Year: 2025
Type: article
Abstract: A profile face is an extreme face pose. This makes the facial poses harder to detect compared to normal facial poses due to a lack of facial feature information. Hence, this paper proposes the combination of three different networks to detect the profile faces in an image. The CNN architecture proposed to be used is SSD MobileNetV2 with FPN, where it is divided into three different networks. MobileNetV2 acts as the backbone of the overall architecture. The main function is to extract feature maps from the input image. The feature map is fed into a feature extractor layer that consists of FPN and generates a feature pyramid that consists of feature maps in different scales. The SSD is the detection network that will detect profile faces in the image and produce candidate bounding boxes. The NMS is applied as a post-processing step to remove duplicate detection and low confidence level bounding boxes. As a result, the final output is the highest confidence level of the detected profile face's bounding box in the image. Based on the experimental results, the proposed work has achieved good accuracy in detecting the profile face view.
Institution: Tun Hussein Onn University of Malaysia
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4409971547