An efficient computation method for the texture browsing descriptor of MPEG-7

Abstract

In this paper, an efficient computation method for computing the texture browsing descriptor of MPEG-7 is provided. Texture browsing descriptor is used to characterize a texture's regularity, directionality and coarseness. To compute the regularity of textures, Fourier transform is first performed. To get more discriminative features for regularity computation, the Fourier spectrum is treated as an image and the Fourier transform is performed again to produce an enhanced Fourier spectrum. A regularity measure based on the variance of the radial wedge distribution is then calculated to determine the regularity of textures. For regular textures, the texture primitives are assumed to be parallelograms, the two dominant directions are extracted by Hough transform. A scale computation method is then provided to determine the scales corresponding to the two dominant directions. In addition, principal component analysis is provided to detect textures with only one dominant direction. Experiments of texture browsing, coarse classification of textures and similarity-based image-to-image matching are performed on the texture images of Brodatz album and Corel Gallery image database to demonstrate the efficiency and effectiveness of the proposed method. The proposed method can be used in the applications of texture browsing and texture retrieval.

Introduction

Texture almost presents everywhere in natural and real world images. Texture, therefore, has long been an important research topic in image processing. Successful applications of texture analysis methods have been widely found in industrial, biomedical and remote sensing areas. Thus, appropriate descriptors for textures could provide powerful means in applications of texture browsing, coarse texture classification and texture retrieval. Three texture descriptors have been specified in MEPG-7 [1]. The Homogeneous Texture Descriptor (HTD) provides quantitative characterization of texture patterns and is useful for images with homogeneous textural properties. HTD can be used for similarity-based image-to-image matching. To provide descriptions for non-homogeneous textures, Edge Histogram Descriptor (EHD) is provided. The spatial distribution of edges is extracted and is useful for image-to-image matching when the underlying texture is not homogeneous. Texture Browsing Descriptor also named Texture Browsing Component (TBC) relates to the perceptual characterization of texture, in terms of regularity, directionality and coarseness. The coarseness is related to image scale or resolution. This descriptor is useful for browsing type applications and coarse classification of textures.

Approaches for estimating the dominant directions of textures in spatial domain [2] or in frequency domain [3] have been proposed. Liu and Picard [4] proposed a texture model addressing the perceptual characteristics of textures mentioned above. The texture model proposed emphasizes the perceptually most salient periodic information. To determine the prominence of periodic structures in a texture, the energy distribution of image autocovariance function is examined. The image autocovariance is computed as the inverse DFT of the image power spectrum. In addition, a computation method of TBC has been recommended in [5], [6]. In this method, an image is filtered using a bank of scale and orientation selective band-pass filters called Gabor wavelet [7]. To compute the dominant directions, directional histograms are constructed from the filtered images at different scales and then the two histogram peaks with the highest contrasts are considered as the two dominant directions. To compute the scale and regularity, the filtered images are projected along the two dominant directions to form two sets of projections. For each set of projections, the autocorrelation function is evaluated. The scale and regularity are then determined based on the peaks and valleys of the autocorrelation function. In addition, consistency checks based on the neighboring relationship of the projections are provided to make the method more robust. As it involves the computation of applying Gabor wavelet filters and autocorrelation function, the method is relatively time-consuming.

In this paper, an efficient computation method of TBC will be provided. The method is based on the fact that for a directional texture image, the magnitudes of its Fourier spectrum will concentrate on a certain direction; for regular, on several directions; for irregular, not on any direction [8], [9]. To compute the regularity of textures, Fourier transform is first performed. The Fourier spectrum is then smoothed to reduce noises. The smoothed Fourier spectrum is treated as an image and the Fourier transform is performed again to produce an enhanced Fourier spectrum. A regularity measure based on the variance of the radial wedge distribution is then calculated to determine the regularity of textures. For regular textures, the texture primitives are assumed to be parallelograms, the two dominant directions and their associated scales of the primitives are determined by Hough transform [10]. In addition, principal component analysis is provided to detect textures with only one dominant direction. Experiments of texture browsing, coarse classification of textures and texture retrieval are performed to test the texture images of Brodatz album [11] and Corel Gallery image database to demonstrate the efficiency and effectiveness of the proposed method.

The rest of the paper is organized as follows. The proposed computation method is presented in Section 2. In Section 3, experimental results and discussion are described. Finally, in Section 4, we give a conclusion.