Image inspection apparatus

1. An image inspection apparatus for performing pass/fail determination of an inspection object on the basis of an image obtained by imaging the inspection object, the image inspection apparatus comprising:

an image input unit configured to input a non-defective product image group and a defective product image group, the non-defective product image group composed of multiple non-defective product images added with an attribute as a non-defective product by a user, the defective product image group composed of defective product images added with an attribute as a defective product by the user;

a classifier generator configured to input the non-defective product image group and the defective product image group, which are input by the image input unit, to a learning processor with a multi-layered structure, plot the non-defective product image group and the defective product image group in a feature space, and set a classification boundary between the non-defective product image group and the defective product image group plotted in the feature space, thereby generating a classifier for distinguishing between a non-defective product image and a defective product image;

a notification unit configured to, in a case in which an additional image that is added with an attribute as either one of a non-defective product and a defective product is newly input to the learning processor by the user, and the additional image is plotted in the feature space, estimate the probability that the attribute of the additional image is wrong, on the basis of a distance between the position of the additional image in the feature space and the classification boundary or a representative point of the images with the other attribute, the notification unit further configured to provide a notification of the probability of wrong attribute of the additional image when the attribute of the additional image is expected to be wrong; and a selection unit configured to receive result of selection whether to correct the attribute of the additional image, by the user who receives the notification of the probability of wrong attribute of the additional image, from the notification unit,

wherein the classifier generator is configured to determine the attribute of the additional image on the basis of the result of selection received by the selection unit and to correct the classification boundary in accordance with the determined attribute.

2. The image inspection apparatus according to claim 1 , wherein the notification unit is configured to expect the probability that the attribute of the additional image is wrong in a case in which the distance between the position of the additional image in the feature space and the representative point of the images with the other attribute is less than a threshold.

3. The image inspection apparatus according to claim 1 , wherein the representative point is a point closest to the additional image, among the multiple images with the other attribute, which are plotted in the feature space.

4. The image inspection apparatus according to claim 1 , wherein the representative point is a centroid of the multiple images with the other attribute, which are plotted in the feature space.

5. The image inspection apparatus according to claim 1 , wherein the notification unit is configured to expect the probability that the attribute of the additional image is wrong, in a case in which, compared with a distance between the position of the additional image in the feature space and a representative point of the images having an attribute as either one of a non-defective product and a defective product, the distance between the position of the additional image in the feature space and the representative point of the images with the other attribute is short.

6. The image inspection apparatus according to claim 1 , wherein, in a case in which the attribute of the additional image is expected to be wrong, the notification unit is configured to determine whether an image having a predetermined degree or more similarly relative to the additional image is already input by the image input unit and to not provide a notification when determining that the image is still not input.

7. The image inspection apparatus according to claim 1 , wherein the notification unit is configured to add a mark that shows the probability of erroneous registration, to the additional image that is expected to have a wrong attribute, and to provide a notification when another additional image is plotted in the feature space and a similarity between the another additional image and the additional image added with the mark is determined as being a predetermined degree or more.

8. The image inspection apparatus according to claim 7 , further comprising a display unit configured to display the additional image added with the mark and the another additional image having the predetermined degree or more similarly relative to the additional image added with the mark, when the notification unit provides the notification.

9. The image inspection apparatus according to claim 1 , wherein the notification unit is configured to calculate a reliability on the basis of the distance between the position of the additional image in the feature space and the representative point of the images with the other attribute, and the notification unit is further configured to provide a notification that the additional image does not contribute to learning, which is performed by the learning processor, in a case in which the calculated reliability is less than a predetermined degree.