A New Hybrid Shape Moment Invariant Techniques for Face Identification in Thermal and Visible Visions

Presently, the extraction of robust facial features is becoming very effective for accurate face recognition especially for smart security surveillance systems. This paper investigates three different moment invariants techniques for robust facial features extraction and then determine how each one of these moments is affected by whether the face image was thermal or on a greyscale with the proposal of a hybrid technique that deals with the robust descriptors of each method. This hybrid technique has improved the results and gave robust facial features for face identification. A feed-forward neural network is trained with these moments' features where the recognized faces are classified to one of the basic faces of IRIS and CARL face datasets which achieved high accuracy reaching 98.1% for thermal images and 81.2% for grey images.