New BarkTex benchmark image test suite for evaluating color texture classification schemes

 

 

Download the test suite

 

Context

Several image test suites are available in the literature to evaluate the performances of classification schemes. In the framework of color texture classification, two have often been used, the OuTex-TC-00013 and Contrib-TC-00006 test suites, that are extracted from the OuTex and VisTex databases, respectively. Accuracies reached by many classification schemes have been compared thanks to these color texture image sets. However, by analyzing the classification results obtained, we have noticed that color histograms obtain better rate of well-classified images than color texture features, although it does not take into account any texture information in the image. This incoherence lead us to rethink the relevance of these two benchmark color texture sets for measuring the performances of color texture classification algorithms. Indeed, the partitioning used to build these two sets consists in extracting training and validation subimages from a same original image. We show that such a partitioning, when it is combined with a classifier such as the nearest neighbor classifier, lead to biased classification results. That is why we propose a new relevant image test suite for evaluating color texture classification schemes, where the training and the validation sub-images come from different original images in order to ensure that color texture images are less correlated as possible.

 

Description

The BarkTex database1 includes six tree bark classes, with 68 images per class. To build the New BarkTex set, a region of interest, centered on the bark and whose size is 128 × 128 pixels, is first defined. Then, four sub-images whose size is 64 × 64 pixels are extracted from each region. We thus obtain a set of 68 × 4 = 272 sub-images per class. To ensure that color texture images used for the training and the testing stages are less correlated as possible, the four sub-images extracted from a same original image all belong either to the training subset or to the testing one: 816 images are thus used for the training and the remaining 816 for the testing stage.

 

 

How to cite this dataset

Please cite this dataset using the following reference:

[1] A. Porebski, N. Vandenbroucke, L. Macaire and D. Hamad, “A new benchmark image test suite for evaluating color texture classification schemes,” Multimedia Tools and Applications Journal, vol. 70, no. 1, pp. 543–556, 2014.

 

Results

[7] - 92.6%

[5] - 89.6%

[6] - 84.31%

[4] - 82.1%

[2] - 81.37%

[3] - 81.37%

[8] - 76.6%

[1] - 75.9%

 

Papers using New BarkTex dataset

[2] A. Porebski, N. Vandenbroucke and D. Hamad, “LBP histogram selection for supervised color texture classification”, 20th IEEE International Conference on Image Processing (IEEE-ICIP'13), Melbourne (Australia), pp. 3239-3243, IEEE Signal Processing Society, September 2013.

 

[3] M. Kalakech, A. Porebski, N. Vandenbroucke and D. Hamad, “A new LBP histogram selection score for color texture classification”, 5th International Conference on Image Processing Theory, Tools and Applications (IEEE-IPTA'15), pp. 242-247, Orléans (France), November 2015.

 

[4] F. Sandid and A. Douik, “Robust color texture descriptor for material recognition”, Pattern Recognition Letters, vol. 80, pp. 15-23, 2016.

 

[5] F. Sandid and A. Douik, “Dominant and minor sum and difference histograms for texture description,” in International Image Processing, Applications and Systems Conference (IPAS), pp. 1–5, 2016.

 

[6] J. Wang, Y. Fan and N. Li, “Combining fine texture and coarse color features for color texture classification,” Journal of Electronic Imaging, Volume 26, Issue 6, 2017.

 

[7] A. Porebski, V. Truong Hoang, N. Vandenbroucke and D. Hamad, “Multi-color space local binary pattern-based feature selection for texture classification”, Journal of Electronic Imaging, Volume 27, Issue 1, 2018.

 

[8] L. Kabbai, M. Abdellaoui and A. Douik, “Image classification by combining local and global features”, The Visual Computer, pp. 1–15, 2018.

 

 

If you have a work that uses this dataset, please send a message informing the reference of your paper and also the obtained results (porebski [at] lisic.univ-littoral.fr).

 

 

1 R. Lakmann, “Barktex benchmark database of color textured images” Koblenz-Landau University, ftp:// ftphost.uni-koblenz.de/ outgoing/vision/Lakmann/BarkTex.