engleski

Generating Artificial Biscuit Tile Images

This paper addresses the problem of creating a large image set with artificially inserted biscuit tile surface defects. Those defects commonly occur in real-life ceramic tile production. However, the occurrence rate of a defected tile is neglectable to the rate of a non-defected tile. To overcome the gap in the size of the two sets of images, with defects and without defects, artificial image generation increases the number of images with defects. This paper enables the future use of machine learning algorithms, which require a relatively equal size of the two image sets. The analysis of surface defects pointed to a similarity of the surface defect to an inverted Gaussian curve. The Gaussian distribution model is proposed for modeling a biscuit tile surface defect. Captured images with surface defects are compared with corresponding artificially created images. Image set without defects is used to create the image set with defects.

ceramic tile ; augmentation ; artificial ; defects ; detection

nije evidentirano