A neural algorithm for variable thresholding of images

A two-stage thresholding for gray scale images is presented in this paper. The first stage is based on a conventional application of the histograms which provides fixed global threshold value. This threshold value is then assigned as the initial state of a set of neurons which will process the image in parallel, in a horizontal scan, producing the binary image at the output. The state of the neurons is updated using the Kohonen self-organizing learning algorithm. This technique has two properties, First it smooths the spike noise, and second the low frequency illumination variation is compensated for and the segmented binary image regions are not affected by lighting conditions. Several examples are processed and presented to show the performance of the algorithm