Applications of Multiscale Geometric Analysis in Image Texture Recognition and Classification

Abstract

With the rapid development of computer vision, the applications of image texture recognition and classification are increasingly prevalent across various domains, particularly in medical imaging, industrial inspection, and remote sensing image analysis, these applications hold significant practical importance. Traditional texture recognition techniques often rely on manually designed feature extraction methods, which tend to perform poorly in complex environments, are sensitive to noise and lighting variations, and are limited when dealing with non-uniform or multiscale textures. To address these shortcomings, this paper introduces two novel texture analysis methods that enhance the robustness of texture features and improve classification accuracy. The first part of the study presents the contourlet-kernel spectral regression (KSR) image texture feature extraction technique, which, by integrating Contourlet transform with Krawtchouk polynomials, effectively enhances the descriptive power and adaptability of features. The second part explores a texture image classification method based on domain-multiresolution cooccurrence matrices (MCM), which significantly improves the accuracy and robustness of the classification process by analyzing the co-occurrence characteristics of images at multiple resolutions. The introduction of these methods not only optimizes texture recognition performance but also advances the application of image processing technologies in complex scenarios.