Iterative Segmentation Scheme from Motion Estimated as Optical Flow

In this paper we propose a segmentation scheme which uses motion information of optical flow to extract each region of moving objects separately. The optical flow can be represented by image velocity distribution which is calculated from two or more frames of motion image. Segmentation scheme based on motion image is expected to be essential in the application of motion recognition, orientation estimate and so on. Our segmentation technique consists of two sections. In the first stage, the M-estimator is used to divide the whole image into several pieces of regions each of which shearing a coherent velocity. In the second stage, the regions are re-combined to one of adjacent regions which has the most similar histogram of velocity components. Experimental results show that the proposed method improves both segmentation and optical flow estimation