Summary: | 碩士 === 國立臺灣大學 === 資訊網路與多媒體研究所 === 105 === Since the smart devices are popular in recent years, the demand for optical lens continues greatly. According to Industrial Economics and Knowledge Center (IEK) [3], the global shipments of optical lens will reach 4.3 billions on 2016, and increase to 4.6 billions on 2017. In order to maintain high quality, the defect detection is necessary on production line. Therefore, we need a robust dirt detection method.
We propose a method for automatic lens dirt detection. Our algorithm is based on high-pass filter, and it could amplify the stain which cannot be discriminated by eyes. Besides, we use this dirt detection method before lens shading correction; it means that our method runs successfully on images with lens shading effect.
Our research mainly has two parts: image enhancement and dirt detection. For image enhancement, we assume the stains are high frequency and adopt our high-pass filter to enhance the stains. We mix unsharp masking, low-pass filter, color inversion, and linear light to simulate the high-pass filter on Photoshop CS3. On the other hand, we remove the noise produced by high-pass filter, enhance the contrast again and convert to binary image to show our result on dirt detection.
|