Algorthm for Human Skn Detecton Usng Fuzzy Logc Mrtunjay Ra, R. K. Yadav, Gaurav Snha Department of Electroncs & Communcaton Engneerng JRE Group of Insttutons, Greater Noda, Inda er.mrtunjayra@gmal.com ravpusad@gmal.com gsnha78@gmal.com Abstract Ths paper deals wth algorthm for detectng and extractng of human skn regon usng Fuzzy Logc and skn segmentaton from a vdeo. Before the detecton of skn regon the detecton of change of shots s to be doneand then extracton of mage (f any) s done from detected shots. Moreover ths paper also explans the method of nterpolaton n order to obtan the hstogram plot. The values obtan from hstogram plot wll provde the exact frame number where the change of shots s takng place. Once the frames s detected from change of shots t wll got converted nto HS color model and then skn segmentaton usng Fuzzy logc s to be done. The whole paper explans the mportance of Fuzzy logc n deo Survellance system. Keywords shot detecton, nterpolaton, skn segmentaton, skn regon extracton, Fuzzy logc I. INTRODUCTION deo survellance system based on magerecognton s wdely appled to areas ncludng nobody defends and ntellgent traffc, whch not only could effectvely prevent many crmes, but also s the mportant hgh tech method and the technology to guarantee socety publc securty. Face skn colour s the mportant tem of human faces. The frst man step n detectng skn pxels s to classfy as skn colour or non- skn colour. A proper colour model s needed to perform the classfcatons. The HS colour model s popular when compared to RGB or YCbCr colour models because t s compatble wth human colour percepton. The second step wll be connectvty analyss to confrm a skn pxel or not, the thrd and fnal man step s to fnd an optmal boundary box to extract the face from the nput mage.in ths paper, we present a fuzzy logc based skn detecton system to detect the presence of human skn n a survellance vdeo. The paper s organzed as follows: In the secton 2 detals of shot detecton process s provded by usng hstogram dfferences. Smulaton results show that the system can produce accurate shot change detecton, whch s very useful for vdeo retreval. Secton 3 provdes a complete objectve assessment framework. More specfcally, the process to select the key frame on measures of pxel value of every shot. In Secton 4 the fuzzy logc rules [1] appled for skn colour dscrmnaton are presented, and techncal detals are brefly dscussed. Fnally, at the end, an evaluaton of the overall performance of ths approach as well as ts lmtatons takes place n the last sesson. II. PROPOSED ALGORITHM FOR SHOT DETECTION USING HISTOGRAM DIFFERENCE A.HS Hstogram Dfference HS colour model has a good lnear scalablty, whch can understand the colour dfference and colour corresponds to the knd of weght value proportonal to the Eucld dstance, so ths model more acceptable to human than the RGB colour model, n order to get best effect the RGB space s changed nto HS space frst.as the three colour components of the HS colour space are ndependent, frst, we compute the hstogram dfferences of consecutve vdeo frames on each colour component respectvely. And then, the maxmum hstogram dfference [2] s used as the detecton feature. Where, H D b1 h HD = max(h,s, ) H h H j. (1) S D b1 h S h S j. (2)
D b1 h h j H, S, are the hstogram dfferences of H,S, component respectvely. D s the number of bns, and j are the frame number.. (3) The flowchart of shot detecton process s shown n Fg.1. Fgure 1: Flowchart of Shot Detecton process B.Shot Detecton by HS Hstogram Dfference A shot s detected when the HS hstogram dfference of frames s hgher than the threshold. Here the threshold value s(m*n/25).where m and n are number of rows and columns neach frame. Fg. 2 shows three transton frames from a vdeo. Fgure 2: Three dfferent detected shots from a vdeo All the frames lyng between the two transton frames are consdered as a sngle shot. Ths technque