PURPOSE: A robust face recognition method through a statistical learning of a regional property is provided to maximize a face recognition rate by dividing each of photographed face images, extracting a regional property from the divided images, and statistically learning the extracted properties. CONSTITUTION: A face recognition device divides multiple training face images respectively into M images, obtains a local property descriptor of each divided image through a SIFT(Scale Invariant Feature Transformation) extracting method, divides one test face image into M images, and obtains the local property descriptor of the each divided image through the SIFT extracting method(S1,S3). The device calculates a distance between the training face image and the test face image by combining a distance between the local property descriptors with a weighted value about the divided image of the each training face image(S6). [Reference numerals] (S1) Obtain a local property descriptor through an SIFT extracting method after dividing each of multiple training face images into M; (S2) Average/dispersion of the local property descriptors; (S3) Obtain a local property descriptor through the SIFT extracting method after dividing a single test face image into M; (S4) Calculate a distance between the local property descriptors of the training face image and the single test face image; (S5) Calculate a weighted value about the divided image of the training face image using the calculated average/dispersion; (S6) Calculate a distance between the training image and the test image by combining the distance between the local property descriptors and the calculated weighted value
展开▼