OPTIMIZATION OF IMAGE ENHANCEMENT USING AN ARTIFICIAL IMMUNE SYSTEM

Transcription

1 Interntionl Journl of Mechnicl Engineering nd Technology (IJMET) Volume 8, Issue 8, August 7, pp. 84 9, Article ID: IJMET_8_8_ Avilble online t eme.com/ijmet/issues.sp?jtype=ijmet&vtyp pe=8&itype=8 ISSN Print: nd ISSN Online: IAEME Publiction Scopus Indexed OPTIMIZATION OF IMAGE ENHANCEMENT USING AN ARTIFICIAL IMMUNE SYSTEM G. K. Jgtheswri Assistnt Professor, Deprtment of Science nd Humnities, PES Institute of Technology-Bnglore South Cmpus, Bnglore, Indi Murugesn R Professor, Deprtment of Physicl Sciences, REVA University, Bnglore, Indi ABSTRACT Using fuzzy logic technique, new method is proposed for the enhncement of imges. The suggested method is introduced for the improvement of color imges using combintion of fuzzy logic method nd rtificil immune system optimiztion lgorithm. The fuzzifiction of underexposed region hs been crried out by Gussin membership function nd for fuzzifying overexposed res, tringulr membership function is found suitble. An estimte clled exposure, which serves s divider hs been defined for both underexposed nd overexposed prts of the imge. The hue, sturtion nd intensity (HSV) color spce is used for the trnsformtion of the imge from RGB to HSV spce without chnging the true color composition unblemished. Two functions, nmely prmetric sigmoid function nd power-lw opertor finds good suitbility for the enhncement of the underexposed nd overexposed regions. An objective mesure involving entropy, the contrst nd the visul fctors re constructed for the computtion of the prmeters present in the opertors nd re optimized using rtificil immune system lgorithm. The results re compred with genetic lgorithm (GA) bsed imge enhncement nd the proposed method when correlted with recent GA- bsed pproch is found to be generting better results. Key words: Artificil Immune System, contrst fctor, exposure, entropy, fuzzifier, imge enhncement, intensifiction, overexposed imge, underexposed imge, visul fctor. Cite this Article: G. K. Jgtheswri nd Murugesn R, Optimiztion of Imge Enhncement using n Artificil Immune System, Interntionl Journl of Mechnicl Engineering nd Technology, 8(8), 7, pp MET/issues.sp?JType=IJMET&VType=8&ITy ype=8. INTRODUCTION Due to common defects like insufficient lighting, the size of the perture, the shutter speed nd trnsmission errors, the visul qulity of the imges gets ffected. The defect found in recorded imge is shown in the rnge nd shpe of the gry-level histogrm. The imge 84 editor@ieme.com

2 Optimiztion of Imge Enhncement using n Artificil Immune System enhncement techniques plys significnt role in improving the qulity of the originl imge by incresing the potency of some fetures or by decresing the uncertinty between different regions of the imges nd provides more dt tht ws not present in the originl picture.a lot of imge enhncement lgorithms like logrithmic trnsformtion, piecewise liner trnsformtion nd histogrm processing techniques re described in [].Velde [] ttempts to enhnce the color imge in the LUV color spce, where ech prt is pplied to get the grdients. Color sturtion djustment for mking more nturl colors is implemented by To nd Asri [3].Nevertheless, these techniques cnnot be generlized for color imges s they do not retin the originl colors in the enhnced imge. To overcome the limittion of mbiguity or uncertinty in imge processing, fuzzy set theory bsed imge processing techniques hs been developed by vrious reserchers. A generlized repetitive fuzzy enhncement lgorithm is projected by Dong-Ling nd An-Ke [4]. Another fuzzy frmework in the field of imge enhncement nd smoothing is by using IF-THEN-ELSE rules [5]. These fuzzy rules re pplied to set of neighboring pixels in imge enhncement problems. The trnsformtion function s-curve is used in [6] nd its prmeters re computed by optimizing the entropy. To modify the Gussin membership function, Hnmndlu et l. [7]exmined prmetric sigmoid function relted to the optimiztion of entropy tht contin new intensifiction opertor (NINT) nd single fuzzifier which is extended to color imges[8].hnmndlu nd Jh [9] ddressed the issue of underexposed nd overexposed imges by proposing globl intensifiction (GINT) opertor which hs prmeters tht re exmined by optimizing the imge entropy nd which holds good to enhnce underexposed imges only. The process in [] is extended for imge enhncement using combined fuzzy logic technique nd bio-inspired optimiztion technique []. The contrst of color imges is enhnced in [] by introducing new objective function clled contrst informtion fctor which is optimized using prticle swrm optimiztion. In this proposed method, the pproch in [] is extended by introducing n enhncement to ll clssifictions of degrded imges. Bsed on the objective mesure clled exposure, the imge cn be ctegorized into underexposed nd overexposed regions nd cn be discretely be ltered by both sigmoid function nd power lw opertors. The prmeters present in both the opertors re clculted by involving the contrst nd visul fctors of the imge. The objective function with the opertors is optimized to enhnce the imge by using n AIS optimiztion lgorithm. AIS is computtionl intelligence method to problem solving inspired by theoreticl immunology nd observed immune functions, principles, nd models. The min ppliction domins of AIS re computer nd web security, fult nd nomly detection, optimiztion computtion, dt nlysis nd informtion mining.the lgorithm does not represent mutul ction reltion between the ntibodies nd ntigens, which led the popultion diversity in the serch process.. IMAGE CLASSIFICATION, IMAGE FUZZIFICATION AND INTENSIFICATION Every imge is considered to be combined type of imge including certin percent of underexposed or overexposed res before proceeding to the enhncement process.the V component of HSV represented in gry levels is pplied for the use of delinetion of n imge in underexposed or overexposed res.a prmeter clled exposure denoting n intensity exposition to find out the prt of imge gry levels in both underexposed nd overexposed regions is given by 85 editor@ieme.com

