Applications Of Optimization Algorithm In Image Processing

With the development of technology,the application of computer vision technology is becoming more and more widespread.As a direct and simple information carrier,image is favored by many scholars for its intuitiveness and convenience in providing information.However,since most of the devices need to work outdoors,they are inevitably affected by bad weather,resulting in blurred images,overwritten information,and serious quality degradation.In order to fully obtain the image information and improve the image quality,it is necessary to perform image processing on the acquired degraded images.Therefore,in this paper,we propose a new model for each of the three elements we study: image defogging,text image blind deblurring and image rain removal.The contents and innovations of this paper are as follows.For image defogging problem,to solve the problem that the parameters in Convolutional Neural Network(CNN)are not generalizable,we design a trainable end-to-end image defogging algorithm based on convolutional neural network and random walk theory by fusing deep learning and intelligent algorithms.The algorithm first calculates the image atmospheric transmittance using a trainable end-to-end image defogging algorithm based on convolutional neural network(Dehaze Net),and then clusters the atmospheric transmittance using K-means algorithm to make the atmospheric transmittance more uniformly distributed in a certain range.The difference between the mean square error function and the distortion function is used as the objective function to optimize the atmospheric transmittance,and the clustered atmospheric transmittance is used as the initial value of the objective function to solve the optimal atmospheric transmittance using the random walk algorithm,and finally a clear and fog-free image is recovered.For the problem of blind deblurring of text images.In this chapter,a blind deblurring model for text images combining sparse prior and multiscale fusion strategies is designed.The model is solved using semi-quadratic splitting and ADMM algorithm.In addition to considering the sparse gradient prior on the potential clear text images,we also add a sparse prior on the high-frequency wavelet coefficients of the potential clear text images.Also,we consider the effect of the brightness features of the recovered blur kernel on the recovered clear text images.The blur kernel luminance information is boosted using a multiscale fusion algorithm.For the single image de-rain problem.In this paper,we propose a single-image rain removal model combining image decomposition and wavelet transform,which is solved by using ADMM algorithm.Firstly,the first image decomposition is performed using bilateral filtering method.Then the high-frequency image obtained from the first image decomposition is used as the target image to perform the second image decomposition to remove the rain traces contained in the high-frequency image,while incorporating the wavelet transform method to fully consider the sparsity of the high-frequency wavelet coefficients in the clear image under the wavelet transform.Finally,the clear image is recovered by using the low-frequency image obtained by bilateral filtering decomposition and the image after removing the rain marks using image decomposition.