Quality Assessment For Image Color Changing

The 21 st century is hailed as an era of information explosion.Everyday,the amount of information human receive increases exponentially.With the development of technology,images occupy an increasingly important position in the process of information transmission.Image quality assessment technology is to evaluate the quality change of image during its acquisition,transmission and processing,which is of great significance and practical value for ensuring the accuracy of information transmission and enhancing the perceptual experience of the human visual system.In the literature,researchers focus on distortion types related to information carrying ability,such as noise,blur and compression loss.Thanks to the hardware and software improvements in image acquisition,transmission,processing and display,as well as the development of image quality assessment techniques in recent two decades,these distortion types have to some extend been solved.Therefore,the hot area of image quality assessment gradually shifts to higher-level criteria such as image saliency,image authenticity,image aesthetics and so on.Among them,image color is a vital important factor on image quality but relatively lack of research.Hence,this paper focus on the issue of color changing image quality assessment.We study relevant theories in-depth and analyze the existing problems in this area.This paper makes a series of innovative work as follows:1,Create a subjective quality assessment database for color change images.In the field of image quality assessment,the research of objective algorithms depends on corresponding subjective quality assessment database.Though investigation,we find that the existing public image databases are not enough to support the research needs.In order to fill this void,this paper creates a new subjective image database,which has reached the leading level in the field of image quantity,image size,distortion type richness and label support.The new database provides a strong foundation for related research.2,Propose a full-reference quality assessment matrix based on color related features.Feature extraction of the classical image quality assessment matrices focuses on images' structural features,however,the structural features are not suitable for the problem of image color change quality assessment.Therefore,this paper extracts series of features related to color,including the basic indexes of color,the feature of color edge,the features of distribution in HSV color space,the entropy features and so on.A gradient boosting decision tree model is used to achieve the quality assessment matrix of the image color change.Experiments show that the model has a good consistency with the subjective evaluation data.Establish a non-reference quality assessment method based on color harmony knowledge learning.No reference quality assessment is the ideal solution for industrial applications.To achieve this,the relationship between colors should be explored.Therefore,this paper introduces the theory of color harmony,which is widely used in art and design.However,after analyzing the existing color harmony models,it is found that there are some defects to involve it the image quality assessment such as the conflicts between different models,unfitness for wide color range,lack of quantitative formula.Hence,this paper follows only the basic idea of color harmony theory and adopts the concept of experiential learning.We extracts the harmony relationship of color from a large number of images through image segmentation and color theme extraction,and establish the image color quality assessment algorithm based on the obtained data.Experiments show that this method is effective and it can be easily optimized with more data.