how to read the graphs...
Distribution
Shows the normalized distribution of the deviation for bad and good images. Ideally, theses two curves should not be crossing.
The green line represents the threshold below which a piece will be considered as good. On the opposite the red line represents the threshold above which a piece will be considered as bad.
In between those lines is a grey zone, in which the pieces will be considered as intermediate
ROC (Receiver Operating Characteristic)
Shows the separation of the classes for all possible thresholds. One can take a point on the blue line and read the corresponding true positive on the left of the graph (a defected view where the system found a defect, ideally 100%); and false positive rate on the bottom of the graph (a non defected view where the system found a defect, ideally 0%).
The
AUC
(Area Under Curve) value summaries this curve whereas the maximum of 1.00 corresponds to a perfect classifier.