Realization Of Degraded Video Image Recognition And Enhancement Based On Neural Network

Scientific and technological progress has promoted the development of society.While improving the people's living standards,it has also put forward higher requirements for security precautions,making the use of surveillance cameras gradually spread throughout human life.At the same time,monitoring video images bring about the phenomemon of degradation due to various adverse factors,which is not conducive to the analysis and use of surveillance images,and even hinders the monitoring network to achieve better security performance.Therefore,this paper proposes and designs a classification and enhancement method for degraded image recognition based on neural network,and mainly carries out the following work.Firstly,a classification model of degraded type recognition was designed for the degraded images generated by surveillance video.The model is based on convolutional neural network design and was trained and optimized on the Tensorflow framework through self-built image datasets.The experimental results show that the model designed in this paper can accurately identify and classify the degradation types of video images.Secondly,two methods were designed to enhance the degraded images that output by the classification model.For the output images of low-contrast degraded,an enhanced algorithm of histogram equalization combined with radial basis function network is designed.The algorithm by using neural network to establish the nonlinear mapping relationship between input image and certain parameters in histogram equalization algorithm achieving low contrast image enhancement.For the fuzzy degraded image output by the recognition and classification model,an image enhancement method that based on generating and confrontation network is designed in this paper.An optimal enhancement model is generated by redesigning the generator and discriminator and redefining the loss function and training using matched blurred/clear image pairs to train.The final model can directly input the blurred image,and then output the enhanced image can be output,realizing automatic enhancement.Finally,This paper subjectively evaluates and objectively analyzes the two enhancement methods.The experimental results show that the two enhancement methods are simple and efficient,and can achieve the intended research goals.