| 摘要 |
A method for recognizing a facial gesture by using a three-dimensional cylinder head model are provided to offer the pose information of a three-dimensional pose parameter, which is proper to the quick movement and luminance/pose changes, and the recognition information of eye blink through yes/no recognition and AAM(Active Appearance Model) fitting by using an HMM(Hidden Markov Model) recognizer. An image input unit(101) inputs an image, a facial outline detection and feature detection extraction unit(102) detects a facial outline from the inputted image and eye location information from the detected facial outline. A head detection unit(103) obtains a templet image and an image, which is to be matched with the templet image, from the inputted image by using a three-dimensional cylinder head model in a fitting process to obtain a three-dimensional pose parameter which minimizes the difference between the inputted image and the templet image, and minimize the difference between the two images. A HMM recognizer(104) performs yes/no judgment by receiving the value of the three-dimensional pose parameter outputted from the head detection unit. An AAM fitting and eye-blink recognition unit(105) defines the AAM which detects the eye movement when the image reflects the front view of a face, and obtains the size of the major and minor axes of an eye from feature points around the eye after the AAM fitting. Then, the AAM fitting and eye-blink recognition unit judges the opening or closing state of the eye by using a comparison function. |
