Studies On Image Restoration In Video Surveillance

Currently video surveillance system has been widely applied for the surveillance andmanagement of the places such as the bank, the supermarket, the parking area, the residentialquarter, the street and so on. In most cases, the ordinary low resolution video camera can well meetthe need of video surveillance and management, but in some special circumstance, ordinary lowresolution video camera can’t meet the need. Therefore this paper researches in the low resolutionvideo camera condition, how to use the video sequence realize high resolution image reconstruction,in order to satisfy the needs of the analysis and processing.The high resolution image reconstruction by video sequence involves of two key steps that areimage registration and image reconstruction. This paper studied the basic method and the specificimplementation of image registration and image reconstruction in the condition of the singlefixed-focus video camera.First of all, in the image registration, a popular image registration algorithm that based onfeature points the common characteristic points is optimized. Because of the local concentration offeature points, it would be affected that the calculation accuracy of the transformation matrix. Herewe control the distance between the feature points, avoiding the distribution of the feature pointsare too centralized, so as to improve the precision of the transformation matrix calculation.Secondly, in the image reconstruction, this paper puts forward an image reconstructionalgorithm based on the diffusion weighted least squares. The algorithm considers the correlationbetween pixels in the image. When estimate the current pixel value, we not only use theinformation itself but also use its neighborhood pixel information, improving the effect of imagereconstruction effectively. Due to the diffusion mechanism it may lead to excessive smoothing ofthe reconstructed image, we process the diffusion least square objective function to be weightedand control the iteration times by the difference between the current and former reconstructedimage so that avoid the excessive smoothing of the reconstructed image.Finally, combine image registration and image reconstruction to form a complete highresolution image reconstruction system based on video. The first step is carrying out imageregistration on the video sequences, and then reconstructed the image by the former images usingthe method that the diffusion weighted least squares image reconstruction algorithm. Carry out thesimulation by different angles and fixed-focus image sequence, and results verify the effectivenessof the proposed scheme.