An approach to occluded face recognition based on dynamic image-to-class warping using structural similarity index

An approach to occluded face recognition based on dynamic image-to-class warping using structural similarity index

연구 분야: Artificial Intelligence

학회: Applied Intelligence

Face recognition in uncontrolled environments is a challenging problem in computer vision due to occlusion, pose, and illumination changes. While machine learning techniques address occluded face recognition, they require retraining when updating gallery images. The Dynamic Image-to-Class Warping (DICW) technique offers real-time recognition without training, maintaining the natural order of facial features (forehead, eyes, nose, mouth, and chin) to avoid disruptions caused by occlusion. DICW separates face image patches and integrates them into an ordered sequence through raster scanning. It computes the image-to-class distance between query and target faces using optimal warping paths along temporal and within-class dimensions. This paper proposes an improved face recognition approach using DICW and the Structural SIMilarity (SSIM) index, mitigating variations in illumination and contrast to match structural information. A technique for automatic face recognition from video sequences with DICW is also presented. Experiments on the AR Face Database, Chokepoint Database, and uncontrolled environment video sequences show that the proposed approach significantly improves the recognition rates for occluded images. The proposed approach achieved an improvement of around 5-6% in all considered cases compared to other state-of-the-art approaches.