IMAGE PATTERN ANALYSIS WITH IMAGE POTENTIAL TRANSFORM

Pattern analysis with image transform based on potential calculation was considered. Initial gray-scale image is sliced into equidistant levels and resulting binary image was prepared by joining of some levels to one binary image. Binary image was transformed under assumption that white pixels in it may be considered as electric charges or spins. Using this assumption Ising model and Coulomb model interaction between white pixels was used for image potential transform. The transform was calculated using moving window. The resulting gray-scale image was again transformed to binary image using the thresholding on 0.5 level. Further binary images were analyzed using statistical indices (average, standard deviation, skewness, kurtosis) and geometric signatures: area, eccentricity, Euler number, orientation and perimeter. It was found that the most suitable geometric signature for pattern configuration analysis of Ising potential transform (IPT) and Coulomb potential transform (CPT) is area value. Similarly the most suitable statistics is distance statistics between white pixels.