Facial behaviors represent activities of face or facial feature in spatial or temporal space, such as facial expressions, face pose, gaze, and furrow happenings. An automated system for facial behavior recognition is always desirable. However, it is a challenging task due to the richness and ambiguity in daily facial behaviors. This paper presents an efficient approach to real-world facial behavior recognition. With dynamic Bayesian network (DBN) technology and a general-purpose facial behavior description language (e.g., FACS), a task oriented framework is constructed to systematically represent facial behaviors of interest and the associated visual observations. Based on the task oriented DBN, we can integrate analysis results from previous times and prior knowledge of the application domain both spatially and temporally. With the top-down inference, the system can make dynamic and active selection among multiple visual channels. With the bottom-up inference from observed evidences, the current facial behavior can be classified with a desired confidence via belief propagation. We demonstrate the proposed task-oriented framework for monitoring driver vigilance. Experimental results demonstrate the validity and efficiency of our approach.
This paper presents a method to obtain the optimal motion description from two consecutive images including multiple moving parts. It copes with segmentation and motion estimation problems which resemble the well known "chicken and egg" relation. "Segmentation" is necessa y for amotion estimation" of each part, and vice versa. Unkke previous approaches based on the motion similarity between two pixels, we propose t o use an information measure ap roach, based on comparisons between an individual ( r pixel) and a class (or set of pixels,). First, the motion of an edge segment as optimally modeled. Next, merging and splitting processes are iterated until the minimum descriptaon is obtained for the whole image. As a result, the im.age is segmented into several regions, each of which is represented by an edge segment list, and at the same time the maximum likelihood motion estimation is obtained f o r each region. Experiments performed on real images are shown.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.