| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide a pseudo defective image automatic creation device capable of creating automatically in large quantities, when creating simulatively a defective image necessary at the initial learning time, and an image inspection device. <P>SOLUTION: A nondefective image is inputted from a nondefective article input part 11 and learned as learning data on a neural network 10. Though the defective image is inputted from a defective image input part 12 and learned, the learning is not sufficient. Accordingly, differential data 17 between itself and the nondefective image are extracted by a defective image extraction part 13, and various defect creation conditions such as a defect synthesis position are produced from a random number value in a random number generation part 18 by a pseudo data condition setting part 14, and the differential data are synthesized on the nondefective image by a pseudo defective image creation part 15 based on the defect creation conditions, and thereby various pseudo defective images are created and inputted as the learning data on the neural network 10. When synthesizing the data, the differential data are enlarged or shrunk, or synthesized after changing its brightness, to thereby enable to increase the pattern of the pseudo defective images. <P>COPYRIGHT: (C)2005,JPO&NCIPI |
