Automatic face detection based on bidirectional recurrent neural network optimized by improved Ebola optimization search algorithm

Abstract

Face detection is a multidisciplinary research subject that employs fundamental computer algorithms, image processing, and patterning. Neural networks, on the other hand, have been widely developed to solve challenges in the domains of feature extraction, pattern detection, and the like in general. The presented study investigates the DNN (deep neural networks) use in the creation of facial detection operating systems. In this study, a novel optimized deep network has been presented to face detection. In this paper, after using some preprocessing stages for contrast enhancement and increasing the data number for the next deep tool, they fed to a bidirectional recurrent neural network (BRNN). The network is optimized via a novel enhanced version of Ebola optimization algorithm to provide far greater accuracy. The suggested procedure is examined on GTFD (Georgia Tech Face Database) and the results indicate that the proposed technique significantly outperforms other comparative methods, attaining an accuracy of 94.3%, a precision of 93.51%, a recall of 94.53%, and an F1-score of 92.47%. Furthermore, the method exhibits resilience against various challenges, achieving an accuracy of 95.6% under occlusions, 96.3% under lighting variations, 94.8% under pose variations, and 92.4% under low resolution conditions. Simulation results depict that the suggested technique gives far greater accuracy in comparison with the other comparative approaches.

Keywords: Bidirectional recurrent neural network; Deep learning; Face detection; Improved Ebola optimization search algorithm.

Fig. 1

Sample example of the primary image and its corrected outcome: (A-1) raw image, (A-2) histogram of (A-1), (B-1) corrected image, and (B-2) histogram of (B-1).

Fig. 2

Sample example of the primary image and its corrected outcome: (A-1) input image, (A-2) histogram of (A-1), (B-1) enhanced image, and (B-2) histogram of (B-1).

Fig. 3

Some different face pictures from face database of Georgia tech.

Fig. 4

Different face images from the face detection dataset.

Fig. 5

Confusion matrix of the measurement indicators.

Fig. 6

Graphical results outcomes for the proposed method compared with others without preprocessing for (A) GTF, and (B) FDD.

Fig. 7

Graphical results’ mean results for the proposed method compared with others by preprocessing for both datasets.