PURPOSE: To accurately find a speed field in a dynamic image by simple neural network constitution and to precisely and automatically extract a moving body area from the speed field.
CONSTITUTION: This processor is equipped with block division parts 3 and 6 which divide a current frame image and a next frame image of the dynamic image into plural blocks respectively, a block matching part 7 which inputs respective divided block images and calculates a bias input supplied to a neural network by block matching, the neural network 8 which is equipped with plural neuron blocks representing respective two-dimensional speed vectors of the block-divided images and constituted by mutually joining mutually adjacent neuron blocks and calculates a speed vector in parallel according to the bias input, and a speed difference threshold value processing part 12 which automatically extracts a moving body in frame images according to the found speed vector.
ISOBE YOSHINAO