works effcently for hard transton, but t falls mserably when the vdeo contans gradual transtons. Dfferent gradual transton methods proposed by dfferent researchers are fade-out/fade- n detecton [3], wpe detecton [4],Dssolve detecton [5]. In ths work a gradual transton method based on nterpolaton s proposed whch can handle the gradual transtons wth reasonable accuracy. Instead of consderng only two conjugatve frames for smlarty measurement, current frame and ts past frames are used to fnd the smlarty. C.Interpolaton Method: The proposed nterpolaton method s explaned as follows. Fg. 3 shows four nterpolated conjugatve frames. Fgure 3: Interpolated frames upto four frames Let n, n- 1, n- 2 and n- 3 be the feature vectors of the present and prevous consecutve frames. n n 1 n2 n3 avg (4) 4 t n 4 1.5t 0.5t 0.5t 1 avg n1 avg avg n2 avg n3 1.5t (5)
1 6t 3 n n1 3 n2 n3 (6) Where t s the tme dfference between two conjugatve frames If the dfference only larger than the threshold, t should be detected by other algorthms as follows to decde whether a shot change occur or not. III.REFFERNCEFRAMESELECTIONFROMASHOT In ths work key frames [6] are selected by the mean value of colour pxels. We have calculated the mean value of Gray level and R, G and B level. Then we fnd the frame whch one has the closest value of the mean. In the Fg 3 flowchart of frame selecton has shown, Fgure 4: Flowchart of Frame Selecton After gettng the frames of dfferent levels we wll choose that partcular frame whch one s selected as mean n most of the levels. Fg. 4 shows a dagram of frame selecton procedure. Fgure 5: Key Frame of a Shot and ts Selecton I.HUMAN SKIN DETECTION USING FUZZY LOGIC The colour of human skn [7] s totally dfferent from the colour of many other objects and for ths reason the statstcal measurements of ths area are of very bg mportance for face detecton, gesture recognton, and personal dentfcaton. Evaluatng the skn tone (colour) statstcs, t s expected that the human skn colour tones wll be dstrbuted over a dscrmnate space n the RGB plane. Stll now there are many approaches n the lterature used smlar detect procedures ether based on the RGB, chromnance (CbCr)or Hue and Saturaton (HS) space.the man am of ths system s the locaton of human skn areas n the mage usng fuzzy logc n HS space. Fgure 6: Fuzzy Logc Implementaton n Skn Color Detecton
x=0:0.1:10; y=trapmf(x, [1 5 7 8]); plot(x, y) xlabel('trapmf, P=[1 5 7 8]') Fuzzy logc have been successfully appled n varous felds [8] such as automatc process control system, classfcaton of dfferent data, decson analyser expert systems [9], and computer vson. Fuzzy logc s the way of generatng the wde area of mappng from a gven nput to an output usng fuzzy logc. The fuzzy rules are characterzed by a collecton of dfferent fuzzy membershp functons [10] as shown n fgure 6, varous types of logcal operatons, and IF-THEN rules. In the proposed system fuzzy logc s workng upon the mage lumnance or the mage ntensty. In the Fg 5 the processng of fuzzy logc has shown by a flowchart Fgure 6: Flowchart of Skn Detecton usng Fuzzy Logc Fuzzy logc s an extenson of conventonal methods that deals wth the concept of potental truth. It s used membershp functon to measure the problem as human mnd. In the Fg. 6 whch s shown below, orgnal mage and detected skn usng fuzzy logc has provded Fgure 7: Orgnal Image and Output Image after Skn Detecton usng Fuzzy Logc.PERFORMANCE EALUATION The system has used hstogram based algorthm to detect vdeo shots, whch has gven better result compare to othertechnques lke contour based or moton based[11].the changeof hstogram can be detected easly whch s shown n thefgure 7 Fgure 8: Hstogram Generaton from a deo In ths paper we used hstogram dfferences of dfferent frames for shot boundary detecton [12].By usng ths method we have detected dfferent vdeo shots. We have used 10 dfferent vdeos as nput and after gettng and
