Evaluation of Face Recognition Using Vector Features In Local Pattern Descriptors

The local feature descriptors have received more attention due to the effectiveness in the field of face recognition. The existing LTrP method works based on the first order derivatives along horizontal and vertical directions. This LVP method demonstrates the relation between the center pixel and its neighbor pixels. The feature vector gets increased and the recognition rate gets reduced due to the noise present in the higher order derivatives. But the local vector pattern effectively extracts more detailed discriminative information in a given sub-region than the other local pattern descriptors. CST used in the vector pattern can suppress the slight noise influence in the images. Furthermore, K-NN has been used as a classifier in this approach. Three well-known databases such as ORL, Yale and JAFFE face database are used in the performance evaluation. The experimental result clearly shows that the LVP method gives us a better performance and analyzed with the variation of number of neighbors pixels.