Research And Application Of Grayscale Image Colorization Algorithm

Images with rich colors can not only clearly express the characteristics of a thing,but also facilitate information exchange.The color of grayscale images is not only a popular research topic,but also has a certain challenge and practical significance,and its research results have been widely used in all walks of life.In order to solve the problems of long colorization time and color error diffusion in current colorization algorithm,this thesis uses deep learning method to study image colorization,and the specific work and achievements are as follows:(1)The method of image rapid color transformation based on neural network and Reinhard algorithm is studied.Firstly,LBP eigenvalue matrix is used to replace the feature extraction layer of neural network to accelerate the training and image generation speed of neural network.Then,the image gray value matrix and LBP eigenvalue matrix are matched point-to-point,so the classification problem of image matrix is transformed into the classification of two-dimensional coordinate points,which gets rid of the dependence of model on training data set and reduces the model complexity.The advantage of the characteristics of target gray level image details,the most clear,the luminance component after the HSV space transformation and grayscale image fusion itself,so as to find the lost in the process of neural network in image generated texture information,not only effectively removed noise,at the same time to enhance the image resolution,image quality was optimized.Finally,the fast color transfer speed of Reinhard algorithm is utilized to further shorten the running time of the algorithm.Through comparison,it is found that although the proposed colorization method has a small amount of color error diffusion phenomenon,compared with other colorization algorithms,it does not need a lot of training data,and the colorization speed is faster.It not only conforms to human visual habits in effect,but also has strong versatility.(2)In this thesis,the method of color rendering of gray image based on attention mechanism and Welsh algorithm is studied.First,in order to eliminate the image background is invalid area influence on color effect,this thesis puts forward a kind of color of the thought of "divide and conquer",by using the method of attention mechanism and feature fusion for reference color images in the cutout,isolated prospect area effectively,not only reduces the algorithms in the image through participation in the process of computing the number of pixels,And effectively overcome the phenomenon of color mistransmission.Then,Retinex algorithm was used to enhance the image of the separated effective foreground region to enhance the clarity of the dark region of the image and improve the quality of the reference image.Next,the global histogram equalization process is performed on the target gray image to make the brightness value distribution of the gray image more uniform,enhance the contrast of the target image,highlight the local texture structure information of the image,and reduce the color error diffusion.Finally,the improved Welsh algorithm is used for color transfer,which can effectively shorten the running time of the algorithm and achieve the final ideal color effect.The experimental results show that the proposed colorization method not only runs fast,but also effectively overcomes the phenomenon of color mistransmission,and the generated color image is of high quality.The idea of colorization is instructive to the subsequent research of colorization algorithm.