Research On Colorization Algorithm Of Grayscale Image

Image colorization is the process of adding color to grayscale images or video using computer technology,as one of the important research topics in the field of computer vision,it is widely used in many engineering fields such as film,medicine,ancient painting restoration and military.In view of some shortcomings of the existing colorization algorithms,this paper focuses on grayscale images,and the algorithms of coloration based on color marker and coloration based on color transfer are studied respectively.The main research contents are as follows:The optimized colorization algorithm based on color markers is studied.Aiming at the traditional algorithm,when the number of image mark colors is too small or the edge intensity is too low,the problem of color cross-border mis-diffusion occurs,and an improved optimized colorization algorithm is proposed.Linear nearest neighbor propagation algorithm is introduced by the algorithm,and the optimal weight is reconstructed by constructing local linear neighbor matrix,and the initial global color diffusion is achieved according to the process of minimizing the weighted average difference of the chrominance values of the unmarked pixels and their neighboring pixels.At the edge of the image where the intensity is too low,a dedicated edge mark is introduced to further suppress color diffusion.In order to improve the speed of the algorithm,it is proposed to replace the pixels with SLIC superpixels,and then use the bilateral filter to achieve a smooth transition between superpixels.The experimental results show that the algorithm of this paper not only solves the problem of color cross-border mis-diffusion in the case of less marked colors,but also improves the speed of operation.The colorization algorithm based on color transfer is studied.Aiming at the traditional algorithm only global color transfer,without considering the local features of the image,the problem of color mis-transmission occurs,and a colorization algorithm based on Graph-Based image segmentation and region matching is proposed.The target image and the reference image are segmented into a plurality of sub-regions whose luminance and texture features are consistent by using Graph-Based image segmentation.According to an idea presented in this paper,it is considered that “sub-regions with similar brightness and texture have similar color distributions”,a matching relationship is established between the target image and the sub-region of the reference image,and the sub-region is matched by using luminance feature and LBP local texture feature.Color transfer is performed on the matching sub-regions using luminance and texture features in the YCbCr color space.The experimental results show that compared with other algorithms,the algorithm overcomes the problem of color mis-transmission,and the color distribution and texture definition of the image are also improved.