Segmentation-based lossless compression of burn wound images

Segmen a ion-based lossless comp ession o bu n wound images Ca men Se ano Begon ˜ a Acha Pin ˜ e o A ´ ea de Teo ı ´adelaSen ˜ al y Comunicaciones Escuela Supe io de Ingenie os Uni e sidad de Se illa Camino de los Descub imien os s/n. 41092 Se illa, Spain E-mail: [email p o ec ed], [email p o ec ed] Ranga aj M. Rangayyan Depa men o Elec ical and Compu e Enginee ing Uni e si y o Calga y Calga y Albe a, T2N 1N4, Canada Lau a M. Roa G upo de Ingenie ı ´a Biome ´dica Escuela Supe io de Ingenie os Uni e sidad de Se illa Camino de los Descub imien os s/n. 41092 Se illa Spain Abs ac . Colo images may be encoded by using a g ay-scale im- age comp ession echnique on each o he h ee colo planes. Such an app oach, howe e , does no ake ad an age o he co ela ion exis ing be ween he colo planes. In his pape , a new segmen a ion-based lossless comp ession me hod is p oposed o colo images. The me hod exploi s he co ela ion exis ing among he h ee colo planes by ea ing each pixel as a ec o o h ee componen s, pe o ming egion g owing and di e ence ope a ions using he ec o s, and applying a colo coo dina e ans o ma ion. The me hod pe o med be e han he Join Pho og aphic Expe s G oup (JPEG) s anda d by an a e age o 3.40 bi s/pixel wi h a da- abase including ou na u al colo images o scene y, ou images o bu n wounds, and ou ac al images, and i ou pe o med he Join Bi-Le el Image expe s G oup (JBIG) s anda d by an a e age o 3.01 bi s/pixel. When applied o a da abase o 20 bu n wound im- ages, he 24 bi s/pixel images we e e icien ly comp essed o 4.79 bi s/pixel, hen equi ing 4.16 bi s/pixel less han JPEG and 5.41 bi s/pixel less han JBIG. © 2001 SPIE and IS&T. [DOI: 10.1117/1.1383781] 1 In oduc ion Lossless comp ession echniques a e essen ial in a chi al and communica ion o medical images. Inc easing applica- ions o elemedicine in he a eas o de ma ology and pa- hology a e c ea ing new demands, such as ansmission and a chi al o colo images. The Bu n Uni o he Hospi al Uni e si a io Vi gen del Rocı ´o de Se illa and he Biomedi- cal Enginee ing and Signal P ocessing G oups o he Uni- e si y o Se illa a e de eloping a elemedicine p ojec whe e he diagnosis o bu n pa ien s is pe o med wi h digi- al colo pho og aphs. As au oma ic classi ica ion is pe - o med in a p ocedu e whe e colo and ex u e a e essen ial, a lossless comp ession me hod is needed o his pa icula applica ion. Al hough signi ican e o has been di ec ed owa ds he de elopmen o lossless algo i hms o comp essing image da a, mos o such me hods ha e been o ien ed owa ds comp essing g ay-scale o wo- one 共bina y兲images. I is commonly s a ed ha a ed-g een-blue 共RGB兲colo image can be easily comp essed by using a g ay-scale image com- p ession echnique on each o he h ee colo componen s. Howe e , such an app oach does no ake ad an age o he co ela ion exis ing be ween he colo planes. Recen ly, a ew e o s ha e been di ec ed owa ds com- p ession o ue-colo images by aking ad an age o hei spec al co ela ion Singh e al.1p oposed an in ege -based wa ele ans o m me hod, which was shown o be e icien o comp ession o images wi h ine de ails. Memon and Sayood2p oposed se e al me hods based on e o p edic- ion models, and epo ed imp o emen o e he pe o - mance o he Join Pho og aphic Expe s G oup 共JPEG兲 s anda d o abou 1.5 bi s pe colo pixel. Bocks ein3p o- posed a new me hod based on a lossless ans o ma ion om he RGB planes o o he planes, de ined as linea combina ions o he o me , and comp essing each o he new planes wi h a g ay-scale image lossless comp ession algo i hm.4Van Assche e al.5p oposed a echnique based on he Ka hunen–Loe ` e ans o m 共KLT兲 o deco ela e he colo planes; comp ession a es o abou 0.5–2.0 bi s pe pixel be e han hose p o ided by lossless JPEG we e ob- Pape 99083 ecei ed Dec. 22, 1999; e ised manusc ip ecei ed Aug. 7, 2000; accep ed o publica ion Feb. 23, 2001. 