Novel Texture Feature for Content Based Image Retrieval

Abstract

Content-Based Image Retrieval (CBIR) is an image retrieval technique that fetches relevant pictures from a huge image database based on the content feature of a given query image. It has many applications such as satellite image indexing, development of recommendation system in many applications and medical image processing. Feature extraction, indexing and similar image retrieval are few major steps in the development of CBIR system for any applications. Extraction of relevant feature that effectively represents the image is a challenging task. Therefore, the performance of a CBIR system completely depends upon the ability of the feature sets used to represent the image contents. In addition the dimensionality of the feature set used also affect the retrieval performance in terms of response time. The objective of this work is to analyze the retrieval performance of few recently proposed texture features and propose a novel feature extraction technique that has better retrieval efficiency while addressing the issues of dimensionality. In this paper we implemented and analyze the retrieval performance of three recently proposed texture features and also proposed a novel method for texture feature extraction for the design of CBIR system. The evaluation is performed using five popular publicly available benchmark data sets with varied complexities. The data sets include real-life images, texture images, facial images, etc. Extensive experiments are performed to tune the parameter that affects the retrieval performance for each texture feature considered for evaluation. The performance of the proposed method is compared with other methods in terms of average precession and average recall percentage. The performance of the proposed method is found to be best among all irrespective of the dataset used.