Two-level dimensionality reduced local directional pattern for face recognition

摘要

Face recognition can be done efficiently using local approaches. Local directional pattern (LDP) is one such approach that serves as a descriptor for face recognition. It assigns a code for each pixel and the image is encoded. Histogram binning is done on the LDP encoded image to represent the face. A two-level dimensionality reduced local directional pattern (TL-DR-LDP) is proposed in this paper. The proposed TL-DR-LDP is robust in recognising the faces with maximum recognition rate. The proposed descriptor codes the image by dividing the image into regions and for each region, a code is defined. The same process is repeated for one more level and hence named as TL-DR-LDP. At each level, the dimensions of the feature vector are drastically reduced and performance of the descriptor maintains the higher recognition rate. The proposed descriptor is tested on standard benchmark databases like FERET, Extended YALE B and ORL. The results obtained prove that the TL-DR-LDP is exemplary.