Research On Color Correction Methods Of Digital Image

Human visual system has excellent color constancy function,which can eliminate or reduce the influence of illumination change on the color of object surface in the scene and then can help people to obtain the real color property information of object surface.However,unfortunately,imaging devices,such as digital cameras and smart phones etc.,do not have this smart color adjustment ability,which can result in the phenomenon of color cast in the acquired digital images.Therefore,the research of color correction for digital images receives extensive attentions in related research fields and has important significance.In this thesis,unsupervised and supervised color correction methods of digital images are studied,using color information in the standard color checker as the training sample of supervised learning to carry out color correction of color cast images.What's more,the experimental results are evaluated from subjective evaluation and objective evaluation.The main research contents of this thesis are roughly as follows:Firstly,when the scene is lack of color information,grey world algorithm usually has problems of over-correction and poor performance.In this thesis,in order to solve the above-mentioned problems,an unsupervised color correction algorithm of digital image based on standard deviation weighting and image entropy constraint is designed and implemented on the basis of the classical grey world algorithm.Experimental results show that this method has better color correction effect and performance,compared with the traditional grey world color correction algorithm.Secondly,the color mapping model between an unknown illumination and a standard illumination is usually obtained by means of the color information in the standard color checker.At present,human-computer interaction is usually used to identify the color checker in the scene and extract color information,but which takes a lot of time and energy.Therefore,considering the advantages of SIFT algorithm,a method of standard color checker automatic recognition and color information extraction is designed and implemented.Experimental results show that this method can accurately and effectively identify the color checker in the scene and extract the color information.Thirdly,in supervised digital image color correction methods based on neural network,the back-propagation neural network(BP)algorithm is slow in training speed and easy to fall into local optimization.Moreover,the selection of kernel function and related parameters of Support Vector Regression(SVR)algorithm is very complicated,which can only be adjusted by experience and continuous attempts,and the learning rate is relatively slow.In view of theabove-mentioned shortcomings,Extreme Learning Machine(ELM)algorithm has the advantages of simple parameters,fast speed,multiple outputs and good generalization ability,etc.Therefore,ELM algorithm is introduced into the color correction of digital images by means of using the color information of standard color checker in the scene as the training data.Experimental results show that the supervised digital image color correction method based on ELM is not only fast in calculation,but also better in performance and effect than that based on BP and SVR.