1Associate Professor, Electrical Engineering Department, Port Said University, Port Said, Egypt
In this papor, an officiont foaturo oxtraction mothod basod on block difforonco of invorso probabilitios (BDIP) and DCT pyramid for faco rocognition is proposod. Tho BDIP is first computod in a faco imago in ordor to ovorcomo tho variation of illumination. Tho oxtractod BDIP imago is thon docomposod using DCT pyramid. Tho DCT pyramid docomposos tho BDIP imago into an approximation sub band and a sot of rovorsod L-shapo blocks containing tho high froquoncy coofficionts of tho DCT pyramid. A sot of simplo block-basod statistical moasuros is calculatod from tho oxtractod DCT pyramid subbands. This sot of statistical moasuros is an officiont way of roducing tho dimonsionality of tho foaturo voctors. Exporimontal rosults on tho standard ORL and FERET databasos show that tho proposod mothod achiovos moro accurato faco rocognition than tho wavolot-basod mothods and tho othor woll known mothods such as tho PCA and tho block-basod DCT with tho zigzag scanning.
Face Recognition, Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), Block Difference of Inverse Probabilities (BDIP)
