Characterization Of Digital Cameras Based On Luminance Division

With the development of color science and image processing technology,the digital camera has already become an important device in the process of digital image acquisition.What's more,it is an important tool for common users to obtain images.However,due to the default settings of the digital camera,the changes of lighting conditions and data transmission paths may result in color image reproduction different from what you perceived.In response to this problem,a solution for characterizing digital cameras has been proposed.Aiming at the requirements of digital camera in the application of consistency and stability in color reproduction process,this thesis proposes a characterization model based on brightness and conducts a comprehensive and basic exploration.Firstly,the meaning and necessity of digital camera characterization are clarified.Then the theoretical basis of colorimetry and the basic components and principles of digital cameras are studied.The commonly used algorithms for digital camera characterization at home and abroad are studied.Several characterizations are carefully analyzed.When verifying the basic characterization model,we found that the color samples with low brightness would have a large deviation in the process of color reproduction,so we divided training set into high-brightness set and low-brightness set according to the brightness.These two sets were separately characterized.In this experiment,we selected two digital cameras(Canon 60 D and Nikon D610)to take images of samples.24-color standard color card and 140-color standard color card were used as research objects.RAW images and JPG images were selected for research.After the previous hardware selection and experimental platform construction,we designed and verified a new model for digital camera characterization.Programming was used to achieve conversion from device-dependent RGB color space to device-independent XYZ color space.We used the polynomial model and the root polynomial model to transform the two sets when solving the transformation matrix.After many experiments and verification of multiple sets of data,the two sets of the best and most stable model combinations were found.For the low-brightness set the polynomial first-order model was selected,and for the high-brightness set the root polynomial second-order model was selected.As the project progressed,we also tried to restore images from different light sources to the same standard light source.This thesis compared the results with recent articles published by Amiri and Fairchild.The nonlinear regression model they proposed is based on the improvementof the polynomial regression model.The weighting function is used to improve the accuracy of the model,but it also increases the computational time cost.The color difference evaluation of the two characterization models on the test samples further highlights the advantages of characterization of the model we proposed divided by brightness.The model is simple,and the precision of the prediction is high.The research results of this project also provide a new idea for the development of the future digital camera characterization model,which has certain reference significance.