Novel approaches in morphological correlations

Abstract

Morphological correlation is a novel method for obtaining high discrimination ability in pattern recognition applications. It provides also important abilities for image compression and image analysis. The concept is based on slicing the input image and the reference filter into many binary slices, e.g. 255, and correlating them. The morphological correlation is defined as the summation of these correlations. The morphological correlation is characterized by a sharp correlation peak narrower than that exhibited by matched filter. The disadvantages are the requirements of performing many correlations and its very high sensitivity to noise added to the reference image. In this presentation we suggest two methods to solve both drawbacks. First, instead of 255 correlations we suggest to utilize only 8, by representing the grey level of each pixel by its 8 bit binary representation. Then, 8 binary masks are constructed according to the binary representation. In order to address the problem of severe sensitivity to noise, we suggest to sum the 255 correlations of the morphology slices while each slice is multiplied by a weighting factor which equals the correlation peak of that specific slice with noise divided by its correlation peak value when no noise is added. The solutions suggested here were examined by computer simulations demonstrating considerable improvements in the performance of the morphological correlator.