Effective image retrieval based on hybrid features with weighted similarity measure and query image classification

Abstract

An effective image retrieval needs efficient extraction of low-level features, and for this various methods have been recently proposed. Most of these methods use the histogram or some variation for representing colour and other descriptors which require significant amount of space and extra similarity calculation. Here, an efficient content based image retrieval (CBIR) system is proposed, which is based on the fusion of chromaticity-colour moments, and colour co-occurrence-based small dimension features using inverse variance weighted similarity measure. In this measure, property of the varying weights reduces the effect of redundancy and effectively retrieves relevant images. In addition, this paper also proposes a supervised query image classification and retrieval model by filtering out irrelevant class images using multiclass SVM classifier. Basically, this model recovers the category of query images, and this successful categorisation of images significantly enhances the performance and searching time of retrieval system. Descriptive comparative analyses confirm the effectiveness of this work. We obtained 83.83% and 76.9% average precision for 12 and 20 images retrieval using weighted similarity measure together with 85.6% average precision and 84.4% recall for classification framework.