Fig 1.
Examples of images: (a) without defect, (b)& (c) with small defects, and (d:f) with big defects.
Fig 2.
Examples of features extracted from two different images: (a) original images, (b) gradient, and (c) Gabor features.
Fig 3.
Visualization of the output features from the second hidden layer when applying an autoencoder on: (a) grayscale images, (b) gradient images, and (c) the binary images.
Table 1.
Classification performance with different features for representing images and different classifiers.
The best performances are indicated in bold (the lower the better). These results are obtained using 100-fold cross validation.
Fig 4.
ROC curves of the methods with different images features and different classifiers: the proposed method (top), SVM (middle), and KNN (bottom).
Table 2.
Performance of the methods for detecting defective patches.
Rates are obtained using 100-fold cross validation. Here, we provide the average performance and the standard deviation (Std) over the 100 runs.
Fig 5.
ROC curves of the methods with different features for representing images and different classifiers.
Fig 6.
Examples of successful defects detections (a & b) using the proposed method.
Fig 7.
Examples of incorrect detections: False positive (top row) and false negative (bottom row).