Cascade learning from adversarial synthetic images for accurate pupil detection. (April 2019)

• Record Type: Journal Article
• Authors: Gou, Chao; Zhang, Hui; Wang, Kunfeng; Wang, Fei-Yue; Ji, Qiang
• Is Part Of: Pattern recognition. Volume 88(2019:Apr.)
• Page Start: 584
• Page End: 594
• Publication Date: 2019-04
• Subjects: Cascade regression -- GANs -- Pupil detection; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2018.12.014
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We propose a unified framework to learn shape augmented cascade regression models for accurate pupil detection. We exploit the adversarial training to refine the synthetic eyes with texture and appearance from real images. By leveraging the power of cascade regression, the proposed method iteratively estimate the eye related key point locations. The proposed framework achieves the state-of-the-art results of pupil detection on BioID, GI4E and LFW. Abstract: Image-based pupil detection, which aims to find the pupil location in an image, has been an active research topic in computer vision community. Learning-based approaches can achieve preferable results given large amounts of training data with eye center annotations. However, there are limited publicly available datasets with accurate eye center annotations and it is unreliable and time-consuming for manually labeling large amounts of training data. In this paper, inspired by learning from synthetic data in Parallel Vision framework, we introduce a step of parallel imaging built upon Generative Adversarial Networks (GANs) to generate adversarial synthetic images. In particular, we refine the synthetic eye images by the improved SimGAN using adversarial training scheme. For the computational experiments, we further propose a coarse-to-fine pupil detection framework based on shape augmented cascade regression models learning from the adversarial synthetic images. Experiments on benchmark databases of BioID, GI4E, and LFW show that the proposed work performs significantly better over other state-of-the-art methods by leveraging the power of cascade regression and adversarial image synthesis.