A Texture Feature Extraction Technique Using 2D–DFT and Hamming Distance

Texture analysis plays an increasingly important role in computer vision. Since the textural properties of images appear to carry useful information for discrimination purposes, it is important to develop significant features for texture. This paper presents a novel technique for texture extraction and classification. The proposed feature extraction technique uses 2D–DFT transformation. A combination of this technique and a Hamming Distance based neural network for classification of extracted features is investigated. The experimental results on a benchmark database and detailed analysis are presented. 1. Background Texture analysis has a wide range of applications. Millions of digital images are created throughout the World Wide Web, digital cameras, different kinds of sensors, medical scanners etc. Image analysis is based on three main image features: colour, shape and texture. Texture plays an important role in human vision. Texture has been found to provide cues to scene depth and surface orientation. Researchers also tend to relate texture elements of varying size to a reasonable 3-D surface. Although textured image analysis has been a topic of research for the last few decades [1-12], due to the complexity and the lack of ability to clearly define the significant features of texture, a number of challenging problems still need to be addressed. Features that have been used to describe texture images include simple mean and standard deviation, Gabor transforms, wavelet-based features, and Fourier transform based features [511]. In this paper, we propose a feature extraction technique, which uses a 2D-Discrete Fourier Transform (2D-DFT) and investigate it in conjunction with a novel Hamming Distance based neural network to classify the texture features of the images. The proposed feature extraction technique was implemented and tested on the Brodatz benchmark database [12]. 2. Research methodology This section describes in detail the proposed technique for feature extraction and classification. The overall block diagram of texture feature extraction and classification of these features is presented in Figure 1. The proposed technique is divided into two stages. Stage 1 deals with image segmentation and feature extraction from the texture images. Stage 2, deals with classification of features into texture classes. The texture database used to check the proposed technique consists of 96 different texture images. Each image is 512 x 512 pixels in size. The collection of Brodatz textures consists of textures of both a statistical and structural nature. Proceedings of the Fifth International Conference on Computational Intelligence and Multimedia Applications (ICCIMA’03) 0-7695-1957-1/03 $17.00 © 2003 IEEE