Research On Image Artifact Detection And Inpainting In Dirty Camera Modules

Camera module is a critical electronic device for image capture,which is widely used in mobile digital devices and smart phone products.As a core component of the camera,quality inspection of camera modules has been an important part of the industrial camera manufacturing process.At present,the quality inspection is mainly carried out manually,which is inefficient and can cause high misjudgment rate due to subjective judgment of different personnel.In addition,some camera modules with large dirty areas were discarded by the manufacturer because they could not correct the problem,which undoubtedly increases the production costs for manufacturers.So the study of an effective camera module dirty detection and inpainting method is very necessary.The existing algorithms have low detection accuracy and high false detection,and there are edge blurring and jagged effects in the image of the dirty repair results.In this thesis,based on normalized correlation algorithm and Total Variation(TV)and Fast Marching Method(FMM),we mainly focus on the problem of dirty detection and repair of camera modules,and the main work and research results are as follows.First,aiming at the problem of the detection of image artifacts in dirty camera modules,this thesis analyzes the main causes and characteristics of the formation of artifacts,and proposes a detection model combining the normalized mutual correlation algorithm and the region growth algorithm,firstly,the coordinates of the possible existence of artifacts are obtained by the normalized mutual correlation algorithm,and then the image and nonextreme value suppression in the original image using the region growth algorithm,for the detection of artifacts in industrial camera modules provides a new way of thinking,and also provides a repair location for the image artifact repair task in the camera module.Second,aiming at the problem of the image artifact inpainting in dirty camera modules,this thesis first explores the poor edge retention problem of the TV model,and optimizes the gradient descent equation in the model by introducing a weight balance function,so as to perform adaptive inpainting in edge regions with different gradient variations.The experimental results show that the textures of the repaired regions are more natural using the improved algorithm,and the algorithm has good robustness.Third,to address the problem of excessive time complexity in the improved TV model,this thesis introduces the concept of confidence matrix based on the FMM,which calculates and updates the confidence level of each pixel to reduce the error accumulation during the inpainting process and improve the continuity of the image structure.The main gradient direction calculation is also discussed,and the neighborhood range of the pixels is adjusted so that the actual number of gradient directions in the divided direction interval is distributed more evenly.In addition,the inpainting order of FMM is optimized,and the actual restoration order is measured by three factors: pixel confidence,the number of pixels in the neighborhood deviating from the main gradient direction and the distance from the inpainting boundary.The experimental results show that the inpainting algorithm of using constraint matrix processing improves the smoothness of the image and reduces the jagged effect phenomenon that occurs in the original algorithm,and the algorithm has low time complexity and good portability.