An Improved 2DLPP Method on Gabor Features for Palmprint Recognition

We propose an improved 2DLPP method on Gabor features (I2DLPPG) for palmprint recognition in this paper. 2DPCA is first utilized for dimensionality reduction of Gabor feature space maintaining most prominent 2D information. Thus similarity matrix corresponding to elements is easily constructed and the followed 2DLPP can be implemented directly in the reduced feature space. The proposed method preserving more intrinsic manifold structure of feature matrices yields higher recognition accuracy than the existing 2DLPP which treats the Gabor feature matrices as a whole. Meanwhile, fewer coefficients are extracted for image representation and recognition owing to 2DLPP and 2DPCA in the row and column directions simultaneously. Euclidean distance and the nearest classifier are finally used for classification. The recognition accuracy of the proposed I2DLPPG can reach 99.5% with 15 x 5 features. Experiments results demonstrate the effectiveness of our proposed method in both recognition accuracy and speed.