The Blemish Detection Research Of Camera Module Based On Background Difference

Camera module has a very extensive application in digital products, and has become an indispensable part in people's daily life, its quality requirements are also getting higher and higher,however in the various visual defects of camera module, blemish is more common, and the most difficult to detect. Based on the traditional method of relying on human's eyes to detect blemish with low time efficiency, and causing a dispute over the judgment of slight blemish in the process of detection, which is influenced by subjective psychological factors, this method is difficult to meet the requirements of camera module's product quality and production efficiency. Therefore,it has very important research significance and application value for developing a kind of blemish visual inspection method which is accurate fast and free from external conditions.To realize the automatic operation, this paper presents a method to extract blemish targets from complex background intensity through single background modeling. By using different modeling algorithm, split out, then judge and assess suspected target area respectively. This article expounded the research content mainly from four parts: image preprocessing, object region extraction and classification analysis, blemish determination and classification, as well as the analysis and comparison of different modeling algorithm.First collect surface imaging of acrylic board by using camera module, analyze the imaging features of image and performance characteristics of blemish; then through the preprocessing such as: down sampling, contrast stretching and smooth filtering, access to the gray image which target features is obvious, and convenient for observation and processing; After that, execute background modeling respectively used slider scanning, linear interpolation, median filter and B-spline surface fitting, then the segmented foreground pixels are analyzed by connected domain to obtain a suspected target area, simultaneously calculate the main characteristics of contrast,area size and circularity of this area; subsequently, the method of judging suspected targets by decision tree, and completing severity level classification of blemish targets by fuzzy pattern recognition was proposed.Finally, according to the technical route of this paper, the experiment is designed to verify the result. Through detecting a large number of real defect samples and normal images and calculating the correct detection rate and detection error rate under different algorithms, the advantages and disadvantages and robustness of various modeling methods are analyzed.Experimental results show that background modeling algorithm based on linear interpolation and B-Spline surface fitting can segment the target more complete and ensure the targets' shape and gray features, however, that algorithm based on the slide scanning and median filtering will make a difference between the area and the performance of the detected target, especially the details will be lost when the blemish is light. At the same time, these algorithms will inevitably have a certain degree of missed detection for the blemish defects in the vicinity of the dark corner.