Research On Low Quality Video Restoration And Moving Object Detection

In military field such as military image reconnaissance system，missile imageguidance system and civil field such as the urban intelligent transportation system andintelligent security systems. Image degradation is easily caused by the relative motionof the camera and target or low illumination of imaging environment. It will reduce thequality of video image and bring interference to the visual observor and moving targetdetection. In order to remove the effects of low quality image for moving targetdetection, this paper conducted the following research.At first, to enhance low illumination image, a novel low illumination imageenchancement method based on dark-channel prior with the capability of preservingscene color is presented. The intensity of every image pixel with dark-channel prior isestimated directly to estimate the illumination intensity of the scene. For dealing withwrong dark-channel values in white region, a correction method is pproposed. Then, alinear smoothing method is used to refine the light intensity distribution. The simulationexperiment shows the proposed methodis effective.Then, to sovle theerrors in blind image restoration methods caused by insignificantedges, we propose a improved normalizing regularizing image restoration method withinsignificant edges and texture suppression to reduce kernel estimation errors. Firstly,we filter the image to get the gradient image. Secondly, we use Shock filter to removeinsignificant image edges and propose a method to reduce the amplified noise caused byShock filter. Then, a sparse regularization function is used to guide the blurred imagerestoration. Finally, various experiments are conducted to verify that our improved blindimage restoration algorithm is able to deblur the motion blurred images in differentscenes and blurred conditions.Finally, we use a saliency method based on histogram contrast to detect movingobject. We calculate the significant value of each pixel in image and reduce the amountof calculation by decline the number of colors in image. Considering the influence ofspatial relations, region contrast method is introduced to make the significant valuecalculation more accurately.And the moving target and background objects aredistinguished by background modeling method. Through simulation experiment, themethod is proved effective to moving object detection after low quality imageenhancement.