Histogram correlation of the output from a small mask operator: a basis for adaptive texture segmentation

There appears to be no total practical solution to the problem of texture discrimination. Thus, grey level histogram correlation often works well, particularly for stochastic texture fields. But this method is less able to distinguish regions within structured texture fields. In the paper, the basis for a scheme is presented in which the histograms of local texture properties (grey level differences, including directionality) within windowed regions are correlated with the output from a sample window. The authors use the output of a novel texture discriminator based on encoded comparisons between adjacent pixels in a 2×2 neighbourhood. Results with test images, including those from the Brodatz album, suggest that the feature statistics operator would form the basis of a useful discrimination tool for structural textures in an adaptive system