| Summary: | 碩士 === 國立臺北科技大學 === 自動化科技研究所 === 103 === This thesis proposes a novel defect detection algorithm for the surface defects of smartphone that overcomes the difficulty of low-contrast, non-uniform illumination, and low S/N ratio situations. Thers are two parts proposed in the image preprocessing. Firstly, a bright preserving weight clustering (BPWC) is applied to preserve the image brightness and its features/ visualization after image enhancement. Secondly, the anisotropic diffusion method is adopted to not only reinforce the low-contrast defect but also keep the gradient of the high-contrast defects. Then, the region of interesting (ROI) and nonintersecting can be extracted by the gradient direction and the geometric relationship. After preserving the characteristics of ROI, the defects are roughly segmented by region growing method which aggregates the pixels together when the magnitude of gradient is beyond the threshold Furthermore, the accurate position of defects are segmented bythe proposed adaptive region growing algorithm which averts from the influence of the non-uniform illumination. Finally, the experimental results demonstrate that the surface defects can effectively be detected in the conditions of high and low contrast, non-uniform illumination, and low S/N ratio on the surface of smartphones.
|
|---|
