Résumé: Today, the Local Binary Pattern (LBP) has become one of the most widely used texture descriptors thanks to its invariance and efficiency. The basic LBP method encodes local features by considering the difference in the local neighbourhood to represent a given image using the binary pattern histogram. Without performing the histogram step, the LBP method could be used to detect edges in an image. In this paper, two algorithms for edge detection are proposed. They are based on modifying the principle of the LBP method where a local neighbourhood is coded in binary by integrating a criterion of its homogeneity. In this work, we define this criterion as the ratio of the total variation in the whole image to the local variation of the neighbourhood. Thus, a new approach of edge detection is presented in two versions according to the way of calculating the differences in a neighbourhood. Experimental results on a standard natural image database show that the two proposed algorithms significantly improve the MSE, PSNR and SSIM indicators of the famous Canny detector and the improved LBP approach. In noisy conditions, our proposed algorithms present a better robustness to three types of noise

Langue: Anglais