Improving Cassification Engine in Content Based Image Retrieval by Multi-point Queries Via Pareto Approach

Machine learning methods have demonstrated promising performance for Content Based Image Retrieval (CBIR) using Relevance Feedback (RF). However, a very limited number of feedback images can significantly degrade the performance of these techniques. In this work, each image is represented by a vector of multiple distance measures corresponding to multiple features. Each feature is considered a sub-query for RF process. In RF process, we propose to use Pareto method to get Pareto points (also called trade-off points) according to different depths. These points are used as relevant queries for the next RF round. Experimental results show that our proposed approach is very effective to improve the performance of the classification engine.