Texture defect detection detects defects in a texture based on a texture classifier trained with texture samples that do not contain defects. The texture classifier is trained to recognize texture samples that are acceptable in the current inspection. The texture defect detection algorithm accepts an image of a texture surface as an input, identifies texture defects, and returns a binary image of the texture defects. The following figure illustrates typical input and output images.

Shift Variation

Texture defect detection is invariant to shift. For example, if the texture in the inspection image shifts vertically or horizontally from the trained texture samples, the texture defect detection algorithm continues to correctly identify any texture defects.

Rotation Variation

Texture defect detection is invariant to rotation of approximately ±5 degrees. If the texture under inspection can shift more than 5 degrees, you must train the classifier with texture samples at every expected orientation. The following figures illustrate the same texture at distinct orientations that require trained samples for each variation.

Scale Variation

Texture defect detection is invariant to rotation of approximately ±10 degrees. If the texture under inspection can vary more than 10 degrees in scale, you must train the classifier with texture samples at every expected scale variation. The following figures illustrate a difference in scale that require trained samples for each variation.