1017-9909/2001/$15.00 © 2001 SPIE and IS&T. Jou nal o Elec onic Imaging 10(3), 720 – 726 (July 2001). 720 / Jou nal o Elec onic Imaging / July 2001 / Vol. 10(3) Downloaded F om: h p://elec onicimaging.spiedigi allib a y.o g/ on 03/30/2017 Te ms o Use: h p://spiedigi allib a y.o g/ss/ e mso use.aspx ained, bu wi h he disad an age o high compu a ional ime. In 1985, Kun e al.6p oposed a con ou - ex u e ap- p oach o pic u e coding; hey called such app oaches second-gene a ion image coding echniques, and hey ap- plied i o lossy comp ession. The main idea behind his echnique is o i s segmen he image in o nea ly homo- geneous egions su ounded by con ou s such ha he con- ou s co espond, as much as possible, o hose o he ob- jec s in he image, and hen o encode he con ou and ex u e in o ma ion sepa a ely. Because con ou s can be ep esen ed as one-dimensional signals and pixels wi hin a egion a e highly co ela ed, such me hods a e expec ed o a ain high comp ession a ios. Al hough he idea seems o be p omising, i s implemen a ion mee s a se ies o di icul- ies. A majo p oblem exis s a i s e y impo an i s s ep o segmen a ion, which de e mines he inal pe o mance o he segmen a ion-based coding me hod. Mos o he seg- men a ion algo i hms based upon his p inciple a e sophis- ica ed and gi e good pe o mance only o speci ic ypes o images. Shen and Rangayyan7p oposed a segmen a ion-based lossless image coding 共SLIC兲 echnique o adiog aphic images. To o e come he p oblem men ioned abo e in e- la ion o segmen a ion, hey p oposed a simple egion g owing me hod. Ins ead o gene a ing a con ou se , a dis- con inui y map is ob ained du ing he egion g owing p o- cedu e. Concu en ly, he me hod also p o ides a co e- sponding e o image based on he di e ence be ween each pixel and i s co esponding ‘‘cen e pixel.’’ The las s ep in he algo i hm is o code he discon inui y and e o da a wi h he Join Bile el Image expe s G oup 共JBIG兲com- p ession s anda d.8 We ha e ecen ly ex ended he SLIC algo i hm o colo images.9In o de o exploi he co ela ion exis ing among he h ee colo planes, ou algo i hm ea s each pixel o he image as a h ee-dimensional 共3D兲 ec o 共RGB兲and pe - o ms 3D egion g owing. The me hod p oduces a h ee- componen e o image bu only a one-componen discon- inui y map. Wi h a iew o ob ain be e comp ession, we ha e included a lossless colo -coo dina e con e sion s ep om RGB o he luminance, in-phase, and quad a u e- phase 共YIQ兲 ep esen a ion sys em.10 We de o e Sec. 2 o p o ide de ails o he segmen a ion- based lossless colo image coding 共SLCIC兲algo i hm, and p o ide he esul s o i s compa ison wi h he JPEG in e - na ional s anda d o lossless s ill-image comp ession11 in Sec. 3. 2 Segmen a ion-Based Lossless Coding Algo i hm o Colo Images In lossless image comp ession, he ask is usually spli in o wo s ages: one is image ans o ma ion, wi h he pu pose o da a deco ela ion; he o he is encoding o he ans- o med da a. In he SLCIC algo i hm, image ans o ma ion is achie ed in bo h he egion g owing p ocedu e and, la e , in he JBIG algo i hm. The JBIG algo i hm uses an adap- i e h ee-dimensional coding model ollowed by an adap- i e a i hme ic code .8Figu e 1 gi es a block diag am dem- ons a ing he a ious s eps in he SLCIC p ocedu e. 2.1 Region G owing Algo i hm As desc ibed by Shen and Rangayyan7 o single- componen images, he egion g owing p ocedu e s a s wi h a single pixel, called