Research On Blur Image Enhancement Based On Deep Learning

With the rapid development of science and technology,today’s society has ushered in a new era of intelligence.As the carrier of information,images are inseparable from the analysis and processing of images in the construction of smart cities.High-quality images play an important role in the intelligent era.However,in the process of image acquisition,the image will be affected by adverse factors in various situations,resulting in image blurring.Therefore,researching fuzzy image enhancement has very high practical significance.In this paper,based on deep learning technology,the fuzzy image enhancement method is deeply researched and implemented.The following is the main work completed in this paper:1)A single raindrop fuzzy image enhancement model based on attention mechanism is proposed for rainy and fuzzy images collected in rainy environment.By extracting the raindrop blurred image by multi-scale expansion convolution with different expansion factors,the attention mechanism model is constructed to pay attention to the raindrop region in the image,so as to better remove the raindrop and remove the image blur.Experiments show that compared with other methods,the model can better remove raindrops and blurs from raindrop blurred images,and more details of the images are restored.2)A motion blurred image enhancement model based on the generated confrontation network is proposed for the blurred image caused by object motion or camera shake.The feature is extracted from the motion blurred image by using a neural network with multi-scale convolution kernel.Secondly,the local residual learning and global residual learning are combined,and multi-path multi-weight sharing recursive learning is used.Finally,the discriminant network and the generation network are utilized.The confrontation training optimizes network parameters.The experimental results show that when the model is applied to motion blurred images in road traffic scenes,the model can achieve image enhancement and has better image visual effects.3)Design and develop a single image of raindrop image enhancement model based on attention mechanism and a motion blur image enhancement model based on generational confrontation network.The system is built with VS2015 integrated development environment,using Open CV The image processing technology realizes the processing of the image and the call of the trained model.The Qt framework is used to realize the system interface design,and the overall structural framework and functions of the system are introduced in detail.