Abstract: [PROBLEMS] To provide a feature extracting method for quickly extracting a feature while preventing lowering of the identification performance of the kernel judgment analysis, a feature extracting system, and a feature extracting program. [MEANS FOR SOLVING PROBLEMS] Judgment feature extracting device (104) computes an interclass covariance matrix SB and an intraclass covariance matrix SW about a learning face image prepared in advance, determines optimum vectors ?, ? which maximizes the ratio of the interclass covariance to the intraclass covariance, derives a conversion formula for converting an inputted frequency feature vector x into a frequency feature vector y in a judgment space, and extracts judgment features of a face image for record and a face image for check by using a restructured conversion formula. Similarity computing device (105) computes the similarity by comparing the judgment features.

Abstract: A variation image generation means generates a plurality of variation images having different postures, facial positions, and sizes with respect to a normalized image. A characteristic extraction means extracts a frequency characteristic from the plurality of variation images. A discriminant space projection means projects the frequency characteristic on a discriminant space having high discriminant ability that is obtained by linear discriminant analysis. A reference person comparison means performs a reference person comparison to extract a highly discriminant characteristic. A discriminant characteristic is extracted for a match image using the characteristic extraction means and the discriminant space projection means. A score computation means uses a discriminant axis obtained from a registered image, and the discriminant characteristic obtained from the match image to output a match score.

Abstract: A 3D shape estimation system has a storage device, a relative shape analysis module, a feature point location search module and an absolute shape analysis module. The storage device stores first and second learning data which represent illumination bases and 3D shapes of objects, respectively. The relative shape analysis module calculates an “illumination basis” of an object based on a 2D image of the object and the first learning data, calculates a “relative shape function” that is partial differential of a “shape function” indicating a 3D shape of the object from the illumination basis, and outputs a relative shape data indicating the relative shape function. The feature point location search module extracts a plurality of feature points from the input 2D face image based on the 2D image and the relative shape data, and outputs a feature point location data indicating locations of the feature points.

Abstract: Feature decision means (303) decides a set of features appropriate for pattern identification from a plenty of feature candidates generated by feature candidate generation means (302) by using learning patterns stored in learning, pattern storage means (301). The feature decision means (303) successively decides features according to a reference of information maximization under the condition that the decided feature is known while adding an effective noise to the learning pattern and performs information amount calculation approximately and at a high speed while merging the learning patterns into a set of N elements when required. As a result, it is possible to automatically create a feature set appropriate for pattern identification of a high performance without requiring enormous learning. Moreover, by using a transition table (305) containing transitions between sets, it is possible to perform pattern judgment with a high efficiency.