Palmprint Recognition Based On Improved PCA And Multi-feature Integration

Palmprint recognition is an important part of biological feature recognition. Palmprint identification is a method of identification that use palmprint features. PCA is a classic palmprint feature extraction method. But it also has some disadvantages. It will lose some information and rotation, light changes will affect its recognition rate. For these problems we propose two improved PCA algorithm, they are focusing on global features and local features of palmprint. In a study of palmprint recognition system we found that global features and local features have different roles in palmprint recognition process. Therefore, a comprehensive study of these two feature recognition algorithm is very meaningful. In response to these problems, the main work is as follows:(1) For the weakness of PCA algorithm: only main message original image is preserved and it will lose some information. Combined with the strength of the original palmprint image contrast is not obvious, this paper proposes palmprint recognition method based on Fourier and PCA feature combination. Palmprint first passes through a Fourier transform, and then use the PCA to reduce the dimension of the image after conversion, and finally we use the LS-SVM classifier for recognition. Experiments demonstrate the effectiveness of the proposed method.(2) For the weakness of PCA algorithm: rotation, light changes affect its recognition rate. We propose palmprint recognition method based on Module PCA and LS-SVM combination. Palmprint image is divided into 16( 4 ×4) sub-block images, and then use the PCA to reduce the dimension of the images. The advantage of this method is obvious in the single sample palmprint recognition problems, because it extends the sample size effectively. Experiments demonstrate the effectiveness of this method.(3) The main palmprint image information more focused after Fourier transform. ModulePCA will preserve the local information better and enhance the robustness of image. Palmprint image contains not only the global feature also contains a wealth of local features. The paper design a palmprint identification system based on the global and local features after integrated the above two methods. Palmprint image through a Fourier transform to obtain global feature. Go through Gabor transform, and then use the ModulePCA to obtain local features. In the recognition phase, we use a coarse-to-fine serial integration strategy to integrate global and local classifier. Experiments verify that it can take care of speed and accuracy of palmprint recognition and enhance the usefulness of palmprint recognition system.In this paper, the experimental data set that we used is from the Hong Kong Polytechnic University(PolyU).