he seed pixel. Each o he seed’s ou -connec ed neighbo pixels, scanned in a ixed o de , is checked wi h a egion g owing 共o inclusion兲condi ion. I he condi ion is sa is ied, he neighbo pixel is included in he egion. The p ocedu e is ecu si ely con inued un il no spa ially connec ed pixel mee s he g owing condi ion. A new egion g owing p ocedu e is hen s a ed wi h he nex pixel o he image which is no al eady a membe o a egion. The p ocedu e ends when e e y pixel in he image has been included in one o he egions g own. The egion g owing condi ions used in his wo k a e 共i兲 he neighbo pixel is no a membe o any o he egions al eady g own, and 共ii兲 he absolu e di e ence be ween he Fig. 1 Summa y o he SLCIC p ocedu e. Segmen a ion-based lossless comp ession... Jou nal o Elec onic Imaging / July 2001 / Vol. 10(3) / 721 Downloaded F om: h p://elec onicimaging.spiedigi allib a y.o g/ on 03/30/2017 Te ms o Use: h p://spiedigi allib a y.o g/ss/ e mso use.aspx ed, g een, and blue in ensi ies o he neighbo pixel and hose o he co esponding ‘‘cen e pixel’’ is less han a p ede ined e o –le el. The cen e pixel is de ined as ha pixel which is being used as he e e ence o check i s ou - connec ed neighbo s o inclusion in he egion being g own; he cen e pixel would ha e al eady been included in he egion. When a new neighbo pixel is included in he egion being g own, i s e o –le el-shi -up di e ence wi h espec o i s cen e pixel is s o ed as he pixel’s ‘‘e o ’’ alue; his alue is a ec o o h ee componen s. I only he i s o he wo egion g owing condi ions is me , he dis- con inui y index o he pixel is inc emen ed. The discon i- nui y index indica es how many imes a pixel has been es ed o be included in a egion and inally i has no been included. The e o e, a e egion g owing, a ‘‘discon inui y index image da a pa ’’ and a h ee-componen ‘‘e o im- age da a pa ’’ will be ob ained. The maximum alue o he discon inui y index is ou . Mos o he p e iously epo ed segmen a ion-based coding algo i hms include con ou cod- ing and egion coding; ins ead o hese s eps, we use a discon inui y index da a pa and an e o image da a pa . The e o –le el is de e mined by he p eselec ed e o –bi s assigned o each o he h ee componen s o he ec o as e o –le el⫽2e o –bi s⫺1. Fo ins ance, i he e o image is allowed o ake up o 5 b/colo -plane-pixel o each componen (e o –bi s⫽5), he co esponding e o –le el is 16; in o he wo ds, he allowed di e ence ange is hen 关⫺15, 15兴. The e o alue o he seed pixel o each egion is de ined as he ec o o med by he low e o –bi s bi s o each componen . The alue o he high (N⫺e o –bi s) bi s o he h ee componen s o he pixel is s o ed in a ‘‘high-bi s seed da a pa ,’’ whe e Nis he numbe o b/pixel in he o iginal image da a. The abo e h ee da a pa s a e used o ully eco e he o iginal image du ing he decoding p ocess. The egion g owing condi ions du ing decoding a e 共i兲 he neighbo pixel unde conside a ion o inclusion is no in any o he p e iously g own egions, and 共ii兲 he discon inui y index o he pixel is equal o ze o. When he condi ions a e me o a pixel, i s pixel alue is es o ed as he sum o i s e o –le el-shi -down e o alue and i s cen e pixel alue 共excep o he seed pixel o e e y egion, o which he la e is gi en by he high-bi s seed da a pa 兲. I only he i s o he wo condi ions is sa is ied, he discon inui y index o he pixel is dec emen ed. Thus he discon inui y index gene a ed du ing segmen a ion is used o guide e- gion g owing du ing decoding. The egion g owing p ocedu e may be iewed as adap- i e scanning o he image wi h he aim o main aining a localized di e ence o e o alue wi hin a limi ed dynamic ange p especi ied and con olled by e o –bi