Video analysis and synthesis based on a retinal-inspired frame

This paper introduces a novel retinal-inspired filter which is applied on video streams. We mathematically prove that under specific assumptions the spatiotemporal convolution turns into a spatial convolution with a short lifespan temporal kernel. As a consequence, the filter is applied on each image of the video stream separately. We analyze how each image is decomposed into a group of subbands, each one of which approximates the image providing different kind of information. Afterwords, we propose an algorithm to reconstruct each image by exploiting the group of subbands. Finally, we defend our mathematical proofs by providing numerical simulations which show the relevance of our study.