Improving the Performance of CBIR with Genetic Approach and Feedback
Youssef Bourass, Abdelkhalak Bahri, Hamid Zouaki · Lecture notes in electrical engineering · 2016
Today, mass production multimedia is observed with high variability of form and content, which complicates the management of databases. The use the research techniques based on content has become increasingly necessary rather than the metadata, such as keywords, tags, or descriptions associated with the image. To dynamically associate the most appropriate search technic to each type of image, an intelligent system is required. However, it is very difficult to determine the adequate descriptor and distance for the analysis of a given image, the system quickly becomes unstable. In this paper we develop an application for the implementation and test of the most classic color and texture descriptors, in order to combine them using Entropy Impurity and Mutation approach. Our objective is to increase system performance and stability. Our application is based on a web interface, able to perform an experimental comparison of several methods used in image retrieval, in terms of accuracy and relevance of texture and color descriptors. Distances, between different descriptors are also calculated for four references of multimedia databases.