Embedded Approach for Image Texture Classification

In this paper, an image texture classification approach is presented which is based on ARM 9 Single Board Computer (SBC). Texture classification is the process of deciding the texture category of an observed image. The proposed approach involves feature extraction and classification stages. Texture features are extracted using two methods viz. Dominant Local Binary Patterns (DLBP) and Circularly Symmetric Gabor Filters. Dominant Local Binary Patterns method captures features which occur most frequently in the image whereas Circularly Symmetric Gabor Filters capture global features to give more textural information. Both these features are combined and given to classification stage. Classification is performed using Support Vector Machine (SVM) classifier. The proposed approach is implemented on ARM SBC9302. Its performance is evaluated by comparing it with six different feature extraction methods like Daubechies wavelet transform features (DBWP), Rotation invariant Daubechies wavelet transform features(RDBWP), Traditional Gabor filters (TGF), Circular Gabor filters(CGF), Anisotropic circular Gaussian MRFs (ACGMRF), andUniform local binary patterns (LBP). Through experiments, the performance is also evaluated by applying the approach to original textures, histogram-equalized textures, randomly-rotated textures, histogram-equalized and randomly-rotated textures, and texture corrupted by additive Gaussian noise. It is experimentally demonstrated that this method achieves better classification accuracy,is rotation invariant, less sensitive to histogram equalization and robust to noise.