A RandomWalkProcedure forTexture Discrimination HARRYWECHSLERAND MASATSUGUKIDODE, MEMBER,IEEE
Harry Wechsler, Masatsugu Kidode · 1979
We consider theproblem oftexture discrimination. Ran- domwalks areperformed inaplain domain D bounded byanabsorbing boundary r andtheabsorption distribution iscalculated. Measure- mentsderived fromsuchdistributions arethefeatures usedfordiscrim- ination. Bothproblems oftexture discrimination andedgesegment detection canbesolved using thesamerandomwalkapproach. The border distributions andtheir differences withrespect toahomoge- neousimagecanclassify twodifferent images ashaving similar ordis- similar textures. Theexistence ofanedgesegment isconcluded ifthe boundary distribution foragiven window(subimage) differs signifi- cantly fromtheboundary distribution forahomogeneous (uniform grey level) window. Therandomwalkprocedure hasbeenimplemented andresults oftex- turediscrimination areshown.A comparison ismadebetween results obtained using therandomwalkapproach andtherirst- orsecond-order statistics, respectively. Therandomwalkprocedure isintended mainly forthetexture discrimination problem, anditspossible application to theedgedetection problem (asshowninthis paper) isjust aby-product. IndexTerms-Digital image processing, edge detection, pattern recog- nition, texture discrimination.