METHOD AND DEVICE FOR COLLATING IMAGE

PROBLEM TO BE SOLVED: To provide a method and a device for performing image collation at high speed with high accuracy even in the case of an input image in the state of making the direction of a recognition object or the like different from a model image. SOLUTION: The image of a face area is segmented out of the image (S1). A feature point is found out on the basis of respective pixel values, a peripheral local area (window) image is segmented (S2) and dimensional compressed image conversion is performed (S3). These steps S1-S3 are respectively executed to the model image and the input image. Model window data and input window data are compared and collated inside a feature space and the window data of a closest distance inside the feature space are made correspondent (S4). Plural kinds of combination are selected out of correspondent window groups and on the basis of position information inside the image, a geometric conversion parameter expressing a geometric difference between both the images is estimated (S5). Correcting processing is performed on the input image by the obtained geometric conversion parameter (S6) and the similarity of image collation is calculated (S8) through relative position voting processing of the correspondent window (S7).