Autonomous foveating system and integration of the foveated images

As a human-like foveation system, we present an autonomous foveating system based on a pulse-coupled neural network (PCNN) and the integration system for foveated images. This foveating system spontaneously selects the foveation points from the edges and/or optical flows of the input images via the PCNN. The foveation point is defined as the point with the maximum output from the PCNN. Once the foveation point is chosen, the camera should be foveated to the foveation point. In our case, instead of using an actual camera, a log-polar sampled image called a foveated image, which imitates a retinal image, is used as a camera image at the foveation point. The whole-scene image, from which the foveation points are selected, is obtained through a fish-eye lens which gives an equidistant image. The integration algorithm for the foveated images requires two kinds of map which represent the equidistant images for each foveated image: one is the gray image map (G) for the foveated image itself and the other is its resolution map (R). If the resolution of a new foveated image at a cell is equal to, or finer than, the current value of R at the cell, the values of both G and R at the cell are updated by new values. Some experiments demonstrate the performance of the autonomous foveating system and the integration system