Object Segmentation by an Automatic Edge Constrained Region Growing Technique

Region growing is a well known technique for image segmentation. However it suffers from certain limitations such as ambiguity regarding initial seed selection and threshold determination for region growing. In this paper we explore the possibility of using region growing technique for segmenting out the foreground object and propose to solve the imperfect results by incorporating the nearest edge information. The whole procedure is rendered automatic by defining a center window containing some part of the object. From this window the seed selection and threshold value for region growing is determined by simple computations. The results on the ´Car´ category of the challenging PASCAL VOC 2005 dataset indicate a superior performance to Xavier Bresson´s method for object segmentation.