s. Figu e 2 is a simple example o illus a ion o he e- gion g owing p ocedu e and i s esul . The o iginal image is he 512⫻512 24-bi Lena image. A 5⫻5 segmen o he image is shown in Fig. 2共a兲. The alue o e o bi s is se o h ee in his example. Figu e 2共b兲is he esul o egion g owing. The co esponding h ee da a pa s, namely dis- con inui y index image da a, e o image da a, and high-bi s seed da a, a e shown in Figs. 2共c兲–2共e兲. 2.2 Segmen a ion-Based Lossless Colo Image Comp ession P ocedu e The comple e SLCIC p ocedu e is illus a ed in Fig. 1. A he encoding end, he o iginal image is ans o med by he egion g owing p ocedu e in o h ee pa s: he discon inui y index image da a pa wi h one-dimensional elemen s, and he e o image da a pa and he high-bi s seed da a pa composed o 3D ec o s. The i s wo da a pa s a e hen g ay coded, b oken down in o bi planes, and inally JBIG coded. The las da a pa is s o ed o ansmi ed as-is; i needs (N⫺e o –bi s)*3 bi s pe seed pixel. A he ecei ing end JBIG-coded da a iles a e JBIG- decoded i s and hen he g ay-coded bi planes a e com- posed back o bina y code. Finally, he pa s a e combined oge he by he same egion g owing p ocedu e o eco e as in he encoding scheme o eco e he o iginal image. 2.3 Lossless RGB- o-YIQ T ans o ma ion Usually, when coding colo images wi h loss, a ans o ma- ion om RGB colo coo dina es o YIQ is pe o med. This is based upon he obse a ion ha he Y componen is un- co ela ed wi h he ch ominance componen s 共I and Q兲, and ha mos high- equency componen s o a colo image a e concen a ed in he Y componen .12 In lossless comp es- sion, he second ad an age canno be exploi ed, bu he i s one is used o gain imp o ed comp ession. To change he colo coo dina es om RGB o YIQ he ollowing linea ans o ma ion is pe o med12: Fig. 2 Simple example o he egion g owing p ocedu e and i s e- sul wi h e o bi s se obe h ee:(a)a5⫻5 segmen o he 512 ⫻512 24-bi colo Lena image, whe e each alue ep esen s he R, G, and B componen , espec i ely; (b) esul o egion g owing; (c) co esponding discon inui y index da a pa ; (d) co esponding e o image da a pa ; (e) he co esponding high-bi s seed da a pa . Se ano e al. 722 / Jou nal o Elec onic Imaging / July 2001 / Vol. 10(3) Downloaded F om: h p://elec onicimaging.spiedigi allib a y.o g/ on 03/30/2017 Te ms o Use: h p://spiedigi allib a y.o g/ss/ e mso use.aspx 冋 Y I Q 册 ⫽ 冋 0.299 0.587 0.114 0.596 ⫺0.274 ⫺0.322 0.211 ⫺0.523 0.312 册 冋 R G B 册 .共1兲 Howe e , his ans o ma ion canno be applied in loss- less comp ession because i in ol es loa ing-poin ope a- ions, hus equi ing quan iza ion. In o de o a oid quan i- za ion e o , a ans o ma ion based on in ege a i hme ic is equi ed. Such a ans o ma ion is comple ely e e sible and hence lossless, and is gi en by:1 Yl⫽ b b R⫹B 2 c ⫹G 2 c ,共2兲 Il⫽R⫺B, 共3兲 Ql⫽ b R⫹B 2 c ⫺G, 共4兲 whe e b•cdeno es he loo unc ion. The in e se ans o ma ion is gi en by: R⫽Yl⫹ b Ql⫹1 2 c ⫹ b Il⫹1 2 c ,共5兲 G⫽Yl⫺ b Ql 2 c ,共6兲 B⫽Yl⫹ b Ql⫹1 2 c ⫺ b Il 2 c .共7兲 The subsc ip lis used o indica e he lossless na u e o he ans o ma ion. This ans o ma ion pe o ms he YIQ one, bu wi hou losses, and i is p o ed in Re . 1. I is impo an o no e ha in o de o p ese e he lossless na- u e o he algo i hm, one bi mus be added o each o he ch ominance e o componen s 共one o I and one o Q兲 due o he sign ha appea s when con e ing om RGB o YlIlQl. 3 Expe imen al Resul s The SLCIC echnique was es ed using 12 24-bi colo im- ages o di e en sizes and na u e. The pe o mance o he SLCIC echnique was compa ed wi h ha o he lossless JPEG 7 s anda d and he JBIG s anda d. The ini ial es image se includes ou images o scene y 共Lena, Peppe s, Pa o s, and Ge many兲, ou images o bu n wounds, and ou ac al images 共 om h p://di .yahoo.com/A s/ Visual–A s/Compu e –Gene a ed/F ac als/A is s/兲. The bu n wound images ha e been aken ollowing a p o ocol explained in Re . 