into applications directly This has multiple reasons but the main is that OpenCV library can utilize the multiple core architecture which is very common in Android devices better Another important reason is that OpenCV is quite a large library it takes about 20MB so the developer should make sure that the user has it installed only once on his device rather than having it installed for every single application which benefits from OpenCV OpenCV uses Mat object as the basic entity Mat object consists of channels one for each RGB colour which are represented as matrices However the accessibility to a single pixel is very poor and expensive in terms of computation On the other hand Mat objects provide a great support for matrix multiplications and other matrix operations OpenCV also supports conversion from Android Bitmap object which is the basic Android image format to OpenCV Mat object in both ways K means clustering K means clustering 1 is a method which divides the data set into K clusters where ever
34. ibe the details inside well enough Only the red flower in the middle could be described better ad Rd You Clicked Return to Coloring P e 4 Figure 21 detection example Fish 5 of 5 stars No surface is forgotten no gaps Edges are not thicker than necessary Figure 22 detection example 24 Tractor 4 of 5 stars Centres of wheels could be described better Wavy lines are caused by low resolution of input image and therefore this defect should not have any impact on the evaluation Figure 23 detection example Mole 4 5 of 5 stars The source image is simple but contains lot of noise due to low resolution However false edges which are visible on the watering can should not be present Figure 24 detection example 25 Helicopter 3 of 5 stars Edges are pointlessly thick However the overall description of the image is good and the surfaces are cleanly separated Figure 25 detection example Advanced cartoons Advanced cartoon class is slightly more dependent on provided original image Therefore it is not expected to see results of such quality as in Simple cartoon class Lion cub 4 of 5 stars The trunk of the baobab is not separated from the ground Good detection of the lion cub where there could be a problem with shading on his chest Y gt all s lt E s e Pu m e IL za e NON e p E N X lt gt ANS k PERLE ee F n MET po aa a eU M S B 3 C
35. ings Creation of an advanced colouring Slightly advanced colourings can be achieved by using advanced cartoons or their photos Creation of expert colourings The most difficult colouring can be gained from real photos However in many cases the automatic creation will result in failure and therefore it would be advisable to use custom creation 39 Conclusion Future of project The results of this application are far above the initial expectations and therefore it is worth further testing and improvement In near future would like to upload the application to Google Play and make it freely downloadable Possible improvements During a review of the results came up with the idea that this application could be used for gathering data for future detection improvement In future it would be interesting to implement some statistical recognition algorithms for overall improvement based on gained data It could be useful to distinguish different classes of images in more accurate ways Possible usage At the beginning the idea was to make an application for children But when testing GUI realized that it could be used also by patients during stroke recovery It could be useful as an alternative and perhaps a less boring way of regaining fine motor skills 40 Appendices Documentation In this chapter will discuss the implementation philosophy and provide a more in depth concept of application than in Introduction chapter
36. is easy to use even for toddlers Usage is very simple 1 Select Create new colouring 2 Select an image which you would like to Figure 47 Colour it icon colour 3 Colour it Figure 46 create new Do you want to continue in colouring of some previously created and unfinished colouring pages Just simply 1 Click open Colouring 2 Select the desired colouring Grane 3 Continue where you stopped COJOUNNE Figure 48 continue colouring Are you a more experienced user or would you like to know how does it actually work Then why don t you try it yourself For curious users there is the advanced creation available The advanced creation allows to use complicated detection methods in very simple way Just select a method and see what will happen You can compare your detection with the automatic process and compete which is better Usage is again very simple 1 Select advanced Creation 2 Select an image 3 Apply methods from toolkit until you get a perfect colouring Advanced Creation Me 4 Colour it Figure 49 advanced creation Did something went wrong and you would do something differently Feel free to reset and Start again 44 Colouring screen Colouring environment is very simple to use On the left side you can see a palette with colours Figure 50 colouring screen example specifically chosen for this colouring Just click on desired colour and see that square in right bottom corner
