摘要 |
PROBLEM TO BE SOLVED: To reduce a probability for erroneously detecting that an object of a non-detection target is an object of a detection target when detecting the object of the detection target, such as a face included in image data.SOLUTION: An image patch data detection target likelihood model data generation part 7B preliminarily generates detection target likelihood model data for converting feature quantity of image patch data Pd into face likelihood representing the similarity of a face image by using positive instance learning data 12. A two-class discrimination part 5B calculates feature quantity of inputted image patch data Pd, subsequently converts the calculated feature quantity into face likelihood by using detection target likelihood model data 13B, does not discriminate that a class of the inputted image patch data Pd is a face when the converted face likelihood is equal to or less than a prescribed face likelihood threshold, and stops two-class discrimination processing. |