EXPLORE PUBLICATIONS BY COUNTRIES

Relevance feedback, Similarity measures, Content Based Image Retrieval

In this paper we have developed an Interactive Content Based Image Retrieval System which aims at selecting the most informative images with respect to the query image by ranking the retrieved images. The system uses relevance feedback to iteratively train the Histogram Intersection Kernel Based Support Vector Machine Classifier. At the end of the training phase of the classifier, the relevant set of images given by the final iteration of the relevance feedback is collected. In the retrieval phase, a ranking of the images in this relevant set is done on the basis of their Histogram Intersection based similarity measure with query image. We improved the method further by reducing dimensions of the feature vector of the images using Principle Component Analysis along with rejecting the zero components which are caused by sparseness of the pixels in the color bins of the histograms. The experiments have been done on a 6 category database created whose sample images are given in this paper. The dimensionality of the feature vectors of the images was initially 72. After feature reduction process, it becomes 59. The dimensionality reduction makes the system more robust and computationally efficient. The experimental results also agree with this fact.