Research On Optimization Algorithm Of Object Segmentation For Blurred Video Data

As a transmission medium,video plays an extremely important role in the rapid development of computer technology and new media.Because video object segmentation can realize the separation of foreground and background,it has been widely used.For example,in video compression,different encoding can be realized according to the separation of foreground and background to improve encoding efficiency.Therefore,video object segmentation related algorithms have gradually become a research hotspot.In real life,the collected video will always be blurred,which seriously affects the application of video object segmentation technology.Therefore,as a preprocessing step of video target segmentation technology,video deblurring technology is also very important.At present,there are still some problems in related fields.For example,video object segmentation will be negatively affected by object occlusion,violent motion,shape and illumination changes,and current video deblurring research has a large amount of parameters,long processing time,and insufficient accuracy.In response to the above problems,the main work of this thesis is as follows:(1)Aiming at video deblurring,this thesis proposes a Haar and Attention Video Deblurring algorithm(HAVD).This algorithm is based on scale-recurrent network and introduces the haar two-dimensional wavelet transform which is used as preprocessing to deblur the video image in the wavelet domain.At the same time,the spatial attention mechanism and the channel attention mechanism are combined into the overall network framework to improve the feature expression ability of the network framework.Finally,the residual Inception structure is introduced to extract the multi-scale features of the video image and the network training time is accelerated through the jump connection module to achieve a better video deblurring effect.The HAVD algorithm has performed simulation comparison experiments on two benchmark data sets and a self-built data set and it performances better in terms of peak signal-to-noise ratio and structural similarity.(2)Aiming at video target segmentation,this thesis proposes an Attention and Morphology Video Object Segmentation optimization algorithm(AMVOS)based on attention mechanism and morphology.The algorithm is based on the network framework of One-Shot Video Object Segmentation(OSVOS)and firstly combines self-attention.The mechanism is to model the key information of the video frame.The mechanism can capture the key area of the video frame,improve the result of video segmentation in the spatial domain,and quickly capture the long-range dependence by calculating the relationship between the pixels.Information is introduced into the network model to improve video segmentation results.Secondly,morphology is used to optimize the results of the algorithm to make the final segmentation feature expression more compact.The AMVOS algorithm has been simulated and compared in three benchmark data sets and it has better performance in terms of regional similarity,contour accuracy and time stability.