3 G. K. Jgtheswri nd Murugesn R exposure= l l l r( x) x r( x) where stnd for the gry-level vlues of the imge, ( )denotes the histogrm of the whole imge, nd l denotes the totl number of gry levels. If the vlue of exposure is found to be more thn.5 for nygiven imge, then there is n overexposed region thn underexposed region nd imge is found to be plesnt. Different types of opertors re pplied for enhncing the underexposed or overexposed imges. A new fctor given by in the rnge [,l -] prtition the gry levels into [,-] for underexposed imges nd for overexposed imges from [, l -] =l (-exposure) () The imge informtion to be processed in different imge processing pplictions is uncertin or mbiguous. To overcome this difficulty, fuzzy techniques gives effective tools tht quickly del with the uncertinty nd mbiguity of imges. The three min steps of fuzzy imge processing re imge fuzzifier, modifiction of membership vlues nd defuzzifiction. An imge I of size z in the rnge [, l -] cn be considered s set of fuzzy singletons in the fuzzy set nottion. µ U (3) z I = { µ ( z )} =, () p=,,,p; q=,,.,q µ where µ ( z ) or, represents the degree of membership µ of z, with z being z the color intensity t the (, )th pixel. The membership functions re computed only for the intensity components of the color imge. A Gussin MF described in [] is used to fuzzify the imge in the underexposed region s given below: z µx u( z) = exp mx ( z f vg h z) (4) where z represents the gry level of the underexposed region in the rnge [, -], zmx is the mximum intensity level in the imge, nd z vg is the vlue of the verge gry level in the imge. f h is clled the fuzzifier.the initil vlue is computed from 86 editor@ieme.com

4 Optimiztion of Imge Enhncement using n Artificil Immune System f l h = l ( z z) ( z x) p( z) p( z) For the fuzzifiction of n imge in overexposed region, tringulr MF is defined for z which is given by mx mx 4 (5) z < µx o ( z) = z z L Membership functions help to convert the imge intensity levels from the sptil domin into the fuzzy domin where the vlues re in the intervl [,]. After the process of fuzzifiction, defuzzifiction is done using the corresponding inverse membership functions nd the modified membership vlues re converted bck to the sptil domin. A prmetric sigmoid function is defined for enhncing the MF vlues of the gry levels of the underexposed region nd is given by µ X u( z) = t( µ X ( ) c u z µ + e (6) (7) where t is clled the intensifier nd µ c isclled the crossover point. Another opertor clled power-lw opertor is defined for enhncing the overexposed region of the imge.hence, the function of the power-lw opertor is to enhnce the informtion content by power γ on the sum of the tringulr MF plus smller vlueε for the exposed region is given by µ X ( z) = K ( µ X ( z) + ε ) o o γ (8) Here, we tke K= nd ε = for simplicity. Using these opertors, the improvement cnnot be chieved in the permnently degrded region of the imge nd this problem cn be prtilly resolved bymodifying the sturtion.the intensity is required just for certin degrded imges nd the mount of vrition should depend on the portion of the overexposed or underexposed region in the imge. Another power-lw opertor clled s the sturtion opertor is defined by SO ( z) = [ SO( z)] (.5*exp osure) (9) where SO(z) is the originl sturtion vlues nd SO ( z) is the sturtion vlues of the HSV color spce which is modified. Modified sturtion reinsttes the plesurble nture of such imges editor@ieme.com