13. In his p o ocol a digi al pho og aph came a is used, and he pho og aphs a e s o ed as ue-colo ones, i.e., 24 bi s pe pixel, 8 bi s pe each colo componen 共RGB: ed, g een, blue兲. Figu es 3–5 show he e o and he discon inui y index da a pa s as images o h ee o he es images used. I can be obse ed in all o he cases illus a ed ha he e o image does no con ain any signi ican ch oma ici y in o - ma ion 共ac ually hey look like g ay-scale images being colo ones兲, bu e ains a po ion o he bounda y in o ma- ion o he image. The discon inui y index clea ly co e- sponds o edges in he image. The comp ession pe o mance o he SLCIC me hod wi h he ini ial se o 12 images is summa ized in Tables 1 and 2 along wi h he esul s o JPEG and JBIG comp es- sion. The SLCIC me hod has a unable pa ame e , which is e o –bi s. The SLCIC me hod has ou pe o med lossless JPEG on he a e age o he es image se used by 0.68 bi s pe colo pixel 共wi h he o iginal images ha ing 24 b/pixel in he RGB domain兲, and JBIG by 0.29 b/pixel. As shown in Table 1, when using he YlIlQlcoo dina es, he pe o - mance o he SLCIC is much be e , p o iding an a e age o 9.24 b/pixel. The me hod ou pe o med s anda d JPEG by an a e age o 3.34 b/pixel, JBIG by 2.95 b/pixel, and SLCIC wi h RGB coo dina es by 2.66 b/pixel. In Table 2 a compa ison o he code and decode imes o SLCIC, JPEG, and JBIG is done, whe e he imes o eading and w i ing he images o and om he disk a e also conside ed and, o he case o he SLCIC algo i hm, he ime em- ployed in he con e sion om RGB o YIQ is also in- cluded. Al hough in e nally SLCIC uses JBIG, i ge s as e imes han JBIG alone. This is due o he ac ha JBIG ac ing alone has o code 24 bi planes whe eas JBIG join o SLCIC only has o code he numbe o bi planes de e - mined by he pa ame e e o –bi s. Addi ionally, he egion g owing p ocedu e does no add a signi ican quan i y o he o al compu a ional ime. The JBIG so wa e we ha e Table 1 Compa ison o he pe o mances o SLCIC, JBIG, and JPEG using wel e 24 b images by b/pixel. Image JPEG (b/pixel) JBIG (b/pixel) SLCIC (b/pix) RGB Y l I l Q l Scene y: 1. Peppe s 15.69 15.65 15.99 16.29 2. Ge many 17.73 17.49 16.49 12.35 3. Lena 14.79 15.12 15.08 14.98 4. Pa o s 18.04 12.06 18.01 9.63 A e age o scene y: 16.56 15.08 16.39 13.31 Bu n wounds: 1. Pho o 1 10.8 12.43 10.64 5.87 2. Pho o 2 7.36 8.19 6.56 4.04 3. Pho o 3 7.45 8.68 6.8 4.32 4. Pho o 4 7.33 8.67 6.74 4.22 A e age o bu n wounds: 8.24 9.49 7.69 4.61 F ac als: 1. A lan is 15.12 14.05 13.76 12.76 2. 7 hm 12.09 11.27 11.00 10.97 3. Mandsil 14.58 14.46 13.45 7.90 4. Yinyw hm 9.95 8.17 8.25 7.62 A e age o ac als: 12.94 11.99 11.62 9.81 A e age o all images: 12.58 12.19 11.9 9.24 Segmen a ion-based lossless comp ession... Jou nal o Elec onic Imaging / July 2001 / Vol. 10(3) / 723 Downloaded F om: h p://elec onicimaging.spiedigi allib a y.o g/ on 03/30/2017 Te ms o Use: h p://spiedigi allib a y.o g/ss/ e mso use.aspx Fig. 3 Lena: (a) o iginal image; (b) e o image; (c) discon inui y index image. Fig. 4 Bu n wound: (a) o iginal image; (b) e o image; (c) discon inui y index image. Fig. 5 F ac al: (a) o iginal image; (b) e o image; (c) discon inui y index image. Se ano e al. 724 / Jou nal o Elec onic Imaging / July 2001 / Vol. 10(3) Downloaded F om: h p://elec onicimaging.spiedigi allib a y.o g/ on 03/30/2017 Te ms o Use: h p://spiedigi allib a y.o g/ss/ e mso use.aspx employed is a ee dis ibu ion e sion de eloped by AT&T in 1991. A e examining he Y, I, and Q