Defects on industrial objects can have arbitrary shapes and colors and they can differ in size.

The surface structure and calor of the product can also variate from batch to batch and from part to part.

Besides, in many cases the product can have a complex surface, composed of different structure types

Our method of feature detection and classification combined with machine learning, is robust in detecting defects and is invariant to changes in defect shape as well structural background.

Our machine learning approach also enables teaching the system new defect types as they arise.

Our algorithm is capable of detecting arbitrary shaped defects and anomalies on varying and unstructured surfaces.