A Hybrid PSO and Active Learning SVM Model for Relevance Feedback in the Content-Based Images Retrieval

Relevance feedback (RF) based on Support Vector Machines (SVMs) has been widely used in the Content-based image retrieval (CBIR). However, three problems are confronted: how to choose the optimal input feature subset, how to set the best kernel parameters, and the training data is scare in the RF procedure. To address those problems, an improved relevance feedback system based on hybrid PSO and active learning SVM model was proposed in this text. In the new model, the PSO with/without feature selection can optimal the parameters ( and ) and sub-features in the SVM classifier. And, the active SVM was applied on actively selecting most information images that minimizes redundancy between the candidate images shown to the user. The experimental results show the proposed approach has the speedy convergence and good results in the relevant feedback system.