Improved fractal geometry based texture segmentation technique

Abstract

The problem of natural texture segmentation is considered. The technique is based on four texture features derived using the fractal geometry of images. These four features are fractal dimension (FD) of the original image, FD of above average (high) gray level image, FD of below (low) gray level image, and multifractal of order two. A modified box-counting approach is proposed to estimate the FD and computed features all are normalised in the same range [0, 1]. A feature domain smoothing is activated to reduce the spurious segmentation. Next, a nonsupervised clustering approach is used to segment a scene into the desired number of classes. Some supervised techniques like minimum distance classifier and k-nearest neighbour classifier are also considered. Mosaics of various natural textures are generated and the segmentation results are presented to show the efficiency of the technique.