Heterogeneous face recognition over cross-distance & cross-spectrum

Abstract

This report concluded the final year project, which spread over the whole academic year, as part of the academic requirements of the School of Electrical and Electronics Engineering. It encompassed the project introduction, experiments performed, results obtained, as well as future development of the project.

One of the most difficult challenges in the face recognition field is to match faces captured in different light conditions and distances. The objective of this final year project is to design an automatic heterogeneous face recognition system which achieves effective face matching over cross-distance and cross-spectrum environments. First of all, preprocessing procedures were carried out to filter and screen the raw images. It has been proven that these processes have improved the overall performance. The system then extracted the image features from the processed image database and manipulated the image recognition result for each modality. Image feature extraction algorithms, such as Scale Invariant Feature Transform (SIFT), Local Binary Pattern (LBP) and Local Ternary Pattern (LTP), were computed and compared. The comparison showed that the SIFT algorithm outperformed the other approaches and delivered the best recognition rate.