130 likes | 278 Views
Iterative Fuzzy Clustering Regions of Interest in Skin Lesions. By R. Cucchiara, C. Grana, M. Piccardi Presented by Mohammed Jirari October 23 rd , 2002 Image Processing Lab. Introduction. Definition of Skin Melanoma Systems used for Melanoma diagnosis Recent research.
E N D
Iterative Fuzzy Clustering Regions of Interest in Skin Lesions By R. Cucchiara, C. Grana, M. Piccardi Presented by Mohammed Jirari October 23rd, 2002 Image Processing Lab
Introduction • Definition of Skin Melanoma • Systems used for Melanoma diagnosis • Recent research
Method used in paper • Preprocessing • Karhunen-Loeve transform • Fuzzy c-means clustering • Topological tree
Preprocessing • Convolve the image with a Gaussian kernel with standard deviation of one pixel. • Transform images from original RGB into CIE L*a*b* color coordinates.
Karhunen-Loeve transform • Projection of the vectors to be reduced on the eigenvectors of their covariance matrix using the following:
Karhunen-Loeve transform (cont.) • The Karhunen-Loeve transform of vector x is defined by:
Fuzzy c-means clustering • Use the following 2 recurrent equations:
Topological tree • Bright clusterHealthy skin Dark cluster Lesion
Topological tree (cont.) • Def1:Skin region of Interest(Skin ROI) a set of pixels of the skin image exhibiting 3 properties: uniform color, connected pixels and significant area. • Def2:Topological Tree(TT) a tree whose nodes are skin ROIs and the arcs topological inclusion relationships between skin ROIs
Pseudo-code of the algorithm AnalyzeRegion (region R, node N) {if(not StopCondition (R)){ [C1,C2]=FCM(R); [Cint,Cext]=VerifyInclusion([C1,C2]); if(exists([Cint,Cext])){ Cres=R-Cint-Cext; Nnew=AddNodeToTree(Cext,N); foreach C in ConnectedComponents(Cint) AnalyzeRegion(C+Cres,Nnew);} else { Analyzeregion(R-C1,N); AnalyzeRegion(R-C2,N); }} else AddNodeToTree(R,N);}