A modified method for relevance feedback in high-resolution SAR image retrieval system based on SVM

Abstract

Relevance feedback (RF) is an importance technique in CBIR (Content-Based Image Retrieval) systems to bridge the semantic gap between low-level visual features (eg. color, shape, texture) and high-level human perception. One of the most frequently used methods to do RF is Support Vector Machine (SVM), which has a good generalization ability in pattern recognition. But when the training data is insufficient, the performance of SVM may drop dramatically. In this paper, we proposed a method to alleviate the small sample problem in SVM based RF by using a new piecewise similarity measure function and ensemble learning. We compared our method with standard SVM based RF on a high-resolution SAR (Synthetic Aperture Radar) image database, the experiment results show that our method has a better performance and prove that it's an effective algorithm for RF.