componen images sepa a ely, we ealized ha he I and Q componen s we e much mo e uni o m han he luminance one so, hey would need less e o –bi s o be encoded han he Y componen . We changed he algo i hm in o de o code each componen wi h di e en numbe o e o –bi s, and we go be e e- sul s, ha a e summa ized in Table 3. To a oid con usions we ha e called his modi ica ion o he SLCIC algo i hm, SLCIC2. Wi h he SLCIC2 he imp o emen gained o e JPEG is 3.40 b/pixel on a e age and 3.01 b/pixel o e JBIG. Due o he good esul s ob ained wi h bu n wound im- ages, we we e encou aged o s udy he algo i hm o ou pa icula applica ion. Hence, we es ed he SLCIC p oce- du e wi h a la ge da abase o 20 bu n wound images. The esul s a e summa ized in Table 4, gi en as he mean and he a iance o SLCIC, SLCIC2, JPEG, and JBIG, whe e we can see ha he SLCIC me hod has ou pe o med s an- da d JPEG by an a e age o 4.13 b/pixel and JBIG by 5.38 b/pixel, and he SLCIC2 has ou pe o med JPEG by an a e age o 4.16 b/pixel and JBIG by 5.41 b/pixel. The ea- son o such a good pe o mance wi h bu n wound images can be ound in he special ea u es o he images: hey ha e uni o m colo s, such ha majo i y o i s in o ma ion is loca ed in he luminance componen exclusi ely. Addi ion- ally, e en in he luminance componen , he e a e la ge, almos -uni o m egions. In he same way as he e o ma ix and he discon inui y index ma ix a e coded wi h JBIG a e he egion g owing s ep, we expe imen ed wi h encoding he seed ec o wi h JBIG. Howe e , he esul s we e poo because he numbe o egions ob ained was no la ge. 4 Discussion While he e ha e been some wo ks on lossless comp ession o colo images, e y ew ha e exploi ed he no ion o wha has been called second-gene a ion image coding,14 in pa - icula he segmen a ion-based app oach. In his a icle, a me hod based on segmen a ion 共 egion g owing兲has been p esen ed. When es ed wi h se e al gene al-pu pose im- ages, he p oposed me hod esul ed in an imp o emen o 3.40 bi s pe colo pixel o e he lossless JPEG s anda d and o 3.01 bi s pe colo pixel o e he JBIG s anda d. When es ed wi h a da abase o 20 bu n wound images, he me hod ou pe o med s anda d JPEG by 4.16 bi s pe colo pixel, and s anda d JBIG by 5.41 bi s pe colo pixel. As we said, he eason o such a good pe o mance wi h bu n wound images can be ound in he special ea u es o he images: hey ha e uni o m colo s, such ha majo i y o i s in o ma ion is loca ed in he luminance componen exclu- si ely. Addi ionally, e en in he luminance componen , he e a e la ge, almos -uni o m egions. So, we ge e y good esul s when conside ing di e en numbe o e o –bi s o each componen . The ch ominance compo- nen s 共I and Q兲need less e o –bi s han he luminance one, because hey ca y less in o ma ion. The compu a ional speed o SLCIC is no e y slow compa ing wi h ha o JPEG due o he use o an e icien Table 2 Compa ison o code and decode ime o JPEG, JBIG, and SLCIC-Y l I l Q l . Compu ing ime o less han1sisen e edas0in he able. Image Size (col⫻ ow⫻b/pixel) Code/decode ime (s) JPEG JBIG SLCIC Scene y: 1. Peppe s 512⫻512⫻24 2/0 14/14 12/12 2. Ge many 768⫻512⫻24 2/0 19/19 18/18 3. Lena 512⫻512⫻24 2/0 13/13 12/12 4. Pa o s 768⫻512⫻24 3/0 19/19 16/16 Bu n wounds: 1. Pho o 1 832⫻624⫻24 4/1 23/23 17/17 2. Pho o 2 832⫻624⫻24 4/1 22/22 19/19 3. Pho o 3 832⫻624⫻24 5/1 22/22 17/17 4. Pho o 4 832⫻624⫻24 5/1 21/21 18/18 F ac als: 1. A lan is 320⫻240⫻24 0/0 4/4 1/1 2.7 hm 160⫻120⫻24 0/0 1/1 0/0 3. Mandsil 160⫻159⫻24 0/0 1/1 0/0 4. Yinyw hm 320⫻240⫻24 0/0 4/4 1/1 Table 3 Compa ison o he pe o mances o SLCIC2, JBIG, and JPEG using wel e 24 b images by b/pixel. The SLCIC2 algo i hm is implemen ed using di e en numbe o e o –bi s o he Y, I, and Q componen s. Image JPEG (b/pixel) JBIG (b/pixel) SLCIC2 (b/pix) Scene y: 1. Peppe s 15.69 15.65 16.29 2. Ge many 17.73 17.49 12.32 3. Lena 14.79 15.12 14.97 4. Pa o s 18.04 12.06 9.61 A e age o scene y: 16.56 15.08 13.30 Bu n wounds: 1. Pho o 1 10.8 12.43 5.83 2. Pho o 2 7.36 8.19 4.01 3. Pho o 3 7.45 8.68 4.29 4. Pho o 4 7.33 8.67 4.20 A e age o bu n wounds: 8.24 9.49 4.58 F ac als: 1. A lan is 15.12 14.05 12.69 2. 