Processing and classification of palmprint images

Palmprint segmentation, which involves extracting a suitable part of a palmprint for feature extraction and matching, provides the basis for palmprint identification. Segmentation aims at automatically and precisely extracting the central part of a palmprint for feature extraction. In this paper, a square-based segmentation method is introduced. The main idea is to first define key points on a palmprint. An orthogonal coordinate system is then established by using the three points. During segmentation, a square with fixed size is extracted from a predefined position under the coordinate system. The basic rule in determining the size and position of the square is to ensure that the part of the extracted image is available in all palmprints. To perform palmprint identification, the PCA is applied to the segmented palms for dimensionality reduction. Wavelet transform features were fused with PCA to improve identification results. The hamming distance was used for classification purpose. Results have shown that fusing PCA and wavelet features had led to an identification ratio of 99.4% which is better than previously investigated techniques.