時空間フィルタを用いた運動視知覚のモデル : 画像表示 : 視覚情報 : 画像通信システム(<特集>視覚と画質)

抄録

Several recent studies have proposed models for visual motion perception using spatiotemporal filters. Among these, a model proposed by A.B.Watson and A.J.Ahumada model calculates two dimensional velocity, while others infer only one dimensional motion direction. However, the Watson/Ahumada model has limited accuracy for velocity estimation, mostly because it essentially uses only one spatial frequency channel. In human visual motion perception, it has been shown by various psychophysical studies that there are interactions between spatial frequency channels. In the present study, therefore, we propose an improvement of the Watson/Ahumada model by combining outputs from multiple spatial frequency channels for estimation of two dimensional velocity. The effectiveness of this multiple frequency method was demonstrated with moving random dot patterns. The outputs from three channels were additively combined using the following method: The straight lines representing the locus of possible velocity solutions for each spatial frequency channel were first plotted independently on one velocity field, then local velocity was determined by finding intersections of these lines. Combination of three frequency channels using this simple additive method resulted in an improvement of accuracy in velocity estimation. The positive results achieved here indicate that further implementation of more elaborate interactions between channels will provide even better performance.