Image processing application using CNN with dynamic template

Abstract

In this research, we propose an image processing application using cellular neural networks (CNN) with dynamic template (D-CNN). In D-CNN, the wiring weights of template are dynamically changed at each update by learning. In this study, we also focus on the number of converged value. Thus, some converged values of each cell in the initial input image are changed by effect of next input image in motion picture. Although input image is only still image, we think that if multiple input images are inputted to the D-CNN as motion picture, converged values of each cell are continued to changed. We investigate the converged value by changing input image. The results indicate that D-CNN can be applied to the region segregated processing for the motion picture.