OBJECT SEGMENTATION USING REGION GROWING AND EDGE CONSTRAINTS

Abstract

This thesis focuses upon Object Segmentation which plays an important role in the field of computer vision. The image segmentation problem is concerned with partitioning an image into multiple regions according to some homogeneity criterion. Object segmentation is used to typically locate objects and boundaries in images. The proposed object segmentation method is an integration of region growing and edge information. This method automatically selects the initial seed and determines the threshold with the help of a 20x20 window across the center pixel for single seeded region growing. Automatic threshold is determined by the difference between mean and median of this window, whereas, minimum distance between mean and all pixels of window help in initial seed selection. The grown region is used for object segmentation by placing edge constraints over it to obtain nearest strong canny edges. Further, certain morphological operations are performed to obtain precise results. The proposed algorithm is applied to state-of-the art database PASCAL VOC 2005 and compared to the method proposed by Xavier Bresson based on the Active Contour model. This algorithm produces successful results and accompanying ground truth annotations helps in determining precision and recall parameters. On the basis of evaluation of these parameters it can be said that proposed method produce good segmentation results with high precision.