5 G. K. Jgtheswri nd Murugesn R 3. FUZZY CONTRAST MEASURES In this outlook, fuzzy performnce mesures like contrst nd visul fctors cn be employed to pproximte the ctul imge qulity becuse n pproprite performnce mesure is difficult to determine tht cn exctly retlite to the ctul qulity of the imge nd corresponds to the chrcteristics of the humn optic system. The qulity of the color imges which is enhnced cn be evluted using the objective nd quntittive nlysis nd lso by the subjective visul inspection of the imges. In this pproch, entropy nd visul fctors cn be utilized for quntittive mesuring of the enhnced color imge. The definition in Eq. nd Eq. gives the fuzzy contrst of the modified imge nd the verge fuzzy contrst of the enhnced imge in the underexposed region respectively. The definitions in Eq. nd Eq. 3 give the fuzzy contrst nd the verge fuzzy contrst of the modified imge in the overexposed region respectively. X u ( z)) fu = ( µ () fu = ( µ X u ( z)) l X o ( z)) fo = ( l µ l fo = ( l µ Also the definitions in Eq. 4 nd Eq. 5 give the fuzzy contrst of the originl imge symbolized by s nd the verge fuzzy contrst of the originl imge lso symbolized by s for the underexposed region respectively. Eq. 6 nd Eq. 7 represents the contrst nd the verge contrst of the originl imge symbolized by s for the overexposed region respectively. X o fus = ( µ ( z)) X u ( z)) fus = ( µ X u fos = ( l µ ( z)) l X o ( z)) l fos = ( l µ X o ( z)) () () (3) (4) (5) (6) (7) The visul fctor is computed from the contrst fctor nd the rte of the fuzzy contrst with respect to the verge fuzzy contrst gives the contrst fctor. The verge fuzzy contrst provides the overll qulity of the imge nd the fuzzy contrst gives the spred of the grdient with respect to the crossover point editor@ieme.com

6 Optimiztion of Imge Enhncement using n Artificil Immune System The contrst fctor of the modified imge is defined in Eq. 8for the underexposed region nd similrly for the overexposed region of the modified imge is defined in Eq. 9. fu Q CFu = fu (8) Q = CFo fo fo (9) The respective contrst fctor of the originl imge is defined in Eq. for the underexposed region nd similrly for the overexposed region of the originl imge is defined in Eq.. Q = CFos CFus Q = fos fos fus fus () () 3.. Entropy In the fuzzy domin, entropy is used to ssess the qulity of informtion in n imge nd is given by E = [ ( )ln( ( )) ( ( ))ln( ( ))] + ln µ X u z µ X u z + µ X u z µ X u z l L [ µ Xo ( z)ln( µ Xo ( z)) + ( µ Xo ( z))ln( µ Xo ( z)) ] () The entropy is optimized to determine the prmeters t, µ, f nd γ. 3.. Visul Fctors The visul fctor otherwise clled s the normlized contrst fctor is determined for the underexposed nd the overexposed regions nd it defines the mount of enhncement produced nd both cn be combined using the exposure prmeter. The vlue of the visul fctor lies between. nd.5 for plesing imge which llows scertining the visul evlution of both underexposed nd overexposed regions Optimiztion of Objective Function The entropy function nd the visul fctor yields mesure of uncertinty connected with the imge informtion nd therefore hs to be optimized. The entropy function given by E is optimized with respect to the constrintsv = V. The objective function is setup under c h df f 89 editor@ieme.com

7 G. K. Jgtheswri nd Murugesn R J = E + λ V df V f (3) where λ is the Lgrngin multiplier. The intervls where the prmeters re chosen for optimiztion of the objective function is, µ, nd γ. t c For quick convergence, AIS lgorithm is pplied bsed on positive selection nd clonl selection principles. The clonl selection methodis bsed on choosing the most suitble ntibody from pool of ntibody group nd gets the new ntibody with high similrity from the pool by simple rndom muttion. The clonl selection lgorithm mintins the high ffinity offspring(ntigen) in muttion threshold rdius R wheres low ffinity prents nd offspring ntibody is discrded. The proposed lgorithm considers the entropy function s ntibody nd uses the Gussin nd the chotic distribution for both muttion nd cloning process on the ntibodies. The clonl selection will work ccording to the flowchrt. 4. FLOWCHART Strt Specify the Prmeters Discrd No Fitness < Threshold Rndomly Populte Librry Yes Built Antibody Reinstte Mutted Antigen in Antigen Librry Clculte Fitness Clculte Antigen Librry Discrd No Fitness< Threshold Yes Clone Chromosome with Mximum Fitness Add into Antigen Librry Mutte Antigen Librry 9 editor@ieme.com

