In this thesis, we describe motion parameter coding methods for region based video coding, where the motion of the region is described by using the affine motion model. Because the estimated coefficients are real-valued, we have to quantize them. Also, there are poor correlation and similarity between coefficients of neighboring regions, and each coefficient has different statistical property and dynamic range. To overcome these problems, the triangle method is known to be an effecient method compared with direct quantization.
We present two methods for motion parameter coding in this thesis. First, to improve the triangle method in region based coding, we investigated the effects of the position and the size of the triangle, and proposed a DPCM method between displacement vectors in one triangle. Second, we convert conventional six-parameter to real translation, scaling, and rotation information. The obtained translation motion is differentially coded in temporal direction. Then the rotation and scaling information is coded using vector quantization. Simulation results show that the proposed schemes provide better performance than the direct quantization method.