37. ithout advanced shading Figure 13 an example of simple cartoon class 15 Advanced cartoons This is a more advanced class which contains images with more numerous and smaller surfaces of solid colours than simple cartoons For example there can be gradient colour mixtures or lighting effects 3 An 3 cm ENS m Figure 14 an example of advanced cartoon class Photos of cartoons This is a class of photos of cartoons which had been scanned or photographed previously Some noise is present in these images however there are still clear surfaces even though they are not made of solid colours Figure 15 an example of class containing photos of cartoons 16 Real photos This is the most general class where a lot of noise with no clear surfaces is expected Therefore the result is largely dependent on the source image its quality and content a Figure 16 an example of photo class Colourings This class comprises downloaded colourings prepared to be coloured Images are grayscale and consist of white and black like colour Only real issue is that black colour is often shade of grey and not strictly black Figure 17 an example of colouring class 17 Source recognition As an input to this step a resized image of unknown content is expected At the end we expect a set of settings for a detection algorithm Since the content density source or almost any other information about the source image
38. ms an action based on threshold value In this case binary thresholding has been used In binary thresholding each pixel is compared to a given threshold and if the pixel value is smaller than the threshold then it is set to O while if it is greater or equal then it is set to 255 This corresponds to black or white colours The method is very useful in cases where prepared colourings were downloaded to determine exactly where the areas are This is necessary especially when anti aliasing was previously used on the image Anti aliasing is a method employed to smooth the image so that the contours look more fluent Binary thresholding can be used for reversing an anti aliased image to an aliased one In the OpenCV library this method is also used to count Otsu s threshold which is useful in edge detection Canny edge detection Edge detection 1 is a name for a set of methods which use brightness changes in the image to detect the so called edges There are many approaches to the detection of edges In this Work state of art Canny Edge Detection has been chosen Canny edge detection consists of multiple steps After filtering noise from the image for example by a Gaussian filter the next step is finding the brightness gradients This is done by the Sobel procedure which will not be explained here When we have the gradients a non maximum step is performed This step removes single pixels which had been detected as potential e
39. o well designed class for images it is used for holding and any operations with images in this work Since the Android Java package does not contain any methods for computer vision an additional library will be necessary OpenCV As noted in the previous chapter the Android library does not include any tools for image processing Therefore it is necessary to use some external library which will fill up this gap Several computer vision libraries are available OpenCV javaCV imageJ boofCV yet not all of them offer precisely what will be needed OpenCV is indisputably the most robust most commonly used and most optimized out of all available BSD licenced libraries and it also has a well accessible documentation Therefore the OpenCV library for Android is being used in this work OpenCV library is a widely used computer vision library which was initially written in C language and released in 2000 It is developed by Itseez team Full interfaces for Python C Matlab as well as for Java are now available OpenCV is a large library which covers a wide area of uses such as image processing motion tracking or statistical machine learning For this work we will use only image processing namely Canny Edge Detection Gaussian Blur K means colour quantization and thresholding A description of these methods can be found in the following chapters bellow Even though there is a Java interface the library is composed as a C library which is w
40. olding it can be said and easily proved that Wsource lt Wthresholded Where W determines quantity of white pixels in respective image The decision whether image is a colouring 1 or is not 0 is computed by following function 0 W 0 W d W W 241 lt W t W C 0 else Where c 1 2 is experimentally gained constant 20 Simple cartoons detection For simple images it is not necessary to choose a difficult approach It is specific of this class that there is almost no noise present in the image Therefore the only real danger of using mild thresholds does not apply In combination with dilatation the resulting colourings are clear and the edges are as complete as they were in the original image Advanced cartoons detection When dealing with colour gradients mixtures the Canny edge detection has a weak spot Therefore it is necessary to use colour quantization before Canny can be used Colour quantization will separate the gradient colours into two or more solid surfaces which can be easily detected afterwards After colour quantization the image is very similar to a simple cartoon from the Canny detector point of view However since the image is more complicated better results are achieved if edge prolonging is applied Photos of cartoons detection The result of this class is strongly dependent on the source Since photos especially those from mobile devices are often motion blurred and noisy it is hard to de
