Facial Expression Analysis by Image Processing

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Facial expression provides cues about emotional response, regulates interpersonal behavior, and communicates aspects of psychopathology. Human-observer based methods for measuring facial expression are labor intensive, qualitative, and difficult to standardize. Our interdisciplinary team of computer and behavioral scientists has developed the CMU/Pitt Automated Facial Image Analysis (AFA) system that is capable of automatically recognizing facial action units and analyzing their timing in facial behavior. The quantitative measurement achieved by AFA represents a major advance over manual and subjective measurement without requiring the use of invasive sensors. We envision to use AFA's reliable, valid, and efficient measurement of emotion expression and related nonverbal behavior for assessment of symptom severity in depression. Current methods of clinical assessment of depression depend almost entirely on verbal report (clinical interview and/or questionnaire). They lack systematic and efficient ways of incorporating behavioral observations that are .strong indicators of depressive symptoms, especially those related to the timing of dyadic interaction between clinician and patient, much of which may occur outside the awareness of either individual. AFA is capable of extracting both the type and timing of nonverbal indicators of depression. Our hypothesis is that quantitative measures of the configuration and timing of facial expression, head motion, and gaze obtainable by AFA will improve clinical assessment of symptom severity and evaluation of treatment outcomes when combined with information from interviews and self-report questionnaires. We propose to test this hypothesis in 40 participants participating in a treatment intervention study for major depression. Interview, questionnaire, and video data will be collected at regular intervals over the course of treatment. To measure social dynamics, both patient and interviewer will be video recorded and processed using AFA. Longitudinal multilevel modeling will be used to test study hypotheses. We will improve further algorithms and capabilities of AFA to meet evaluation goals and prepare AFA for use by the scientific and clinical community.