1996 年 50 巻 3 号 p. 373-379
The method proposed in this paper uses the area of the region around each pixel with small matching errors as a criterion for motion estimation and segmentation. This area is calculated for each displacement vector through directional distance transformation of a binary image obtained by threshoulding the matching error at each pixel. On the basis of this area, the proposed method selects candidate vectors at each pixel and determines one of them to be the motion vector by majority rule in a local region. As a result, an image is divided into regions with uniform motion. This method has the advantage of being insensitive to surface textures since it uses only matching errors. The simulation results show that this method can perform well both motion estimation and segmentation even along motion boundaries when regions with coherent motion are relatively large.