Theoretical and experimental comparison of different approaches for color texture classification

Abstract

Color texture classification has been an area of intensive research activity. From the very onset, approaches to combining color and texture have been the subject of much discussion, and in particular, whether they should be considered joint or separately. We present a comprehensive comparison of the most prominent approaches both from a theoretical and experimental standpoint. The main contributions of our work are: (i) the establishment of a generic and extensible framework to classify methods for color texture classification on a mathematical basis, and (ii) a theoretical and experimental comparison of the most salient existing methods. Starting from an extensive set of experiments based on the Outex dataset, we highlight those texture descriptors that provide good accuracy along with low dimensionality. The results suggest that separate color and texture processing is the best practice when one seeks for optimal compromise between accuracy and limited number of features. We believe that our work may serve as a guide for those who need to choose the appropriate method for a specific application, as well as a basis for the development of new methods.

Texture Description Through Histograms of Equivalent Patterns

“…As a first step we established a predefined set of values for the parameters: Δ ∈ [1,32], Δ 1 ∈ [1, 3] and Δ 2 ∈ [4,6]. Then we estimated the accuracy of each method with each dataset for each value (or couple of values) of the parameters with the procedure described in the preceding section.…”

“…As it has been set into evidence in a recent work [6], this classification strategy is particularly suitable for feature comparison purposes due to the absence of tuning parameters, easiness of implementation and other desirable asymptotic properties. A review of recent related literature indeed shows that 1-NN is most commonly adopted in evaluating the relative performance of texture analysis algorithms [16,31,54,70,115].…”

Supervised texture segmentation using 2D LSTM networks

“…In order to validate the proposed model, the different approaches commonly used for segmentation are compared with our model; (1) patch-wise 6 Haralick features (gray) and Naive Bayesian classifier (Gray-Haralick+Naive Bayes) [26], (2) patch-wise 13 Haralick combined with color chroma features (color) and Naive Bayesian classifier (Color(HSV)-Haralick+Naive Bayes) [26,27], and (3) the combination of Gaussian Mixture Model, ExpectationMaximization, and Hidden Markov Random Field (GMM-HMRF) [28]. All the experiments of LSTM networks have been run by using the RNNLIB library [29] The first comparison method, 6 Haralick feature extraction (contrast, energy, homogeneity, correlation, dissimilarity, and angular second moment in four directions, 0 • , 45 • , 90 • , and 135 • ) was performed on 9 × 9 patches and each pixel is classified by a Naive Bayesian classifier.…”

Content-based image retrieval techniques for the analysis of dermatological lesions using particle swarm optimization technique

“…(10) is divided by n c h where c h is the standard deviation of c h . The matrices are averaged for rotation invariance, and the following five features are computed as in [22,23]. Contrast:…”