Research And Implementation Of A Fast Image Restoration Algorithm

This subject comes from the project ‘Development of the remote video surveillance system based on super-resolution technology', which the goal is to add image restoration to video surveillance system in the circumstances of the poor resolution of the image sensor, the surveillance system can still significantly get high quality video.In this thesis, we use image restoration methods to achieve high-quality video acquisition. Image deconvolution based on the point spread function is the key to recovery image. However, there is an ill-posed problem with deconvolution. As the Hyper-Laplacian model can well fit natural image gray gradient distribution, so we can use this priori to solve the ill-posed problem. The main work and achievements are as follows:Firstly, this thesis reviews the present situation of image restoration methods at home and abroad. A summary of these studies found that image restoration research belongs to the current hotspot, but for engineering applications, especially there are few studies about employing image restoration technology to achieve video acquisition.Secondly, this thesis introduces the related concepts of image restoration techniques based on the point spread function, meanwhile this thesis describes a mathematical model of imaging process and the general idea of imaging restoration t, at last we describes and analysis some common deconvolution image restoration algorithms.Thirdly, this thesis focuses on the image restoration algorithm with high-quality small amount of computation. This part analyzes the video gray change under different conditions of field environments. By comparing different methods for solving the point spread function of blurred image, we select gradient cepstrum method to estimate the point spread function of blurred images. We use image restoration model to analyze the important factors ?, ? and ? of regularization convolution algorithm, and based on the analysis of these factors to optimize these parameters, so we get a small amount of computation, high-quality restoration algorithm. The experiment result has showed that using standard blind deconvolution algorithm can get high quality image when the signal to noise ratio is below 33.3dB; it can get a better image by using Richardson-Lucy restoration algorithm than by using standard blind deconvolution algorithm when the signal to noise ratio is up 33.3dB; the image quality restored by using the fast restoration algorithm is better than by using Richardson-Lucy restoration algorithm when the signal to noise ratio is higher than 37 dB. We use the optimization algorithm to process the blurred image, then the image signal to noise ratio has improved 14.87 dB, and the calculation time has reduced by 5.8%.Finally, we have built the video acquisition system, whose core is based on DM8148 processor, and we have finished a quad-core system development platform's estimation. The algorithm has been validated on this platform. The experimental result has indicated that the algorithm can well achieve HD video acquisition in the circumstances of the poor resolution of the image sensor.