Analysis of dimensionality reduced local directional pattern on different scales

This paper affords an analysis of a novel local descriptor-dimensionality reduced local directional Pattern (DR-LDP) on different scales. DR-LDP extracts the features of the face by partitioning the image into 3 χ 3 sub-regions and the sub-region was convoluted with a set of eight Kirsch masks. The single eight-bit code generated for each sub-region. The histogram features are extracted by partitioning the resultant DR-LDP encoded image into 8 × 8 regions. The features of each regions are combined to form a feature vector for the given facial image. For any query image, the same process is carried out to extract the feature vector. A chi-square test is used to measure the dissimilarity of the feature, the dissimilarity of feature vector in the database with the feature vector of query image was determined to recognize the face. The experiments had been accomplished on well-known benchmark databases. In this paper, an analysis of DR-LDP on 3 χ 3, 5 χ 5, 7 χ 7 regions and convultes each region with 3 × 3, 5 × 5, 7 × 7 eight Kirsch masks are performed to test the robustness of it. From the analysis, it is evident that the DR-LDP performs the best for the scale 3 χ 3.