摘要 |
PROBLEM TO BE SOLVED: To provide a device or a method for raising accuracy of image recognition by stably generating many local feature quantities.SOLUTION: A parameter adjustment unit 2 determines first to n-th parameter values for converting a learning image. An image conversion unit 3 generates first to n-th conversion images by converting the learning image by using the first to n-th parameter values. A local feature quantity extraction unit 4 specifies feature areas in each of the first to n-th conversion images, and extracts local feature quantities from the feature areas. A cluster generation unit 5 generates clusters which can be utilized as data for recognition by clustering the local feature quantities. |