Research Of Palmprint Recognition Based On 2-dimensional PCA

The research of Palmprint recognition using computer started in the late 1960s. However, it developed slowly because of the limits of its difficulties and technical conditions. In the recent 20 years, due to the rapid technology development of computer science, signal processing and pattern recognition, Palmprint recognition technology has received more attention and has got an increasingly significant progress both in theory and practice.As the one of the most important biometric recognition, the development of Palmprint recognition has meaningful value of practice. The traditional PCA presented by Liu et al, has been widely used in expressing eigenvector of image. In 2004, Yang et al improved Liu's method and called it 2DPCA later. This dissertation presents a Palmprint recognition method based on modular 2DPCA, which is on the basis of original 2DPCA and combining the partitioned matrix of linear algebra. Finally,we showed the experiment result. The main research works of this paper are as follows.(1) The dissertation reviewed the research background and history of Palmprint recognition, and introduced the traditional algorithm of feature extraction and recognition, especially PCA (Principal Component Analysis) and Fisher linear analysis, and presented the formula proving.(2) After Capturing the Palmprint image, a series filtering which include low-pass filter, image equalization and Binary image in order that the image sample become more clearly and contain less noise, was adopted. Then we relocated and separated the image by using maximum inscribe circle algorithm, for the purpose of decreasing the noise brought by translation or scale variation, and so on, and eventually gain the ROI(Region Of Interest). Then we utilized the Gabor filter to enhance the principal texture line and Wavelet transforms to get different frequency component of the image respectively, finally we can eliminate the irrelevant frequency and gain the low frequency.(3) 2DPCA algorithm is described in detail. Differently to PCA, it will not transform an image matrix into a 1D vector, instead, it directly computes the eigenvectors using the original image covariance matrix without matrix-to-vector conversion. Therefore, the 2DPCA has two advantages accordingly. First, it is easy to estimate the covariance matrix accurately. Second, the features extracted by 2DPCA may be influenced by the number of samples slightly. The paper proposed a palmprint method using modular 2DPCA, aim to decline the complexity through partitioned matrix and enhance the local feature. At last, we can acquire the training samples as well as extracted the feature of testing samples by utilizing the mentioned modular 2DPCA, and matched each other by nearest distance rule. The experiment shows that modular 2DPCA can improve recognition rate more apparently compare to 2DPCA.