Alice Porebski

Haralick feature extraction from LBP images for color texture classification

In this paper, we present a new approach for color texture classification by use of Haralick features extracted from co-occurrence matrices computed from local binary pattern (LBP) images. These LBP images, which are different from the color LBP initially proposed by Maenpaa and Pietikainen, are extracted from color texture images, which are coded in 28 different color spaces. An iterative procedure then selects among the extracted features, those which discriminate the textures, in order to build a low dimensional feature space. Experimental results, achieved with the BarkTex database, show the interest of this method with which a satisfying rate of well-classified images (85.6%) is obtained, with a 10-dimensional feature space.

Iterative Feature Selection for Color Texture Classification

In this paper, we describe a new approach for color texture classification by use of Haralick features extracted from color co-occurrence matrices. As the color of each pixel can be represented in different color spaces, we automatically determine in which color spaces, these features are most discriminating for the textures. The originality of this approach is to select the most discriminating color texture features in order to build a feature space with a low dimension. Our method, based on a supervised learning scheme, uses an iterative selection procedure. It has been applied and tested on the BarkTex benchmark database.

Supervised texture classification: color space or texture feature selection?

The color of pixels can be represented in different color spaces which take into account different properties. However, no color space is well-suited to the discrimination of all texture databases and the prior determination of such a space is not easy. In this paper, we compare the performances reached by two texture classification schemes that use color spaces: (a) the single color space selection approach, that defines a set of texture features and then selects the color space with which the texture features allow to reach the highest classification accuracy, (b) the multi-color space feature selection (MCSFS) approach, that selects texture features which have been processed from images coded into different color spaces. Experiments carried out with benchmark texture databases show that taking advantage simultaneously of the properties of several color spaces thanks to the MCSFS approach improves the rates of well-classified images with lower learning and decision processing times.

LBP histogram selection for supervised color texture classification

In this paper, we propose a Local Binary Pattern (LBP) histogram selection approach. It consists in assigning to each histogram a score which measures its efficiency to characterize the similarity of the textures within the different classes. The histograms are then ranked according to the proposed score and the most discriminant ones are selected. Experiments, which have been carried out on benchmark color texture image databases, show that the proposed histogram selection approach is able to improve the classification performances.

Comparison of feature selection schemes for color texture classification

In this paper, we propose to compare the performances of two sequential feature selection schemes used for supervised color texture classification. We focus this study on the sequential forward selection (SFS) scheme and the more complex sequential forward floating selection (SFFS) scheme which avoids the “nesting effect”. These schemes retain Haralick features extracted from chromatic co-occurrence matrices of images coded in different color spaces. We experimentally study the contribution of these two feature selection schemes with three benchmark color texture databases.

A new benchmark image test suite for evaluating colour texture classification schemes

Several image test suites are available in the literature to evaluate the performance of classification schemes. In the framework of colour texture classification, OuTex-TC-00013 (OuTex) and Contrib-TC-00006 (VisTex) are often used. These colour texture image sets have allowed the accuracies reached by many classification schemes to be compared. However, by analysing the classification results obtained with these two sets of colour texture images, we have noticed that the use of colour histogram yields a higher rate of well-classified images compared to colour texture features. It does not take into account any texture information in the image, this incoherence leads us to question the relevance of these two benchmark colour texture sets for measuring the performances of colour texture classification algorithms. Indeed, the partitioning used to build these two sets consists of extracting training and validating sub-images of an original image. We show that such partitioning leads to biased classification results when it is combined with a classifier such as the nearest neighbour. In this paper a new relevant image test suite is proposed for evaluating colour texture classification schemes. The training and the validating sub-images come from different original images in order to ensure that the correlation of the colour texture images is minimized.

Selection of color texture features from Reduced Size Chromatic Co-occurrence Matrices

In this paper, we present a feature selection scheme which builds a low-dimensional feature space for texture classification. These features are extracted from texture descriptors called Reduced-Size Chromatic Co-occurrence Matrices (RSCCMs) which result from color quantization. Thanks to experimental results achieved with VisTex and OuTex databases, we show that the analysis of Haralick features extracted from these RSCCMs, themselves computed from color images coded in 28 different color spaces, provides satisfying classification results while significantly reducing the processing time.

A new LBP histogram selection score for color texture classification

This paper presents and compares a new adapted version of the Laplacian score used to select LBP histogram for color texture classification. During a supervised learning stage, we first compute a similarity matrix between images using the true class labels of these images. Then, a score is attributed to each histogram. This score allows to measure the capability of the histogram of preserving the similarity matrix. The histograms are then ranked according to the proposed score and the most discriminant ones are selected. Experiments are achieved on benchmark color texture image databases in order to show the interest of the proposed score for histogram selection.

Texture analysis of lace images using histogram and local binary patterns under rotation variation

The images of lace textile are particularly difficult to be analyzed in digital form using classical image processing techniques. The major reasons of this difficulty emerge from the complex nature of lace which generally has different textures in its constituents like the background and patterns. In this paper, we study the behavior of Image Histogram (HistI) and Local Binary Patterns (LBP) on image extracts of lace in presence and absence of rotation. We further evaluate two variants of LBP; primarily the LBP Histogram (LBPB) and secondly the Fourier Transform applied on the LBP Histogram (LBPFFT). Consequently, we analyze the contribution of data fusion on feature level and score level in the different experimentations. The classification rate evaluates the discrimination degree of each descriptor via the k nearest neighbors kNN classifier. Experimental results indicate that the LBPB, LBPFFT and HistI combined at score level generate the better performance in absence of transformations. Whereas, LBPFFT and HistI combined at the same level generate the better classification rate, in the presence of rotation.

Multi-color space local binary pattern-based feature selection for texture classification

Abstract. This paper deals with multi-color space texture classification. Two approaches are proposed and compared: a multi-color space histogram selection (MCSHS) and a multi-color space bin selection. These approaches select local binary pattern (LBP) histograms or LBP bins that have been processed from images coded in multiple color spaces. On the one hand, the proposed LBP-based feature selection scheme overcomes the difficulty of choosing a relevant color space, and on the other hand, it takes advantage of the specific properties of several color spaces by combining them. Experiments show that the MCSHS approach is relevant for color texture classification issues that require good performances whether in accuracy or classification computation time.