Camera Module Defect Detection Using Gabor Filter and Convolutional Neural Network

Defect detection in camera module is closely related to the yield of camera module production line. As a result, many researches are being done regarding defect detection in camera modules. Conventional methods for detecting defects use template matching or edge filter with lens shading algorithm applied as preprocessing. Template matching method requires a non-defective image to use as a template which is not easy to acquire. Also, due to some defects occurring identical to noise most of the methods were only able to detect prominent defects. To overcome this problem, we propose a defect detecting method which can detect weak defects using a single image. The proposed method detects PP (Pixel Particle) defects by absolute differencing the original image with the blurred image of itself. Stain and crack defects are detected by dividing the image into subimages then comparing each subimage with the nearby subimages. Detected defect areas are fed to convolutional neural network for classification and verification. Experimental results show that the proposed method was able to detect even weak defects while correctly classifying the detected defects.