Improved fuzzy c-means and k-means algorithms for texture and boundary segmentation

Image segmentation is one of the most significant and inevitable task in variety areas ranging from face/object/character recognition and medical imaging applications to robotic control and self-driving vehicular systems. Accuracy and processing time of image segmentation processes are also prominent parameters for quality of such computer vision systems. The proposed method incorporates three main pre-processing techniques such as Down Scaling/Sampling, Gamma Correction and Edge Preserving Smoothing so as to achieve accuracy and robustness of the segmentation. Pre-processing techniques are performed for both Fuzzy C-means (FCM) and K-means algorithm and all RGB information of image are taken into consideration while segmenting the image rather than using only gray scale. Performance analysis are performed on real-world images. Experiments show that, our method achieve higher accuracy levels and feasible processing time results compared to conventional FCM and K-means algorithms.