WEAK HYPOTHESIS GENERATION DEVICE AND METHOD, LEARNING DEVICE AND METHOD, DETECTION DEVICE AND METHOD, EXPRESSION LEARNING DEVICE AND METHOD, EXPRESSION RECOGNITION DEVICE AND METHOD, AND ROBOT DEVICE

摘要

PPROBLEM TO BE SOLVED: To provide an expression recognition system that can recognize an expression with robustness, for example, to a shift of a face position included in an image and with high precision by using a face detection device improved in learning efficiency and thus detection processing speed in detecting an area showing a detection target by ensemble learning; and a learning method therefor. PSOLUTION: When learning by AdaBoost, the face detection device repeats a processing of selecting high performance weak hypotheses from all weak hypotheses, of generating new weak hypotheses from them according to statistical properties, and of selecting one with the highest discrimination performance from them, to thus generate weak hypotheses sequentially until obtaining a final hypothesis. In detection, every time the determination result of one weak hypothesis is output, with an abortion threshold learned in advance, whether an obvious denial of a face is possible or not is determined, and if the determination is affirmative, the processing is aborted. From the detected face image, a given Gabor filter is selected by the AdaBoost technique, and support vectors only of feature values extracted by the selected filter are learned for expression recognition. PCOPYRIGHT: (C)2005,JPO&NCIPI