Research On Key Technology Of Removing Image Obstacles Based On Motion Field

Removing image obstacle refers to a technique of removing an obstacle layer from a mixed image and obtaining a target background layer image.According to the transparency of the media parameters,imaging obstacle can be divided into two categories: transparent and opaque.Transparent imaging obstacle includes glass reflection,impurities on the glass,etc.The opacity imaging obstacle refers to the presence of occlusion objects before the target scene,such as fences of various shapes.The presence of interfering objects makes the captured images blurred,obscured,lost some useful information and other issues,The removal of the obstacle components can enhance the recognition of the image and provide more information on the application of image reconstruction,recognition and tracking.In view of imaging obstacle removal,the current research methods mainly have the following problems: First,many methods require manual interaction,so the automation is poor;Second,the high-resolution image processing takes a long time;Third,the transparent imaging and opacity imaging obstacle as different algorithms,most of the methods focused on a specific type of obstacles to remove;In this paper,we study the technique of image obstacle removing,it based on sequence image.We complete the construction of two kinds of imaging obstacle removal algorithms,which are transparent reflection and opaque occlusion.We complete the framework of imaging obstacle removal system based on motion field.The main contents of this paper are as follows:(1)The implementation of removing image obstacle model is described in detail.The estimation of sparse motion field includes sequence image acquisition,preprocessing,edge extraction and the model of sequence image sparse motion field.Our method is based on the removal technique of sequence image.In this paper,we use the normalized cross correlation method based on the edge pixel to estimate the motion vector between the sequence images,and achieve the removing of the obstacle layer by the complementary information between the images,and comply removing automation.In the pretreatment stage,the lower sampling process is added,which effectively reduces the time of calculating the edge motion field between the subsequent images,and solves the problem of high resolution image processing time.(2)This paper completes the initial removal of the barrier based on the playground and the precise reconstruction of the target background.Firstly,the RANSAC algorithm is used to fit the sparse motion field of the sequence image.It can achieve the separation of sparse motion field between target background layer and the barrier layer.Secondly,this paper proposes to use BP neural network algorithm to interpolate sparse motion field.Experimental results show that the neural network to achieve the surface interpolation speed and the effect is better.After image registration,we achieved initial removal of the two imaging obstacles.Finally,in the optimization of the iterative part,we select the initial separation of the target layer and the obstacle layer as input,increase the constraints,and to achieve accurate removal of imaging obstacles.(3)In this paper,the similarities and differences between the transparent reflection and opaque fence occlusion are analyzed in detail.We combine the removal of two imaging obstacles into the same theoretical system.On the basis of the same main frame,the specific implementation steps of the algorithm are different for the characteristics of two different media.In this paper,the experimental part of each module is analyzed by MATLAB.The experimental results are qualitatively and quantitatively evaluated.Finally,the prototype of the system based on the motion field is developed on the PC side.