Research On Automatic Color Adjustment Of Color Images

With the ubiquity of cameras and mobile phones,images have become a popular medium of visual information in daily lives and scientific research.High-quality images can vividly present the scene seen by the human eye and provide robust basics for scientific research.However,due to the lack of photography skills,casual consumers cannot take photos with the best camera settings,and their images are prone to color distortion and degradation,which can degrade the quality of images.How to restore the degraded image and improve its quality by automatic color adjustment is a problem of great significance in research and application.To solve this problem,this study proposes automatic color correction,exposure correction and color enhancement methods to respectively restore color biases,exposure errors and improve the visual quality of an image.The works of this study can effectively restore image color defects and improve the visual quality of an image.The main works of this study are as follows:(1)This study proposes a color correction method based on improved grey pixels to correct the color bias of an image in the imaging stage.This study has extended the concept of grey pixels to the improved grey pixels(IGP),which is a subset of an image and contains pixels located in the same square area with the illuminant in the discretized rg-chromaticity space.In addition,this study has proposed a method to extract IGP for statistic-based methods from an image by a neural network.The exprimental results have shown that the proposed method can significantly improve the performance and robustness of statistic-based methods.(2)As for the color bias of a post-capture image,this study proposes a color correction method based on the estimation of polynomial coefficients.First,this study has simplified the flow of image signal processing and derived the initial linear model for color correction.This study then uses polynomial expansion and a residual term to refine the initial model.This study subsequently proposes a light convolutional model to estimate the transform matrix in the model for color correction.In the comparison of accuracy and efficiency,this method has surpassed existing methods by a large margin.(3)This study proposes an exposure correction method based on the estimation of intensity transformation curves.This study has firstly analyzed the distribution of the brightness component of over-and under-exposed images and concluded that the common problem of these images is the shift and centralization of brightness.Based on this conclusion,this study proposes a model composed of γ transformation and logistic function to compensate image brightness and enhance image contrast.This study then proposes a deep model composed of a γ estimation module and a logistic estimation module to generate the image-specific correction curve.The experimental results show that this method can handle both over-and under-exposed images with a unified model and has achieved the best performance.(4)This study proposes a Transformer-based color enhancement method to improve the visual quality of an image.The proposed method utilizes a sequence of Transformer encoders to model the global dependency between local areas in an image and construct the feature representation of these non-overlapping areas.Then,it uses a fully connected network to estimate several local color transform matrices and uses the weighted sum of these matrices as a global matrix,which is applied on the polynomially expanded image to enhance its color.In quantitative comparison,the performance of the proposed method is better than existing methods.In summary,the works of this study have made some progress on the methods for automatic color adjustment of color images.In the future,the works of this study can be applied to the following three scenarios:(1)The works can be merged into the image signal processing flow to improve the quality of images;(2)Developing a software to provide consumers with convenient image quality enhancement tools;(3)Based on the works of this study,designing and developing a prepress image processing software for personalized printing to improve the efficiency of printing enterprises.