comparng all the results the all over result has been made. The result has shown below by usng two dfferent tables, whch s showng the percentage of accuracy for dfferent output. TABLE 1. Accuracy Measurement for Dfferent Shots Hard Cut Detecton Gradual Transton Detecton Fade Change Detecton 89% Accuracy 88% Accuracy 68% Accuracy TABLE 2. Accuracy Measurement for Overall Detecton No shot Change No Detecton 85% Shot Change and Detected 83% No shot Change but Detected 35% Shot Changed but Not Detected 37% After the frame selecton fuzzy logc s used for skn detecton. Usng the fuzzy membershp functon n HS planewe get detected human skn. There are some outputs n thetable whch s shown below. TABLE 3: Detected Human Skn Usng Fuzzy Logc Orgnal Images ImagesAfter Skn Detecton I.CONCLUSION AND FUTURE WORK An effcent algorthm for human skn detecton based on fuzzy logc from vdeo has been proposed and the overall performance of the system s found to be qute satsfactory on dfferent vdeos. Interpolaton technque proves to be a better approach for gradual transton detecton n comparson to the other exstng technque [5, 6]. Fuzzy approach s effectvely retrevng the skn regon from the selected frame. But there are several areas on whch further development can possble. The algorthm s tested on the vdeos where the lghtng condton s proper. The algorthm dd not work effcently when the vdeo s taken n under/over lghtng condton. Secondly snce the skn color vares from race to race, neuro fuzzy technque may work more effcently for dfferent races. II.REFERENCES [1] J. Jantzen, Desgn Of Fuzzy Controllers, TechncalUnversty of Denmark, 1998. [2] Zhou Shunyong&XeWenlng A System of deo Shot Detecton Usng Mult-stage Algorthm Informaton Technology and Computer Scence, 2009.ITCS 2009. Internatonal Conference on Dgtal Object Identfer: 10.1109/ITCS.2009.121 Publcaton Year: 2009, Page(s):558-561 [3] R Lenhart, Relable transton detecton n vdeos: a survey and practtoner s gude.internatonal Journal on Image Graphc,vol. 1,no. 3,pp 469-486,2001. [4] Y.G.Jan and C.W.Ngo Towards Optmal Bag-of-Features for Object Categorzaton and Semantc deo Retreval,n Proceedngs of ACM Internatonal Conference on Image and deo Retreval,pp 494501,2007 [5] R Lenhart, Relable Dssolve detecton, n Proc. SPIE Storage Meda Retreval Meda Database, Jan 2001,vol. 4315,pp. 219-230. [6] Wayne Wolf Key Frame Selecton by Moton Analyss Acoustcs, Speech, and Sgnal Processng, 1996. ICASSP-96. Conference Proceedngs., 1996 IEEE Internatonal Conference on olume:2 Dgtal Object Identfer: 10.1109/ICASSP.1996.543588.Publcaton Year: 1996, Page(s): 1228-1231
[7] Pham The Bao,Jn Young Km &Seung You Na FAST MULTI-FACE DETECTION IN COLOR IMAGESUSING FUZZY LOGIC, Proceedngs of 2005 Internatonal Symposum on Intellgent Sgnal Processng and Communcaton Systems,Decmber 13-16,2005,Hong Kong [8] M.H. Yang, D.J. Kregman, N. Ahuja, "Detectng Faces n Images: A Survey," IEEE Transactons on pattern analyss and machne ntellgence, vol. 24, No. 1, January 2002 [9] Chun-Mng Tsa and Zong-Mu Yeh Contrast Compensaton by Fuzzy Classfcaton and ImageIllumnaton Analyss for Backlt and Front-lt Color Face Images IEEE, ol. 56, No. 3, August 2010 [10] Takag T and Sugeno M., Fuzzy dentfcat on of systems and ts applcatons to modelng and control, IEEE Trans. on Systems, Manand Cybernetcs.SMC-15 (1) pp.116-132.1985. [11] G. Lupatn, C. Saraceno& R. Leonard Scene break detecton: a Research Issues In Data Engneerng, 1998. 'Contnuous-Meda Databases and Applcatons'.Proceedngs., Eghth Internatonal Workshop on Dgtal Object Identfer: 10.1109/RIDE.1998.658276 Publcaton Year: 1998, Page(s): 34 41. [12] Xang Fu and Je-xanZeng An Effectve deo Shot Boundary Detecton Method Based on the Local Color Features of Interest Ponts Electronc Commerce and Securty, 2009. ISECS '09. SecondInternatonal Symposum on olume:2 Dgtal Object Identfer:10.1109/ISECS.2009.140 Publcaton Year: 2009, Page(s): 25-28.