41. olourings from an image 11 however it is only a web application so the user needs an internet connection and a browser did not find any application of this type on Google Play which is the main source for Android applications or at any other place on the internet Therefore have decided to create a new colouring books application which implements this feature with a fully or partially depending on the user s decision automated creation process from images for Android tablets on my own For the purpose of creating the application have decided to implement edge detection and other image processing methods One of main reasons why have decided to use such methods is that think that it is important to show that computer science is not only an academic or high tech research field but also one that can be introduced into everyday life and can serve as a powerful tool even for the youngest of us Goal The goal of this work is an application which will provide automatically customized methods based on an initial image These methods should be used to create accurate and simple colouring pages Since there are not any universal object detection methods and it is still an unsolved problem this application will focus on detecting in specified areas of images The application should be able to handle simple cartoon images flawlessly It should also handle the import of prepared colouring pages or detection over average cartoon images
42. ow us your creations Image Vybrat soubor 45172 cart a bear jog Type Black and white drawing Photo or drawing with several colors a a o oo En 27 Figure 11 source house 33 es 39 Figure 12 Scrap Coloring detection house Since the SC colouring environment is a Y L c Copyright 2014 http scrapcoloring com A Get it on Share your pA bd a on coloring on ENE PENES L a and more Facebook webpage it can also be used from Android devices It does not require flash or any other plugins which usually cause problems on the Android system SC offers a large variety of colours and for experienced users there is also an implementation of gradient mixing of two colours or the use of a texture Export of the result is also supported It is possible to print or download it however the resolution is always 600x600 pixels which might be sometimes insufficient and may modify the picture which was originally wider or taller The colouring environment is simple and easy to understand Nevertheless the creation of a colouring is more complicated and would require someone skilled enough to provide the source image Comparison have described three approaches to colouring tasks and now will summarize the advantages or disadvantages of these solutions and compare them Toddler coloring book The colouring interface is simple and easy to understand However there are not many pages avail
43. pages may not be perfect However a fundamental disadvantage is the need for internet connection because the user has to wait for the server response This can take several seconds depending on the connection after each action which can be considered quite slow Recommendation For this work have decided to combine the design philosophies of the above applications in order to achieve a better result The most common user interface philosophy is implemented in Toddler coloring book Since our target group is pre school children the colouring environment has to be kept simple 13 As far as choosing right page is concerned the best approach is represented by the Coloring book application Sliding trough previews of pages seems best in terms of using the potential of the touch display The approach of selecting source image and letting the rest on device seems to be the only way for this work s target age category However there is a potential problem since the selection of new image cannot be implemented in simpler way and it still might be too complicated for some young users Therefore we assume that there might be a second more experienced user a parent for example who will help the child to create new pages With the experienced user in mind the work should offer advanced creation options beside the basic ones which would let the user influence the process to achieve even better results 14 Implementation In this c
44. r it can be confusing since each menu item has a different colour and a toddler may assume for example that it is not a menu but a colour picker itself The function for the picking of colour is hidden 10 behind the pencil the palette or the yellow button depending on the user s preferences because each offers slightly different possibilities Since performing the basic function of picking the colour requires multiple touches of the right places the target customer category is definitely not toddlers and is not suitable for pre school children either On the other hand CB offers multiple advanced functions such as the zoom which is almost mandatory to use when colouring hard images or the undo function which reverses the last action the user made Another advanced function is exporting the colouring for the purposes of sharing or printing A paid version is also available but it is not so well rated or downloaded In the paid version there is an interesting feature of importing prepared colouring pages Scrap coloring Figure 8 Scrap Coloring icon Scrap coloring SC is a webpage focused on colouring Since it is a webpage it can be used only with internet connection It contains a database of prepared colourings for online colouring or printing but also a section for creating new colourings from the pictures provided by the user The results of this feature are largely dependent on the input image An example of on
