GPU Accelerating for Color-Texture Image Clustering Based on Neuro-Morphological Approach

In this paper, we present a GPU acceleration of color-texture features extraction method, followed by a proposed neuro-morphological segmentation approach. in fact, we extract both texture and color information from each pixel of the image, the extraction phase is executed on an NVIDIA GPU using the CUDA environment to compute the color-texture features in parallel for each pixel. In this study, we chose to use Fractal and Haralick texture features combined with color features presented in RGB color space. The proposed neuro-morphological segmentation approach is based on the presentation of these features on a Kohonen map, followed by a morphological segmentation of this map using our watershed transformation. In addition to validating this segmentation approach by comparing it with k-means, the experimental results highlight the performance gap between the proposed GPU-based implementation of the extraction phase and the sequential implementation based on CPU.