A Local Patterns Weighting Approach for Optimizing Content-Based Image Retrieval Using a Differential Evolution Algorithm

In Content Based-Image Retrieval (CBIR), low-level visual characteristics like color, texture and shape are used to search for relevant images. However, the result images returned to the user are generally not satisfactory to his expectations. This is due to the gap between the low-level features of the image and the semantic (high-level) concepts given by the user to the same image. To overcome this challenge, we propose in this paper a mechanism that improves CBIR performance and consequently reduce the semantic gap. In that regard, our work involves the optimization of CBIR using a specific mechanism for weighting the extracted textural characteristics of the image. The extraction of the latter is carried out by some local patterns methods. Then, the generation of the weights associated with the local patterns, is realized using the Differential Evolution algorithm. To evaluate our approach, we tested it on Wang’s database (Corel-1K). In addition, we adopted the precision as performance evaluation measure and we used Manhattan and Euclidean distances for comparing the local patterns histograms. The results of the carried-out experiments show that the obtained precisions by the weighted local patterns methods are better than those of the conventional methods.