GPU Implementation of Haralick Texture Features Extraction Algorithm for a Neuro-morphological Texture Image Segmentation Approach

The goal of this paper is to present a parallel implementation of Haralick features extraction technique for our unsupervised texture image segmentation approach. The gray-level co-occurrence matrix (GLCM) and Haralick features are computed in parallel for each pixel of the Image, by using the CUDA environment on NVIDIA GPU. This enhanced implementation is followed by our clustering approach, which based on the representation of a Kohonen map trained by features parallel extracted from each pixel of the image, and extraction of modal regions from that map. The experimental results focus on the performance comparison between the parallel GPU implementation, and the sequential technique based on CPU. In addition, segmentation rate results obtained by applying our approach are compared to the result of the K-means method.