Modes Detection of Color Histogram and Merging Algorithm by Mode Adjacency Graph Analysis for Color Image Segmentation
Résumé
In this work we present an approach for color image segmentation based on pixel classification. Such methods are based on the assumption that meaningful regions are defined by homogeneous colors and give rise to compact clusters in the color space. Each cluster defined a class of pixels which share similar color properties The construction of the pixel classes is performed by detecting the modes of the color histogram of the image. To identify these modes, mathematical morphology techniques are used. The application of watersheds on the color histogram leads to an over partitioning of the color plane, which can be processed by mode merging algorithms based on mode adjacency graph analysing. Depending the merging criterion we present in this paper two merging algorithms, the first relies on the gravity centers of the modes as a merging criterion, and in the second we introduce a new merging criterion: the spatial-color compactness degree.