8 Optimiztion of Imge Enhncement using n Artificil Immune System Scene Figure () Originl Imge b) Enhnced imge using AIS lgorithm (c) Enhnced imge using entropy-bsed method Flower Figure () Originl Imgeb) Enhnced imge using AIS lgorithm (c) Enhnced imge using entropy-bsed method Tble Visul fctors of AIS bsed pproch is compred with the Entropy-Bsed pproch Imges V f of Entropy pproch V f of proposed pproch Scene Flower RESULTS AND CONCLUSION The proposed method is evluted bsed on visul fctors nd entropy which served s objective mesures nd the results re obtined using Mtlb. The mount of modifiction in the imge is found by optimizing the prmeters t, µ, f nd γ using the optimiztion of c h, the objective function. A few test imges, nmely Scene nd Flower re considered in the subjective evlution of the ppernce nd these figures contins both underexposed imges nd overexposed imges which hve poor brightness nd higher brightness respectively. The proposed method is compred with entropy-bsed method [3] nd the comprison of the AIS bsed lgorithm with entropy-bsed lgorithm is shown in Figs. -. It is found from Tble tht visul fctors obtined by entropy-bsed pproch re less thn those of the proposed pproch. From these figures, it my be noted tht the enhncement ttined with the AIS method is more plesing in nture s the entropy-bsed pproch do not consider underexposed nd overexposed regions for the enhncement of imges. 9 editor@ieme.com

9 REFERENCES G. K. Jgtheswri nd Murugesn R [] R.C.Gonzlez, R.E. Woods, Digitl Imge Processing, 3 rd Ed., Person Eduction, 8. [] K.V.Velde, Multi-scle color imge enhncement, in Proc. IEEE Int.Conf. Imge Process, 999, Vol.3, pp [3] L. To nd V. Asr, Modified luminnce bsed MSR for fst nd efficient imge enhncement, in Proc. IEEE 3 nd AIPR, 3, pp [4] P. Dong-Ling nd X. An-Ke, Degrded imge enhncement with pplictions in robot vision, in Proc. IEEE Trns. Imge Process., vol. 4, no. 8, pp [5] M. Russo nd G. Rmponi, A fuzzy opertor for the enhncement of blurred nd noisy imges,ieee Trnsctions on Imge Processing, vol.4, no. 8, Aug. 995, pp [6] H. D.Cheng, Y.H. Chen, nd Y.Sun, A novel fuzzy entrpy pproch to imge enhncement nd thresholding, Signl Process., vol. 75, no. 3, pp. 77-3, Jun [7] M. Hnmndlu, S.N. Tndon, nd A. H. Mir, A new fuzzy logic bsed imge enhncement, Biomed. Sci. Instrum., vol.34, pp , 997. [8] M. Hnmndlu, D.Jh, R.Shrm, Color imge enhncement by fuzzy intensifiction, Pttern Recognition Letters, vol. 4, 3, pp [9] M. Hnmndlu nd D. Jh, An optiml fuzzy system for color imge enhncement, IEEE Trns. Imge Process., vol, no., pp , Oct. 6. [] M. Hnmndlu, Om Prksh Verm, Nukl Krishn Kumr nd Murlidhr Kulkrni, A Novel Optiml Fuzzy System for Color Imge Enhncement Using Bcteril Forging, IEEE Trns. on Instrumenttion nd Mesurement, vol., 58, no. 8, August 9. [] Mnish Kumr Sini, Deepk Nrng, Cuckoo Optimiztion Algorithm bsed imge enhncement, in Proc. of Int. Conf. on Advnces in Signl Processing nd Communiction, Elsevier, 3. [] M. Hnmdlu, Shvet Aror, Gurv Gupt, Ltik Singh, Novel optiml fuzzy color imge enhncement using prticle swrm optimiztion, in Proc. IEEE on Contemporry Computing, Sep.3 [3] J.Giffler, G.L.Thompson, Algorithms for solving problems, Oper Res., vol. 8, no. 4,pp , Aug. 96. [4] L. Shym Sundr Singh, Dr. Anil K Ahlwt, Dr. Kh. Mnglem Singh nd Th. Prmeshwor Singh, Imge Enhncement Techniques On Fuzzy Domin: A Review. Interntionl Journl of Computer Engineering & Technology, 8(), 7, pp editor@ieme.com