Concerning the problem of image subtraction methods which are sensitive to matching precision and disturbances in product fabricating process when it is applied to defect detection

and problem of missing detection with single mask algorithm

an improved image subtraction strategy based on double masks is proposed for the detection of industrial products surface defects. Edges of template and test images are extracted and dilated

which are used for the template mask and the test mask. After fusion process on the two masks

the synthesized one is convoluted with the template and the test images

the surface defects are detected with two convoluted images subtraction. Experiments show that the methods can detect multi-defects on surfaces

and satisfies real-time requirements.