Boundary extraction in texture mosaics using discontinuity information in feature space

In this study, boundary extraction between textures is examined. The overall system consists of three stages. In the first stage, the gradients in feature space are estimated using a modified version of gray-level edge detection operators. For comparison purposes, both the Prewitt and Sobel operators are used. The second stage involves application of a threshold value to obtain a binary image displaying edges found in the first stage. In the last and crucial stage, some morphological post-processing operations are applied on the binary edge image to remove spurious pixels inside regions and to thin the thick edges occuring due to both rough thresholding and the use of large displacement value in edge detection. To discriminate between textures, four different features are used. The first three features are the fractal dimension (FD) of original image, constrast-strecthed image and top-hat transformed image, respectively. The fourth feature is the entropy which is a parameter obtained from the spatial gray-level co-occurrence matrix of the image. The experimental results are presented for mosaics with different number of textures from Brodatz album.