Abstract:
Image classification is an important research problem that has vast applications in the area of machine vision, artificial intelligence and pattern recognition. Feature extraction, feature selection, and classifier design are three components of image classification system. In recent years, wavelet transform based approaches have been used extensively in image classification. The key issue in wavelet based classification methods is selection of the best sparse representation out of the features set generated by over complete dictionary. In this paper, the authors proposed a framework of sparse representations to address the problem of feature selection using over complete dictionary for image classification. In sparse representation, an important factor is discrimination power as it improves the accuracy of classifying corrupted images. The authors proposed a new approach in the framework of the sparse representations by combining the discrimination power and sparseness in a single cost function. Orthogonal Matching Pursuit algorithm is employed to extract the sparse features of images. LibSVM (Support Vector Machine) classifier is used for the classification. Efficiency and robustness of the developed algorithm is demonstrated with environments of different SNRs (Signal-Noise Ratio) and occlusion levels.
Page(s):
467-476
DOI:
DOI not available
Published:
Journal: Mehran University Research Journal of Engineering and Technology, Volume: 28, Issue: 4, Year: 2009