Example-Based Super-Resolution Reconstruction And Its Application In The Defect Detection Of PCB

The image has become one of the most widely used information carrier in people's daily life,and 70 percent of information obtained by people comes from it.However,with the development of internet technology,multimedia applications have been emerging constantly.There is higher demand for image and video technology.In actual imaging process,there are blur,noise and artifacts in most images,which not only have great influence on images' quality,but also introduce many difficulties for image classification,recognition,analysis and understanding.Therefore,it is necessary to improve the resolution of the image.There are two technologies,hardware technology and software technology,to increase the resolution of obtained images.Improving the performance of the hardware may be not feasible due to the increasing cost.Software method,super resolution reconstruction technology,is flexible and cost-effective.The thesis makes a deep research on example-based super resolution,in which the philosophy of regression is adopted to achieve neighbor selection,deviation matrix learning,mapping relationship estimation and global regression matrix learning.The major contributions are as follows:(1)A deviation learning based image super resolution method is proposed.Image super resolution is a highly ill-posed problem.A low resolution image may correspond to many high resolution images in super resolution reconstruction.We can solve the problem by a projection matrix between the high dimensional and the low dimensional feature space.However,there is a high frequency detail error between the high resolution image obtained by projection matrix and the original image,which can affect the quality of the reconstructed image.To solve the problem,the thesis proposes an image super resolution reconstruction method based on deviation learning.A deviation mapping matrix is calculated by searching for the relationship between the error and the low resolution image.Finally,the projection matrix and the deviation matrix are both utilized to reconstruct a high-resolution image.In this way,there are more high frequency details and better visual effects in the reconstructed image.(2)An image super resolution method based on global regression is proposed.The projection matrix between the high and the low dimensional feature space is calculated by image patches' local neighbors.Therefore,it only includes image patches' local information and ignores the contribution of global information,which can't generate realistic texture details easily.To solve the problem,the thesis proposes one method of image super resolution reconstruction based on global regression.The method utilizes global information of the dataset to increase image details.Finally,local and global information is combined to reconstruct high resolution image with high quality.(3)The thesis indicates the effectiveness of proposed methods,and the best parameters are selected through experiment comparison for each approach.Compared with several state-of-the-art methods,our methods improve the objective quality evaluation index and there are better visual effects in reconstructed images.(4)Both of the example-based super resolution methods proposed in this thesis are effectively applied to PCB defect detection.The proposed methods are utilized to deal with PCB images before defect detection,obtaining PCB images of high quality.Then the defects such as short circuit,open circuit,bulges,depressions and holes can be detected.