45. s Capenglr 2014 sc azerlering com 35 scrspodorng com 36 You Clicked Return to Coloring Copyright 2014 scrapcoonng com T Ps H A Copyright 2014 scrapcelonng com y Pio Ape e vilu V L edr Copyrigt 2014 scrspestoring com Copyrigt 2014 serspestoring com rde CAT DEL 1 441 les E 7 3 TUM GU LT en C PAL AE Figure 44 a b c d e f g h i j k Scrap Coloring and Colour it comparison 37 Discussion This chapter discusses if and to what extent the goals were accomplished The discussion will be presented for each class separately This chapter also provides a cookbook suggesting which type of image to select when a particular type of result is required Result evaluation Simple cartoons In most cases the results of simple cartoon pictures are flawless If the evaluation should be strict then it can be said that some bounds are too thick and thinner ones would be better Advanced cartoons The most common defects consist in the fact that in the results often contains fake lines in the middle of gradient mixtures This is not a big problem however for younger users it could be confusing A bigger issue might be the possible gaps in bounds In the simple cartoons the edges were without gaps because there were always two solid colours around some continuous part of edge This does not work for advanced cartoons The problems are
46. sed for detecting the correct surface to be coloured or picking the colour For every Android application you can set its behaviour in landscape and portrait orientation This application is locked in landscape mode for better usage of the screen which is already too small in many cases That is also the reason why this application is recommended only for devices with the diagonal 7 and greater This practically eliminates all smartphones and leaves tablets only The reason is that colouring on a smaller device would lead to unpleasant results where the colouring page would be too small to be coloured properly However this is only a recommendation because it is not possible to forbid installation on devices which do not meet this requirement Every Android application has a minimal and a target version of the Android system The minimal version is the lowest possible Android version required to install that application Since image processing methods which will be described later need quite a lot of computational time have decided to set the minimal Android version to 4 0 which should guarantee a sufficient amount of computational power The Android library also provides a large set for working with Bitmap which is the main object class for images in Android Therefore there are also static methods available for working with them such as creating of new Bitmaps or changing target pixels in others Since the Bitmap is the default and als
47. t appear Whenever a manager is called it uses tools from utility tools or detection package to perform desired actions while being responsible for all exception handling as well At this level the manager must make sure to use tools properly since tools themselves have no exception handling built in Those tools are usually short functions like resizing bitmap and so on For better understanding present an example Let s say a user is in the middle of colouring a page Therefore there is some state of colouring page some selected colour Let s call this knowledge a State The user performs an action to colour a new area by touching the screen at the desired location The active Activity recognizes that input is at the location X Y and that the user wishes to colour the area Therefore it calls Paint Manager PM with providing pointers to current state and location of user input PM checks if its input is valid and calls a method which fills the target area and returns a new state Then PM checks the result and sets a new state as the current and sends a response back to Activity that computation is over At this moment Activity switches the old colouring for anew one which has the desired area filled And the application is waiting for the next user input 43 User manual Colour it Colour it application allows to create and colour infinite number of new and unique colourings Due to simple and intuitive graphical layout it
