I am working on the development of a defect detection task. When using a segmentation model for detection, I've found that it is not effective in detecting defects when copper lines are completely covered.
Therefore, I aim to compare a design blueprint (CAM) with the original image (after its transformation into binarized images) to identify the differences. The challenge lies in the fact that the circles in the design blueprint are outlined, while the binarized version of the actual image renders them solid. Consequently, subtracting the two images does not yield a meaningful result.
The example of original image and binarization image:
The Original image(its transformation into binarized images)
Are there any algorithms or neural network approaches that can address this type of problem?