Thesis (M.Sc.) - Cairo University - Faculty Of Computers and information - Department Of Information Technology

The purpose of this thesis is to explore the effectiveness of introducing fuzzy logic to the process of image thresholdingCrisp thresholding techniques always fail to segment fuzzy images (where object and background share common gray values) One of the fuzzy logic theory contributions in the area of image segmentation was the measure of fuzzinessMeasures of fuzziness can be used in the representation of knowledge about uncertain variablesVarious measures of fuzziness have been reported for image thresholding such as local and conditional entropies , fuzzy correlation and divergenceAnother aspect we covered in this thesis was the effectiveness of considering other image attributes (egtexture , local average{u2026}etc) during the thresholding processOne of the main problems of thresholding techniques occurs when a pixel's gray level is randomly distributed across the image (ieobjects and background share common gray level values) In such fuzzy images , considering only the gray level attribute may give unsatisfying resultsHowever , considering other image attributes in connection with pixels' gray level may improve the segmentation resultsAdding such attributes to the image histogram yields a 2 - D (or higher) histogram with a dimension represents gray values and the other represents the added attributeUsing 2 - D histogram , pixels having same intensities but different spatial features can be differentiatedOne of our contributions in this study was the introduction of the concept of line thresholding to fuzzy segmentation processMost 2 - D histogram thresholding techniques segment the histogram using the intersection point (t1 , t2) yielded from the two thresholds on both dimensionsAny image pixel having features values greater than t1 on the first dimension and t2 on the second dimension is considered as belonging to object class , and any pixel having features values less than t1 and t2 is considered as belonging to background class (assuming light object on a dark background) Such classification may ignore pixels having feature values greater than the first threshold but less than the second threshold or vice versaThese unclassified pixels are considered as noise pixelsThresholding the 2 - D histogram using only the point (t1 , t2) is called point thresholding and has the serious problem of arbitrary classifying noise pixels where those pixels that do not satisfy object's or background's thresholding conditions are arbitrary assigned to one of the two classesUsing point thresholding for 2 - D histogram segmentation often yields objects

Issued also as CD