7 hm 12.09 11.27 10.80 3. Mandsil 14.58 14.46 7.58 4. Yinyw hm 9.95 8.17 7.55 A e age o ac als: 12.94 11.99 9.66 A e age o all images: 12.58 12.19 9.18 Segmen a ion-based lossless comp ession... Jou nal o Elec onic Imaging / July 2001 / Vol. 10(3) / 725 Downloaded F om: h p://elec onicimaging.spiedigi allib a y.o g/ on 03/30/2017 Te ms o Use: h p://spiedigi allib a y.o g/ss/ e mso use.aspx egion g owing me hod. Mos o he compu a ional cos o SLCIC is due o JBIG coding included in he algo i hm. The egion g owing p ocedu e does no add a signi ican quan i y o he o al compu a ional ime. In a SUN Ul a SPARC 共120 MB memo y, 2.1 GB disk, clock 167 MHz, 112 KB cache memo y兲SLCIC equi ed 1–20 s o encode/ decode he images used; JPEG equi ed 1–5 s o he same images. We a e explo ing possibili ies o u he imp o emen in he pe o mance o SLCIC by analyzing he na u e o he e o image. One imp o emen unde conside a ion is, in- s ead o changing colo coo dina es o he whole image, o change he ep esen a ion o only he e o image om he RGB planes o hue, sa u a ion, and alue 共HSV兲o o YlIlQl. We a e also explo ing he possibili y o using ech- niques o he han JBIG, a e he egion g owing s ep, o code he e o and discon inui y pa s in o de o ake in o accoun he indi idual cha ac e is ics o each pa . Acknowledgmen s This wo k was suppo ed by g an s om CICYT 共TIC-96- 0500-C10-08兲and Jun a de Andalucı ´a o Spain and he Na u al Sciences and Enginee ing Resea ch Council 共NSERC兲o Canada. We hank Liang Shen o help wi h he segmen a ion-based lossless image coding echnique o monoch ome images, and Mihai Ciuc o assis ance wi h colo image p ocessing. We also hank D . Toma ´sGo ´mez- Cı ´a o he Unidad de Quemados o he Hospi al Uni e si- a io Vi gen del Rocı ´o de Se illa o p o iding he bu n wound images. Re e ences 1. I. Singh, P. Aga hoklis, and A. An oniou, ‘‘Lossless comp ession o colo images using an imp o ed in ege -based nonlinea wa ele ans o m,’’ P oc. IEEE In . Symp. Ci cui s Sys ., 2609–2612 共1997兲. 2. N. D. Memon and K. Sayood, ‘‘Lossless comp ession o RGB colo images,’’ Op . Eng. 34共6兲, 1711–1717 共1995兲. 3. I. M. Bocks ein, ‘‘An adap i e me hod o lossless comp ession o colo images,’’ P oc. SPIE. Image P oc. Compu . Op ., 2363, 300– 306 共1994兲. 4. I. M. Bocks ein, ‘‘A me hod o lossless image comp ession,’’ in Pa - e n Recogni ion and Image Analysis, W. Poelzlei ne and E. Wenge , Eds., pp. 92–98, Sch i en eihe de Oes e eichischen Compu e Ge- sellscha , Wien, Aus ia 共1993兲. 5. S. Van Assche, W. Philips, and I. Lemahieu, ‘‘Lossless comp ession o p e-p ess images using linea colo deco ela ion,’’ P oc. IEEE Da a Comp ession Con ., 578 共1998兲. 6. M. Kun , A. Ikonomopoulos, and M. Koche , ‘‘Second-gene a ion image-coding echniques,’’ P oc. IEEE 73共4兲, 549–574 共1985兲. 7. L. Shen, and R. M. Rangayyan, ‘‘A segmen a ion-based lossless im- age coding me hod o high- esolu ion medical image comp ession,’’ IEEE T ans. Med. Imaging 16共3兲, 301–307 共1997兲. 8. ‘‘P og essi e bi-le el image comp ession,’’ In e na ional Teleg aph and Telephone Consul a i e Commi ee 共CCITT兲. Recommenda ion T. 82, 1993. 9. B. Acha, C. Se ano, R. M. Rangayyan, and L. M. Roa, ‘‘Lossless comp ession algo i hm o colou images,’’ Elec on. Le . 