Approach of Using Texture and Shape for Image Retrieval

Introduction

Ever since the 1970s, the further development has been made in content-based image retrieval. The basic idea of content-based image retrieval is to extract and describe the discriminative features from an image, and use these features to index the image. Then similarity measure is used for indexes to locate the same or similar images in an image database. Feature extraction and description is one of important components in content-based image retrieval, and is used to extract a set of discriminative features from an image to describe the content in the image. How these features are described and organized effectively is emphasized to describe the content in the image better. Furthermore, it will affect similarity measure directly. Currently, low level visual features are usual, and can be divided into texture features, shape features, color features, etc. According to low level visual features, the feature extraction and description approaches can be divided into texture feature extraction and description approaches (Haralick et al, 1973; Won et al, 2002; Wu et al, 2000; Jhanwar et al, 2004; Shi et al, 2005), shape feature extraction and description approaches (Zhang & Lu, 2002; Mokhtarian & Bober, 2003; Teague, 1980; Xin et al, 2007), color feature extraction and description approaches (Swain et al, 1991), etc. As being carried out for single feature, these approaches are called single feature extraction and description approach.

Texture and shape features are the commonly used low level visual features. Texture can describe the details of object surface, and shape can describe the contour of object effectively. That texture and shape are used together can describe the features of the image more effectively. Usually, texture feature extraction and description approaches can be divided into structure approaches, statistic approaches, model approaches, and transform approaches (Materka & Strzelecki, 1998). Gabor wavelet transform is one of commonly used transform approaches. The discovery of orientation-selective cells in the primary visual cortex of monkeys almost 40 years ago and the fact that most of the neurons in this part of the brain are of this type triggered a wave of research activity aimed at a more precise, quantitative description of the functional behavior of such cells (Hubel & Wiesel, 1974). Gabor wavelet transform is a computational model which simulates the principle of operation of these cells. Its basic idea is to use Gabor function as mother wavelet to compute a set of wavelets where each wavelet captures energy of specific frequency and direction in a window. Then energy is used for texture features to index an image. Texture features which are invariant to direction and scale can be extracted and described by the invariance of Gabor function to rotation and of wavelet transform to scale. The wavelet function set from the transformation of mother wavelet forms a set of non-orthogonal basis in Gabor wavelet transform approach, which means that redundancy information are involved in computation of texture features in the feature extraction and description (Arivazhagan et al, 2006). To reduce the redundancy information, a set of orthogonal wavelet function basis is introduced into Gabor wavelet transform approach (Manjunath et al, 2000; Ro et al, 2001). The Gabor wavelet transform approach is of minimum information redundancy, and used for the texture feature extraction and description in the paper.