Abstract: An extraction section extracts, from each of a plurality of non-defective product images which show the appearance of the inspection target determined to be a non-defective product, the non-defective vector representing a feature of each non-defective product image. A generation section generates a transformation matrix by using a plurality of non-defective product vectors extracted by the extraction section. The transformation matrix is a matrix representing sequentially performing first mapping for mapping the feature vector to a feature space and second mapping for mapping a result of the first mapping to the whole space to which the feature vector belongs. An adjusting section adjusts each element of the transformation matrix generated by the generation section by using, as learning data, the feature vector extracted from a pseudo defect image.

Abstract: An extraction section extracts, from each of a plurality of non-defective product images which show the appearance of the inspection target determined to be a non-defective product, the non-defective vector representing a feature of each non-defective product image. A generation section generates a transformation matrix by using a plurality of non-defective product vectors extracted by the extraction section. The transformation matrix is a matrix representing sequentially performing first mapping for mapping the feature vector to a feature space and second mapping for mapping a result of the first mapping to the whole space to which the feature vector belongs. An adjusting section adjusts each element of the transformation matrix generated by the generation section by using, as learning data, the feature vector extracted from a pseudo defect image.