35共3兲, 214–215 共1999兲. 10. C. Se ano, B. Acha, and R. M. Rangayyan, ‘‘Segmen a ion-based lossless comp ession o colo images,’’ in P oc. Ten h In e na ional Con . on Image Analysis and P ocessing (ICIAP’99), Venice, I aly, pp. 90–94 共1999兲. 11. G. K. Wallace, ‘‘The JPEG s ill pic u e comp ession s anda d,’’ Com- mun. ACM 34共4兲, 30–44 共1991兲. 12. C. S. Lim, Two-dimensional Signal and Image P ocessing, P en ice Hall Signal P ocessing Se ies, P en ice-Hall, Englewood Cli s, NJ, 共1990兲, pp. 422–423. 13. C. Se ano, L. Roa, and B. Acha, ‘‘E alua ion o a elemedicine pla - o m in a bu n uni ,’’ P oc. o IEEE In e na ional Con . on In o ma- ion Technology Applica ions in Biomedicine, Washing on DC, pp. 121–126 共1998兲. 14. S. Van Assche, W. Philips, and I. Lemahieu, ‘‘Lossless comp ession o p e-p ess images using a no el colo deco ela ion echnique,’’ P oc. SPIE 3308, 85–92 共1998兲. Ca men Se ano ecei ed he MS in Tele- communica ion Enginee ing om he Uni- e si y o Se ille, Spain, in 1996. In 1996, she joined he Signal P ocessing and Com- munica ion G oup a he same uni e si y, whe e she is cu en ly an associa e p o es- so . He esea ch in e es s conce n image p ocessing and, in pa icula , colo image segmen a ion and comp ession, mainly wi h biomedical applica ions. Begon ˜a Acha Pin ˜e o s udied Telecommu- nica ion Enginee ing a Enginee ing Fac- ul y o he Uni e si y o Se ille om 1991 o 1996. She has been wo king as an associ- a e p o esso since 1996 in he Signal P o- cessing and Communica ions G oup o he Elec onic Enginee ing Depa men o he Uni e si y o Se ille. She is cu en ly wo k- ing owa d he PhD. He cu en esea ch ac i i ies include wo ks in he ield o image p ocessing and i s medical applica ions. Ranga aj M. Rangayyan ecei ed his Bachelo o Enginee ing deg ee in Elec- onics and Communica ion in 1976 om he Uni e si y o Myso e a he P.E.S. Col- lege o Enginee ing, Mandya, Ka na aka, India, and his PhD in Elec ical Enginee ing om he Indian Ins i u e o Science, Banga- lo e, Ka na aka, India, in 1980. He was wi h he Uni e si y o Mani oba, Winnipeg, Mani oba, Canada, om 1981 o 1984. He is a p esen a P o esso wi h he Depa - men o Elec ical and Compu e Enginee ing (and an Adjunc P o- esso o Su ge y and Radiology) a he Uni e si y o Calga y, Cal- ga y, Albe a, Canada. His esea ch in e es s a e in he a eas o digi al signal and image p ocessing, biomedical signal analysis, medical imaging and image analysis, and compu e ision. His cu - en esea ch p ojec s a e on mammog aphic image enhancemen and analysis o compu e -aided diagnosis o b eas cance ; egion- based image p ocessing; knee-join ib a ion signal analysis o nonin asi e diagnosis o a icula ca ilage pa hology; and analysis o ex u ed images by ceps al il e ing and soni ica ion. He is he winne o he 1997 Resea ch Excellence Awa d o he Depa men o Elec ical and Compu e Enginee ing, and he 1997 Resea ch Awa d o he Facul y o Enginee ing, Uni e si y o Calga y. He was ecognized by he IEEE wi h he awa d o he Thi d Millennium Medal in 2000, and elec ed as a Fellow o he IEEE in 2001. Lau a M. Roa: Biog aphy and pho og aph no a ailable. Table 4 Compa ison o he pe o mances o SLCIC, SLCIC2, JPEG, and JBIG by b/pixel using 20 bu n wound images. Each im- age is o size 832⫻624 pixels, wi h 24 b/pixel in he RGB ep esen- a ion. 20 bu n wound images JPEG (b/pixel) JBIG (b/pixel) SLCIC (b/pixel) SLCIC2 (b/pixel) A e age 8.95 10.20 4.82 4.79 Va iance 2.16 2.07 0.54 0.54 Se ano e al. 726 / Jou nal o Elec onic Imaging / July 2001 / Vol. 10(3) Downloaded F om: h p://elec onicimaging.spiedigi allib a y.o g/ on 03/30/2017 Te ms o Use: h p://spiedigi allib a y.o g/ss/ e mso use.aspx