Multi-resolution Texture Classification Based on Local Image Orientation

Abstract

Abstract. The aim of this paper is to evaluate quantitatively the discriminative power of the image orientation in the texture classification process. In this regard, we have evaluated the performance of two texture classification schemes where the image orientation is extracted using the partial derivatives of the Gaussian function. Since the texture descriptors are dependent on the observation scale, in this study the main emphasis is placed on the implementation of multi-resolution texture analysis schemes. The experimental results were obtained when the analysed texture descriptors were applied to standard texture databases.

Local binary patterns versus signal processing texture analysis: a study from a performance evaluation perspective

“…In this regard, we have conducted the experiments using four Outex databases defined by 24 texture classes and three Brodatz databases defined by 36 texture classes and the classification results indicate that the performances of the LBP technique and multi-channel GF are comparable. Our experimental results strengthen the conclusion that orientation is an important characteristic of the texture and best results are achieved when this texture property is accurately sampled by the texture descriptors (this conclusion is also confirmed by the results reported in our recent papers where the discriminative power of two texture analysis techniques based on the evaluation of the local image orientation at different observation scales Ghita et al, 2008) was quantitatively measured when the analysed techniques were applied for texture classification tasks). Another important finding resulting from this investigation is the fact that although the concepts behind the LBP and multi-channel filtering techniques are different, we have demonstrated that their behaviour in modelling oriented and isotropic textures shares many similarities.…”

Unsupervised Image Segmentation Based on the Multi-Resolution Integration of Adaptive Local Texture Descriptors

“…In Ghita et al, 2008) a texture descriptor based on the evaluation of the dominant image orientation calculated at micro and macrolevel was proposed. In this section, experimental results that quantify the performance of the image orientation based texture descriptor in the segmentation process are provided.…”

“…Typically, the central frequencies are selected to be one octave apart and for each central frequency is constructed a set of filters corresponding to four (0 0 , 45 0 , 90 0 , 135 0 ) or six orientations (0 0 , 30 0 , 60 0 , 90 0 , 120 0 , 150 0 ). Texture extraction using the dominant image orientation at micro and macro-levels is an approach defined in terms of the distribution of the dominant edge orientations at micro and macro-level and was introduced in Ghita et al, 2008). In this regard, the orientation for each pixel in the image is extracted using the partial derivatives of the Gaussian function (G) while the main focus is centred on the evaluation of the local dominant orientation.…”

“…Recently a novel hybrid statistical-structural approach was proposed where the texture is described in terms of the distribution of edge orientations calculated at micro and macro-level for all pixels in the image Ghita et al, 2008). The quantitative evaluations were conducted on standard texture databases and the results indicated that the local image orientation based descriptor has a high discriminative power in the context of texture classification.…”