48. ur count apply Gaussian blur reset 46 Toolbox consists of following methods 1 Reduce colour count It reduces number of colours present in the image to selected number 2 Apply Gaussian blur Is your image noisy Then it is time to use this It will help to the rest of tools in their performance 3 Detect initial use sliders to properly select what is important and what is not 4 Thick edges detected edges contains small gaps Then you might want to have the detected edges a bit thicker Feel free to apply it multiple times 5 Prolong edges detection is still missing some key points By prolonging the edges you might find some missing contours 6 Reset did something went wrong Feel free to try it again with different setup 7 Save colouring is your detection flawless Save it so you or your children can colour it any time later Installation The installation of this application is done by an apk file which after transferring to device automatically installs the application However since this application uses OpenCV which is distributed as a self standing application it is necessary to download the OpenCV library as well The easiest way to do this is to connect the device to Google Play and then start the Colour it application If there is no library installed it will pop a window which will ask for installing it If you allow this it will be downloaded and installed without any other actions Howe
49. ver internet connection will be required 47 Resources Testing images were collected across Pixabay com web which provides pictures without reuse restrictions 1 onka M Hlav V Boyle R Image Processing Analysis and Machine vision 3rd edition Thomson Learning Toronto Canada 2007 ISBN 2 Robert Lagani re OpenCV 2 Computer Vision Application Programming Cookbook 1st edition Packt Publishing Ltd Birmingham UK 2011 ISBN 978 1 849513 24 1 3 Pavel Herout U ebnice jazyka Java 5th extended edition Kopp publishing Cesk Bud jovice Czech Republic 2013 ISBN 978 80 7232 398 2 4 Nudelman G Android Design Patterns 1st edition Wiley Indianapolis USA 2013 5 L Lucchese S K Mitra Color Image Segmentation A State of the Art Survey online 2007 accessed 15 5 2015 available at http ultra sdk free fr docs Image Processing filters Color9620Image9620Segmentation A 20State of the Art 20Survey pdf gt 6 Itseez team OpenCV library documentation online 2015 accessed 10 5 2015 available at lt http docs opencv org java gt 7 Google Inc Android developer online 2015 accessed 14 5 2015 available at lt http developer android com training index html gt 8 David Arthur Sergei Vassilvitskii k means The Advantages of Careful Seeding online 2007 accessed 5 3 2015 available at lt http ilpubs stanford edu 8090 778 1 2006 13 pdf gt 9 A Greensted
50. y single data vector in this case the coloured pixel the 3 dimensional colour vector RGB is aligned to the nearest cluster The method consists of an iteration over two steps after initiation In the initiation the starting centres are selected based on a chosen strategy The most common strategy is a random choice of K vectors from the dataset which obviously does not guarantee any kind of optimality and it has been shown 15 that in the worst case the running time can be even super polynomial This is why in this thesis a K means method has been used to choose initial centres The initiation is followed by an assignment step In this step the nearest cluster centres are found for all data vectors so the data set is divided into K clusters The nearest cluster centre is defined as the one with the lowest least square distance When the evaluation step is completed the next step is the update of centres For each cluster a new centre position is chosen It is chosen to minimize the sum of distances of all vectors from that cluster distance is defined as least square distance After this we repeat the assignment step and the evaluation step respectively until no cluster centre changes its position This method will not guarantee the global optimum because it converges to local optimum only The optimality of the result is strongly dependent on the initial distribution of cluster centres That is the reason why K means initial distribution is us
51. y images The idea is to find cores of real edges with strict thresholds and iteratively prolong them 19 At the initiation the method sets some strict threshold which will produce just core pieces of edges With each iteration the threshold bounds are loosened a bit One iteration consist of the following steps 1 Loosen the thresholds a bit 2 Detect new potential edges with the loosened threshold 3 If any of potential edges are connected to real edges from previous iterations append such edges to real edges 4 If a sufficient number of new edges has been found continue with a new iteration else terminate it The main advantage of this approach against a simple edge detection is that it allows to find weaker parts of edges which would be otherwise covered with noise A critical part of prolonging is how to set correct initial thresholds In this work they are computed based on Otsu s threshold Colouring class detection Otsu s threshold OT is a division marker which indicates which pixels of a grayscale image will be black and which will be white By comparing the original source image with the one created by thresholding it can be said how much it has changed Based on that it can be determined if the source image was meant to be a colouring or just an ordinary grayscale image The comparison of the source and thresholded image is performed by computing over white pixels only Based on what we know about thresh
Download Pdf Manuals
Related Search