Open AccessHybrid Feature Approach of Face Recognition based on Pixel Binary Segmentation
Author(s) -
Rangsee Pattarakamon Rangsee*,
AUTHOR_ID,
K B Raja,
Venugopal K R,
AUTHOR_ID,
AUTHOR_ID
Publication year - 2022
Publication title -
international journal of engineering and advanced technology
Language(s) - English
Resource type - Journals
ISSN - 2249-8958
DOI - 10.35940/ijeat.c3361.0211322
Subject(s) - grayscale , artificial intelligence , pixel , pattern recognition (psychology) , computer science , binary number , histogram , segmentation , feature (linguistics) , discrete wavelet transform , decimal , computer vision , face (sociological concept) , artificial neural network , facial recognition system , mathematics , wavelet transform , image (mathematics) , wavelet , arithmetic , social science , linguistics , philosophy , sociology
The pose, illumination, and expression variations are challenging tasks in Facial Recognition (FR) and are a popular research area nowadays. We introduce novel nibbles of pixel technique and hybrid features from nibbles in this paper. The color images are converted into grayscale and then converts decimal values from each pixel into eight-bit binary values. The novel concepts of segmenting eight-bit binary into two groups of four-bit binary as Leftmost Nibble (LN) and Rightmost Nibble (RN) is presented. The nibble concept increases the computational speed and decreases the complexity of the system in the case of a real-time system as the 256 shades of grayscale images are decreased to 16 shades for LN and 16 shades for RN i.e., totally only 32 shades instead of 256 shades. The LN and RN binary is converted back to decimal values. The LL subband which is obtained from the applied Discrete Wavelet Transform (DWT) technique on the LN matrix is considered as the most important information while the Histogram of Oriented Gradients (HOG) is applied on the RN matrix to detect the edge information. The linear convolution of DWT and HOG results in the final hybrid features. In the matching part, the Euclidean Distance (ED) matching and the Artificial Neural Network (ANN) are selected to classify the features and calculate the proposed algorithm's performance parameters. The experimentation is implemented on six standard face databases, which demonstrates an outstanding performance by getting higher accuracy with less computation time compared with the existing techniques.