Palmprint recognition with improved two-dimensional locality preserving projections

Recently, two-dimensional locality preserving projections (2DLPP) was proposed to extract features directly from image matrices based on locality preserving criterion. Though 2DLPP has been applied in many domains including face and palmprint recognition, it still has several disadvantages: the nearest-neighbor graph fails to model the intrinsic manifold structure inside the image; large dimensionality training space affects the calculation efficiency; and too many coefficients are needed for image representation. These problems inspire us to propose an improved 2DLPP (I2DLPP) for recognition in this paper. The modifications of the proposed I2DLPP mainly focus on two aspects: firstly, the nearest-neighbor graph is constructed in which each node corresponds to a column inside the matrix, instead of the whole image, to better model the intrinsic manifold structure; secondly, 2DPCA is implemented in the row direction prior to 2DLPP in the column direction, to reduce the calculation complexity and the final feature dimensions. By using the proposed I2DLPP, we achieve a better recognition performance in both accuracy and speed. Furthermore, owing to the robustness of Gabor filter against variations, the improved 2DLPP based on the Gabor features (I2DLPPG) can further enhance the recognition rate. Experimental results on the two palmprint databases of our lab demonstrate the effectiveness of the proposed method.