Development of an enhanced support vector machine face recognition system

Face recognition biometric authentication focuses on uniquely recognizing human facial appearance based on inherent physical traits of the face for application in access control. The use of face for recognition has been proven to be highly reliable and effective. This research performed a performance evaluation of SVM-based variants in the recognition of facial images. Six facial expression images, each from sixty individuals, were locally acquired using a Canon EOS 2000D digital camera at 200×200-pixel resolution, 240 images were used for training, while 120 images were used for testing. The acquired images were converted into grayscale and normalized using the histogram equalization method. Features classification was carried out using a Support Vector Machine for PCA-PSO and PCA, respectively. The performance of the two techniques was evaluated and compared at a 0.42 threshold using Recognition Accuracy (RA), Precision (P), Sensitivity (S), and Recognition Time (RT). The validation of the techniques was done using t - a t-test at a significant 5% level. The RA, P, S, and RT were 97.50%, 97.80%, 98.89%, 1487.16 s, and 3.80s for PCA-PSO-SVM, while the corresponding values for PCA-SVM were 95.83%, 96.70%, 97.78%, 1861.79 s, and 22.96 s, respectively. The paired t-test was P = 0.001 with a mean difference of 2.5%. The PCA-PSO-SVM technique performed better than PCA–SVM for all metrics. A face recognition system based on PCA-PSO-SVM is a more reliable security surveillance system than PCA-SVM