Multi-focus Image Fusion Technology And Its Application In Traffic Image Processing

In today’s society,the transportation industry has made more and more contribution to the economic development of our country.The traffic image based on computer vision is widely used in traffic video,and the corresponding traffic image processing and analysis has become the focus of researchers.However,obtaining traffic images only through a single sensor is far from meeting the development of intelligent transportation systems.The current intelligent traffic image monitoring system generally includes two aspects of traffic image monitoring and traffic signal control.However,due to technical limitations,traffic image signals often have the drawback of small video field of view and low image resolution.In this regard,this paper proposes to divide the traffic image signal into several multi-focus images,and use the advantages of image quality at each focus to be significantly higher than other places.The method can be applied to traffic image monitoring of toll stations,multi-lane intersections and overpasses,which require large video fields and high image resolution.In addition,the same type of imaging equipment under different imaging conditions,the resulting traffic images usually have large differences;and the same type of imaging equipment under a single imaging condition,the traffic images collected are difficult to reflect the scenes people need information.Therefore,in order to achieve the application purpose of obtaining multiple image detail information on one image,improve the utilization of image information,and make intelligent monitoring possible,research and solve multiple multi-focus traffic images under different imaging mechanisms or different imaging conditions.The issue of integration has become an important research topic.The key to multi-focus traffic image fusion is to accurately locate the clear regions in the source image.However,the multi-focus image fusion method cannot accurately locate the clear regions in the source image.In order to solve the current multi-focus traffic image fusion method,the clear region in the source image cannot be accurately positioned,resulting in the problem that the fused image is not ideal.This article focuses on the key technologies such as multi-focus image fusion and how to effectively extract clear regions in the source image.Based on the idea of decision graph in image fusion,this paper establishes a direct mapping between the source input image and the decision graph,analyzes the discrete regions in the decision graph,uses mathematical morphology methods to optimize the decision graph,and strives to obtain the best fusion image.The main research results and innovations of this article are as follows:(1)A multi-focus image fusion algorithm based on image segmentation is designed.This method treats the decision-making graph as the segmentation task of the clear region and non-clear region in the source image.The designed method uses a multi-scale convolutional neural network to implement segmentation of the source image and performs multi-scale analysis of each input image.At each scale,feature maps of clear regions and non-clear regions are obtained,and multiple feature maps are merged to generate an initial decision map.For the problem of clear discontinuities in the initial decision graphs,initial segmentation and mathematical morphology are used to post-process the fused decision-diagrams.Experimental results show that the proposed method can obtain higher contrast and sharper image fusion results.(2)Aiming at the defect that the existing multi focus image fusion methods can not use the activity level measurement and fusion rules uniformly,a fully convolutional neural network is designed to obtain a high-quality decision graph.In the designed total convolution neural network,the maximum pooling layer in CNN structure is replaced by a convolution layer,which makes up for the shortcomings of the existing convolutional neural network pooling layer that it is easy to lose the original information of the image,and improves the accuracy of the fusion decision map Sex.The experimental results show that,compared with the existing advanced image fusion methods,the decision graph obtained by using the fully convolutional neural network is highly reliable,and the image clarity and fusion effect are better.(3)In order to solve the problem that a single-scale dictionary cannot capture the different structural information of the input image well,a multi-scale adaptive sparse representation(MASR)traffic image fusion method is designed.In this method,two important factors,fusion rules and activity level measurement,are considered.Using the proposed multi-scale adaptive sparse representation method,a direct mapping between the source image and the decision graph can be obtained.In the designed multi-scale adaptive sparse representation model,according to the different texture features of the input source image,high-quality image blocks are used to learn a set of compact multi-scale sub-dictionaries.In the fusion stage,different sub-dictionaries are selected according to the texture characteristics of the sub-blocks.Experimental results show that the proposed MASR method can obtain higher contrast and sharper image fusion results.(4)Aiming at the shortcoming of impulse-coupled neural network in judging part of the focus area incorrectly,a pulse-coupled neural network and a multi-focus image fusion algorithm with improved decision diagrams are designed.The designed method uses a pulse-coupled neural network stimulated by directional information to judge the input image and obtain an initial decision diagram;for the problem of discontinuity in the focus area of the initial decision diagram,the initial decision diagram is optimized by using advanced post-processing technology.Get the modified decision diagram;get the final fusion image according to the decision diagram.The experimental results show that the proposed OI-PCNN method obtains better clarity and fusion effect of fused images.(5)Aiming at the shortcomings of losing the input image information during the inverse transformation of the weight gradient fusion method,a semi-weight gradient and self-similarity multi-focus image fusion method is proposed.The semi-weighted gradient method designed can obtain a direct mapping between the focus area and the input image,that is,the initial decision diagram;in this method,the self-similarity of the image is used to adjust the decision diagram,and the adjusted decision The image is optimized to obtain a high-quality fusion image;the designed method takes advantage of the multi-scale weighted gradient,and overcomes the shortcoming of losing information during inverse transformation.The experimental results show that the proposed HWGSS method obtains better clarity and fusion effect of fused images.The rich research work and fruitful research results in this paper can not only provide effective theoretical and technical support for multi focus image fusion,but also provide reference and reference for the fusion of traffic image and infrared image.The fusion method proposed in this paper can effectively improve the utilization rate of image information in practice,expand the working range of imaging system in traffic image and other fields,the usability of the system has also been effectively enhanced,and the description of effective target information in different scenes is more accurate.