Study On Objective Quality Evaluation Method Of Gamut Mapped Images

With the increasing popularity and wide application of multimedia devices,the propagation and sharing of color images between different devices will be inevitable.In order to ensure the quality of the reproduced images,a large number of gamut mapping algorithms have been reported.Gamut mapping is a method of mapping color coordinates in a source image or device to a target device or image.The primary purpose is to reduce the difference of color reproduction between the source and target gamuts,thereby achieving high-fidelity color reproduction.Due to the color mechanisms of different devices are different,it is difficult for a single color gamut mapping algorithm to produce high-fidelity mapping between any two gamuts.Therefore,an accompanying question is how to evaluate the performance of gamut mapping algorithms and the quality of color gamut mapped images.This thesis studied the quality evaluation of color gamut mapped images,analyzed the distortion characteristics and distortion principles of gamut-mapped images,based on which we proposed two effective no-reference quality evaluation methods for gamut mapped images.The main research contents are as follows:Through experimental analysis,it is found that the distortion can cause the change of the statistical distributions of the wavelet coefficients and gradient coefficients;at the same time,the gamut mapping cause not only the grayscale distortion but also color distortion.Based on this,we proposed a blind quality index for gamut mapped images based on Natural Scene Statistics(NSS).Firstly,the images are converted to the S-CIELAB color space to extract the three color attributes(e.g.,brightness,chroma,and hue).Then,for each attribute,NSS features are extracted in the wavelet and gradient domains,respectively.Subsequently,the extracted features and the subjective labels of the training images are fed to a Back Propagation Neural Network(BPNN)for training the quality prediction model.Finally,the model is used to predict the quality of gamut mapped images.Extensive experiments conducted on three gamut mapping databases prove the effectiveness of our metric in evaluating the quality of gamut-mapped images.By analyzing the mapping principles of different gamut mapping algorithms,we find that there are mainly color and structural distortions in gamut mapped images.Based on this,we propose a no-reference quality evaluation method for gamut mapped images based on color and structural distortions.For color distortion,we calculate the rate of abnormal hue and the Kullback-Leibler Divergence between the statistical distribution of the three components of the image(e.g.,R,G,and B)and the ideal uniform distribution.In terms of structural distortion,the entropy and the fourth-order moments are extracted,and statistical features are extracted in brightness and saturation components.Then,combined with the subjective scores and extracted features,the BPNN is used to train the quality prediction model.Finally,the model is employed to evaluate the quality of gamut mapped images.The experimental results prove that the proposed method is superior to the existing quality evaluation models in evaluating the quality of gamut mapped images.The paper has 23 charts,12 tables and 93 references.