Image Classification Schemes for Statistical Moments of Wavelet and Gradient Matrix

Abstract

There are many different pattern recognition applications that have a feature-like look, and one of such applications is texture categorization. In order to complete a greater grade of accuracy in the classification of textures, this study employs two distinct methods. In the first step of the process, directional texture properties are extracted with the use of a gradient matrix. There are two different kinds of features that have been suggested: discrete wavelet transforms and statistical distributions of various relative moments. Discrete wavelet transforms, as well as statistical distributions of various relative moments, and (iv) discrete wavelet transforms are all included in the first order gradient feature vector as well as the max-min gradient feature vector respectively. Every single one of these feature vectors is investigated on its own. In order to classify the information, we utilized all four distinct kinds of Euclidean distance metrics. The proposed method was evaluated by putting it through its paces by employing the 475 texture classes and 16 photographs provided by the Salzburg Texture Image Database. Seven different color bands were utilized in order to differentiate each image (red, green, blue, gray, and so on …). Calculating the inter/intra scatter evaluation for every feature using the theories of average as well as standard variation allowed researchers of identify the attributes that were most suited for discrimination. GM completed with a score of 98.3 on the testing set and a score of 96.2 on the training set, whereas DWT finished with a score of 99.98 on the testing set and a score of 99.71 on the training set.