Local Gabor Dominant Direction Pattern for Face Recognition

We propose a novel face image representation – Local gabor dominant direction pattern (LGDDP) for face recognition. The face image is convolved with the Gabor filters, resulting in multiple response images of different orientations and scales. The response images’ each pixel is encoded by the LGDDP descriptor from the pixel’s dominant neighboring one or two pixels. The image formed by the LGDDP descriptor is partitioned into multiple regions and the histogram is extracted from each region. All the histograms are concatenated into the spatial histogram. The nearest neighbor classifier and the weighted intersection histogram similarity measure are used for face image classification. The advantage of the proposed LGDDP method lies in the high recognition rate performance and the low computational complexity. Extensive experiments are performed on FERET face image database and the experimental results verify the LGDDP’s effectiveness by comparing LGDDP with other well-known published face recognition methods.