Research On Image Enhancement Algorithm Based On Color Transfer

Color transfer technology is an image enhancement method that has been widely applied and developed in various fields,such as artistic creation,visual effect enhancement,aerospace,and medical image processing.The goal is to apply the color characteristics of a reference image to the source image without changing its structure and texture information,achieving more visually pleasing effects.However,existing color transfer methods face challenges such as exposure and saturation issues in generated images,color inconsistencies due to insufficient semantic resolution capabilities,and difficulty in obtaining paired data in deep learning methods.To address these challenges,this paper proposes an efficient and accurate color transfer method specifically designed for portrait photos.The method fully considers the characteristics and requirements of foreground and background when processing portrait photos,independently processes the separated foreground and background layers,and optimizes their color distribution and lighting characteristics.At the same time,by using an unsupervised generative adversarial network(GAN)framework,it effectively solves the problem of obtaining paired data in existing methods.This method covers three key parts: preprocessing,foreground image enhancement,and background image enhancement.In the preprocessing stage,the portrait image’s foreground and background are separated,and face alignment technology is used to focus on the image center,laying the foundation for subsequent processing.The foreground image enhancement part includes color generation and color propagation modules,responsible for performing fine color transfer operations on the face image and applying the transfer strategy to the entire foreground image,achieving natural color effects.The background image enhancement part adopts a color transfer processing method based on statistical data and lighting optimization techniques,aiming to comprehensively adjust the background image’s color distribution and lighting conditions.This thesis conducts extensive experiments to evaluate the performance of this method and compares it with several widely recognized color transfer methods.These comparisons include an analysis of image visual quality and various quantitative performance evaluation metrics,such as peak signal-to-noise ratio,structural similarity index,and deep neural network-based image quality evaluation indicators.Experimental results show that this method has a significant performance advantage in improving the visual quality of portrait photos,bringing a new visual experience to portrait photo processing.At the same time,in terms of quantitative indicators,this method demonstrates superior performance in multiple evaluation metrics.Finally,this study conducted validation experiments on two types of aerospace image datasets.By adjusting the preprocessing methods and fine-tuning the training,the color transfer function was successfully applied to these datasets,further verifying the versatility of the method proposed in this thesis.