Affine parameter estimation with gradual segmentation

We propose an extraction method of the structural motion information for compact image representation. The advanced motion analysis needs segmentation and motion velocity information. Conventional segmentation methods by using edge detection with high-pass filtering tend to fall into over-segmentation. The morphological filter is used to obtain edge information avoiding the selection of small oscillation. In the filtering process, various sizes of structural element (SE) makes possible gradual segmentation with respect to the object size. Then in each divided region, the simultaneous equations of motion velocity are solved to get the affine parameter (AP) which can describe the object's movement of translation, rotation, expansion and reduction. If the intensity error in a region is larger than the threshold, a smaller structural element is used to segment the region. This enables one to get affine motion parameters without the waste of over-segmentation.