PROPOSAL OF A UNIFIED ROBUST DESCRIPTOR FOR PATTERN IDENTIFICATION

Abstract

In the field of computer vision, pattern identification plays a major role in the development of the system. To identify the patterns, there is a need of effective feature descriptor that describes and represents the characteristics of the object. The aim of this thesis is to develop a unified global descriptor using texture and shape information of the pattern. The texture information is extracted using Gabor Wavelet Transform (GWT) by developing a Gabor filter bank employing different orientation and scale. The shape information is obtained by applying Radon Transform on the image and the features are obtained by applying Discrete Wavelet Transform (DWT) on each of the Radon’s projections. Since the DWT is being employed, the concept of discrete Radon transform is introduced in which the number of projections is dependent on the size of the image. By assimilation of Gabor feature with Radon feature, a unified descriptor is constructed and the performance of descriptor is evaluated by conducting an experiment using standard data sets. The performance is measured in terms of Average Recognition Rate (ARR) and further for its effectiveness the comparison of ARR is carried out with the similar state-of-the-art. The comparison of ARR shows the effectiveness and superiority of the proposed descriptor. The classification accuracy is evaluated through two widely used classification models, i.e. k-Nearest Neighbour (KNN) and Support Vector Machine (SVM).