Clifford Algebra and Gabor filter for color image texture characterization

Abstract

The first texture descriptors are proposed in 1973 by Haralick[1] and in 1975 by Marie Galloway[2] are still used today for image classification or segmentation in various domains. The majority of these features are defined for gray level images. Many papers have proposed different approaches among them, a parallel study of color and gray level texture characterization when other combined the two groups of features by defining joint features[3][4][5]. Combining features or defining them jointly are outperformed [6] even though these features impose some constraints such as the color system, measurement similarity methods. The aim of this paper is to use Clifford Algebra to represent multi-component images and to propose a color image texture.

Statistical Learning for Texture Characterization

“…We shall discuss and compare with 3 papers [5], [9] and [20] that have proposed modeling color-texture characterization (rest of the papers are on grayscale characterization). Unlike the methods mentioned in the above three papers, our paper proposes texture characterization at two levels, global and local.…”

“…Global abstraction is done at the whole image level and local abstraction at the image block levels. The characterization in [5], [9] and [20] considers only color features and not grayscale. Our method has the advantage of capturing both grayscale and color textural variance present in an image (Refer Section 3.3).…”

“…Our experimental setup is more general and flexible in the sense that it consists of a collection of images taken from different benchmark datasets (Berkley for panoramic images, Vistex for categorical natural images, Corel for synthetic and natural textures). On the other hand [5], [9] experiment on purely texture images and do not consider any scenic images. The paper [20] considers scenic images where an image segment is described by a color Gaussian Markov Random Field model.…”

“…al. in [5] proposed Clifford Algebra and Gabor filter for color image texture characterization to represent multi-component images and [9] presents a scheme for colored texture image classification based on wavelet correlation signatures containing the energies of each color plane and cross correlation between different planes. Most of the texture characterization is done either keeping a specific domain (synthetic or natural images specific to a category) or is application specific (segmentation on specific benchmark dataset or image retrieval).…”

A Unified Approach for Representing Outdoor Scenes

“…This can also be extended further to smaller blocks if required. Figure 9 shows a Vistex grass image (referred in two papers by [23] and [24])The proposed block illustrations are as follows :…”