An Integrated Framework for Designing Content based Image Retrieval System using Support Vector Machine with Enhanced Biased Maximum Margin

Abstract

With the growth of new era and development a wide variety of applications are being developed in the field of CBIR. There are different applications of relevance feedback approaches to improve the performance of image retrieval. The main concept of using relevance feedback is to overcome semantic gap between low level features and high level concepts. Among the most popular approaches to RF support vector machine is the key to interactive image retrieval. The reason to employ SVM with RF is to make image retrieval more efficient. Despite of the fact that SVM improves the performance efficient methods and techniques are still needed to be developed that can efficiently retrieve more relevant images. The main problem with the RF approach with SVM is it considers positive and negative samples equally. The two samples are different in their properties and cannot be treated equally. In this paper we will present enhanced biased maximum margin support vector machine. In this method negative samples of RF approach are efficiently sampled by shared nearest neighbor and it is optimized by ant colony optimizer with biased maximum margin. This concept mainly maps positive samples closer and negative samples are distinguished from them by maximum margin. It thus improves the performance by bringing most relevant images nearer less relevant images farer.