Hierarchical spatiotemporal moving object segmentation based on a binary split algorithm

To segment moving objects in head-and-shoulder image sequences, a new morphological image segmentation algorithm is proposed using a hierarchical structure based on a binary split algorithm. The split method is performed using the combination of spatial and temporal similarities to reduce the residual errors between the original and segmented image. Spatial similarity is used to split those regions with the most different intensity values, while temporal similarity, based on the displacement vectors between two images, is used to split those regions with different motions. As a result, the proposed hierarchical segmentation method, based on a binary split algorithm, can increase the subjective image quality with only a small number of regions. Furthermore, a new contour simplification technique composed of three filtering steps is repeatedly applied to the segmented image, thereby eliminating any noisy contours of segmented regions. (C) 2003 Society of Photo-Optical Instrumentation Engineers.