Automated Industrial Inspection of Capacitor Chips Using Computer Vision

This research proposes a discrete cosine transform (DCT) based approach for automated inspection of small surface defects on capacitor chips. A spatial domain image is transformed into the DCT frequency domain. From the energy concentration analysis of the DCT domain, we can disassembly the frequency matrix into four disassembled matrices. We select the proper number of larger frequency values to represent the random structure features of the capacitor chip surface. Then, we set the selected frequency values to zero and reconstruct the image. For a defective capacitor chip, the small surface defects will be reserved and the random patterns will be eliminated in the restored image. Finally, the entropy method is used to set the threshold for distinguishing between surface defects and uniform regions in the restored image. Experimental results show that the proposed method achieves a high defect detection rate on random textured surfaces of capacitor chips.