On the Gabor-space geodesic active contours algorithm

Texture segmentation algorithm inspired by the approach outlined in [1] is presented. We represent our image as a two-dimensional manifold, embedded in a higher dimension feature space. Then we obtain our feature space via the convolution of the image with a set a Gabor wavelets tuned to the set of orientations, scales, and modulating frequencies. After that, we derive an induced metric via the pullback procedure. We smooth and regularize our metric via the coupled Beltrami flow and convolution with a large Gaussian kernel. Then, we use an evolution equation derived in the context of Osher-Sethian formulation to extract an edge detector. In our work we pay most of attention to the actual implementation of the algorithm.