| Color is an important component of human visual perception.With the development of computer technology and hardware,color digital images have been widely disseminated.As an important technical means to assist humans in appreciating and interpreting image information,image color transfer technology has gradually become a research hot topic in the field of computer vision,and has been widely used in film and television,medicine,remote sensing and cultural relic restoration.Based on a comprehensive analysis of existing methods,this study aims at the problems of loss of texture details,color mistransmission and lack of layering,and neutral color distortion that commonly exist in traditional color transfer methods.Three stages of method research were carried out,and the research contents are as follows:(1)A color transfer framework based on image eigen decomposition is constructed,and a local color transfer method based on image eigen colors is proposed.The method improves the accuracy of eigen color segmentation and local color transfer while maintaining the image detail texture,and enhances the color hierarchy of the color-transferred image.(2)Aiming at the problems of high computational cost of five-dimensional clustering and the destruction of image structure by traditional clustering and segmentation methods,a drift clustering method based on 3D grid is proposed.The iterative operation speed,and the residual pixels of the two-dimensional plane are eliminated through connectivity enhancement,maintaining the structural integrity of image segmentation.Aiming at the problem that image style preservation and local color rationality cannot be comprehensively considered in eigen color matching,a balance factor for chromatic aberration and specific gravity is introduced to match eigen colors according to the similarity of eigen colors of color and spatial contribution,which improves the accuracy and rationality of local color transfer.(3)Aiming at the problem of neutral color loss caused by color transfer,the saturation weight map is designed by using the source image saturation information,and a neutral color enhancement method based on saturation weighting is proposed,which realizes the neutral color enhancement of the transmission image weighting process.Neutral color correction and tone correction to ensure the consistency of transferred tones solve the problems of neutral color loss,color cast and oversaturation in color migration at extremely low computational cost,and improve the visual quality of images.(4)Qualitative and quantitative experiments prove the effectiveness of the improved and designed image eigen decomposition color transfer framework,eigen color cluster segmentation,and neutral color enhancement methods.Subjective and objective evaluation experiments show that the method can effectively preserve the details and eigen colors of color-transferred images. |
