Abstract
This paper proposes a novel local texture description method which defines six human visual perceptual characteristics and selects the minimal subset of relevant as well as nonredundant features based on principal component analysis (PCA). We assign six texture characteristics, which were originally defined by Tamura et al., with novel definition and local metrics so that these measurements reflect the human perception of each characteristic more precisely. Then, we propose a PCA-based feature selection method exploiting the structure of the principal components of the feature set to find a subset of the original feature vector, where the features reflect the most representative characteristics for the textures in the given image dataset. Experiments on different publicly available large datasets demonstrate that the proposed method provides superior performance of classification over most of the state-of-the-art feature description methods with respect to accuracy and efficiency.
Highlights
Texture feature description plays fundamental roles in many computer vision applications, especially general texture classification that can be widely used in material surface inspection [1], medical imaging [2,3,4], object recognition [5,6,7], scene recognition [8], and image retrieval [9, 10]
To evaluate the performance of the proposed PCAorthogonality key local Tamura’s texture description method (LT+principal component analysis (PCA)-ORTH), different texture description methods combined with different feature selection methods are compared to classify textures in the given testing texture databases
There are three image databases used in our experiments to evaluate different texture descriptors: (i) 32 texture images from the Brodatz texture database [55]; Figures 3(a), 3(b), and 3(c) show three example textures in the database; (1) Initialize L = 1, the number of selected features, sselect = {1, 2, . . . , p}, the selected features set, ortho = 0, the orthogonality of the principal components computed from the selected features, k is the principal component coefficients matrix with p rows corresponding to the original features
Summary
Texture feature description plays fundamental roles in many computer vision applications, especially general texture classification that can be widely used in material surface inspection [1], medical imaging [2,3,4], object recognition [5,6,7], scene recognition [8], and image retrieval [9, 10]. The local binary pattern (LBP) [19] and its variants [20,21,22,23,24] exploited the direct comparison between the pixel intensity and its neighbours’ intensities within a local area These methods, the so-called low-level texture descriptors, described the texture by the intuitive features of the texture itself, including the occurrence and direction of the edges or corners, the shape of the textures, the occurrence frequency of the texture primitives, the variance of the pixel intensities, and other primitive level features. Since low-level descriptors typically lacked in giving significant attention to explore the relationship between the texture characteristics they described and human visual sense, image descriptors for high-level features that described how human beings observed the different texture regions have attracted more attention in recent years. The high-level texture descriptors still need to solve two problems:
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
