| 摘要 |
PROBLEM TO BE SOLVED: To reliably recognize an object even when characteristic points of the object are insufficiently extracted from an input image when characteristic points are extracted from both the input image and a standard image and the characteristic quantity of each point is calculated to recognize the object (standard image) in the input image on the basis of the characteristic quantities of both images. SOLUTION: Characteristic points are extracted from the standard image in advance, and the characteristic quantity of each characteristic point is calculated by adding position information (a position vector of the characteristic point to a reference point) of the characteristic point to a scale invariant characteristic quantity. Then, characteristic points are extracted from the input image and the characteristic quantity of each point is calculated. A characteristic point extracted from the standard image and a characteristic point extracted from the input image, which have similar characteristic quantities to each other, are set as a corresponding point. Then, for each corresponding point (characteristic point) in the input image, a reference point is calculated on the basis of the characteristic quantity (including position information, etc.) of the corresponding characteristic point in the standard image, and then a vote is casted to an input image plane. If a voted value greater than or equal to a predetermined threshold value exists in a certain small region, an image recognition apparatus determines that an object exists in the input image. COPYRIGHT: (C)2009,JPO&INPIT
|
