Abstract
The process of face detection in real-world scenarios is influenced by lighting conditions, background environment, and the possibility of classical techniques to produce false positives, whereas deep learning models require more computing power. This paper presents an integrated face detection algorithm based on the combination of the fast face proposal process through the Haar Cascade detector with wavelet-based feature extraction, followed by CNN-based face validation. In the initial phase of this work, face regions are proposed using a Haar Cascade detector. These selected regions are then further enhanced by utilizing the Haar Discrete Wavelet Transform (DWT) process, and only the LH, HL, and HH components are used for feature extraction. The generated multi-channel features are then fed to a CNN classifier. The performance of the proposed method is tested using a dataset of 2400 face images under different poses, illumination settings, and backgrounds. An accuracy of 96.8%, precision of 95.9%, recall of 94.7%, and F1-score of 95.3% was achieved, in addition to an AUC score of 0.97. A very low false positive rate of 3.8% and an average processing time of 28 milliseconds per image indicate high efficiency and reliability. The novelty of the suggested method is its ability to incorporate the process of enhancing wavelet features with classical Haar-based detection and verification based on deep learning techniques for detecting faces accurately. This combination makes it possible to achieve an excellent balance between precision and efficiency.