Abstract
This paper presents three novelty aspects in developing biometric system-based face recognition software for human identification applications. First, the computations cost is greatly reduced by eliminating the feature extraction phase and considering only the detected face features from the phase congruency. Secondly, a motivation towards applying a new technique, named mean-based training (MBT) is applied urgently to overcome the matching delay caused by the long feature vector. The last novelty aspect is utilizing the one-to-one mapping relationship for fusing the edge-to-angle unimodal classification results into a multimodal system using the logical-OR rule. Despite some dataset difficulties like Unconstrained Facial Images(UFI) which includes varying illuminations, expressions, occlusions, and poses, the multimodal system has highly improved the accuracy rate and achieved a promising recognition result, where the decision fusion is classified correctly (84, 92, and 72%) with only one training vector per MBT in contrast to (80, 62, and 68%) with five training vectors for Normal matching. These results are measured by Eucledian, Manhattan, and Cosine distance measure respectively.
Highlights
Phase congruency (PC) is a contrast and illumination invariant measure of valuable features.Unlike gradient-based feature detectors, which can only detect the first or second derivative features, phase congruency correctly detects all kind of phase angle related features, and not just step features having 0 or 180 phase angle degrees
mean-based training (MBT) training performance The Euclidean classifier based unimodal system satisfied 96.2% maximum accuracy while the decision fusion improved this rate to 98%
The Manhattan classifier based unimodal system satisfied 92.8% maximum accuracy while the decision fusion improved this rate to 96%
Summary
Phase congruency (PC) is a contrast and illumination invariant measure of valuable features. The complexity in calculating the PC features, it is considered an immune method against noise, contrast variations, and illumination. One of the biggest challenges in this work is the removing of the feature extraction stage for reducing the compuations complexity and considering only the detected PC features. The multi-focus image fusion applied new fusion rule and complex Gabor wavelet to obtain the benefits of PC sharpness in finding new focus measure [10]. Two different multi-spectral image fusion rules for nonsubsampled contourlet transform (NSCT) was introduced with: PC, principle component analysis (PCA), directive contrast, and entropy for developing integrity model [11]. The sum modified Laplacian (SML), contrast features measure, and non-subsampled shearlet transform (NSST) were integrated to reconstruct a model for better fusion of edge information [13]. The negatively impact was avoided using the same fusion rule on different scales, a Gaussian filter with multi-scale decomposition (MSD) of total variation (TV) and PC were proposed for designing the modal [17]
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: Bulletin of Electrical Engineering and Informatics
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
