Abstract:
This paper addresses the problem of locating an object in a 2D grey level image. The approach implements a temporal Kohonen map that self-organises input data based on the spatio-temporal properties. Images are divided into blocks that are represented as a sequence of pixel intensity values. After training, the network maps the blocks into different parts of the image, e.g. edges, background. The network is tested by a set of images where it outputs a sequence of labels corresponding to the neurons indices indicating the location of the object.