Enhancing performance of Image Retrieval Systems using Dual Tree Complex Wavelet Transform and Support Vector Machines

This paper presents a novel image retrieval system (SVMBIR) based on dual tree complex wavelet transform (CWT) and support vector machines (SVM). We have shown how one can improve the performance of image retrieval systems by assuming two attributes. Firstly, images that a user needs through query image are similar to a group of images with the same conception. Secondly, there exists non-linear relationship between feature vectors of different images and it can be exploited very efficiently with the use of support vector machines. At the first level, for low level feature extraction, we have used dual tree complex wavelet transform because recently it has been proved to be one of the best for both texture and color-based features. At the second level, to extract semantic concepts, we grouped images of typical classes with the use of one against all support vector machines. We have also shown how one can use a correlation-based distance metric for comparison of SVM distance vectors. The experimental results on standard texture and color datasets show that the proposed approach has superior retrieval performance over the existing linear feature combining techniques.