Research On Repair Technology Of Old Film Scratches And Plaques

Old film are the cultural heritage of mankind.Due to physical damage and the chemical nature of the film itself,old film have problems such as scratches and plaques.Convolutional neural networks have achieved good results in the field of video image processing,and it is of great significance to study old film restoration techniques based on convolutional neural networks.The main research contents of this paper are as follows:1)The detection effect of traditional scratch detection algorithms is greatly affected by the threshold,and it is difficult for a single threshold to meet the detection requirements when used in a video sequence.The scratch shape is simple and fixed and similar to medical images.In this paper,U-net,which is widely used in the field of medical image processing,is used for scratch detection.Create a scratch data set,improve the network according to the characteristics of the narrow width of the scratch,and train the network parameter model to achieve convergence.The image with scratches is preprocessed and sent to the network,and finally the semantic segmentation image of the scratches is filtered and complemented to obtain the mask image damaged by the scratches.Experimental results show that this method has a better detection effect on obvious scratches.2)Due to the random distribution of plaques,plaque detection needs to use motion compensated inter-frame information.The traditional two-dimensional block matching motion estimation algorithm has better search quality,but the search speed is not fast enough.Aiming at the insufficiency of the search speed of the traditional block matching method,a one-dimensional block matching motion estimation algorithm is proposed.Experimental results show that this algorithm significantly improves the search speed of motion estimation while ensuring the quantitative evaluation of the peak signal-to-noise ratio.Patch detection needs to use a lens segmentation algorithm to preprocess the video sequence,and select the reference frame for patch detection based on the lens segmentation result.Through the morphological operation of the difference between the reference frame after motion compensation and the current frame,the mask image of the patch is obtained.Experimental results show that the algorithm can effectively detect the patch area,but it will also misdetect the motion area in the video as patch area.3)Aiming at the large-area misdetection problem in the local motion area in the video,this paper uses the Edge-connect network for damage repair.The network adopts a staged restoration method of repairing the texture information first,and then the color information.Make a data set to train the color repair network.Experimental results show that the algorithm can effectively repair scratches and plaque damaged areas,and it has a good repair effect on large areas of misdetection.This paper combines the above-mentioned scratch and plaque detection and repair algorithms to design a complete scratch and plaque detection and repair system.Experimental results show that the system has a good detection and repair effect on scratches and plaques,and has a high degree of automation.