摘要 |
PROBLEM TO BE SOLVED: To reduce noise to be generated when a memory color region is presented as a stochastic mask, and eliminate an unnatural mask boundary part.SOLUTION: An image processing method includes: a learning stage (a feature selection processing part 1, a bin size determination processing part 2, and a probability density calculation processing part 3) for selecting a feature and calculating a separate hyper-curved surface, by use of a learning image set; and a recognition stage (the probability density calculation processing part 3, and a post-processing part 4) which processes a new input image by use of the feature and the separate hyper-curved surface obtained by learning. In order to select a feature suitable for an object to be detected, a canonical discriminant coefficient calculation processing part 11 calculates a canonical discriminant coefficient of each feature. A feature selection processing part 12 selects a feature in descending order of the canonical discriminant coefficient. |