Nibble-Based Face Recognition Using Convolution of Hybrid Features
Pattarakamon Rangsee, K. Bommanna Raja, K R Venugopal · 2019
Face Recognition technique is used to identify human beings effectively without their intervention. In this paper, we propose a novel nibble-based face recognition using convolution of hybrid features. Face images are converted from color to grayscale anda decimal values of each pixel is represented using 8-bit binary. The novel technique of converting 8-bit binary pixel into Left-Side Nibble (LSN) and Right-Side Nibble (RSN) is introduced to increase computational speed. The 4-bit LSN is converted to a decimal value varying between 0 and 240. DWT is applied on LSN matrix and only LL band coefficients are considered as first set of transform domain features. The 4-bit RSN is considered and converted into decimal values ranging between 0 and 15. HOG is applied on RSN matrix to generate second set of spatial domain features. The final hybrid features are generated by performing linear convolution on spatial- and transform-domain features. The Artificial Neural Network (ANN) is used to compare and classify the database image features and test image features to compute performance parameters of proposed algorithm. It is observed that the performance of the proposed method is better than the existing methods.