Study On Objective Quality Assessment Of Retouched Images

High-quality images play an important role in every fields of life,such as medical community,films and television circles,industry,etc.However,due to the complexity of the external environment,present image acquisition system still can not obtain high-definition images in accord with the requirement of people.This problem promotes the development of a lot of algorithms and software for retouching images.Image retouch is to obtain images with better perceptual feeling by appropriately adjusting several features of the images obtained by the image acquisition equipment,namely improving some factors relative to the image quality.At present,image retouch has been widely used in every aspects of the image processing field,so it is fundamental to evaluate the quality of retouched images.In this work,based on traditional image quality evaluation metrics,two quality assessment methods for retouched images are proposed by adopting the color similarity and structure similarity.Among them,the chroma similarity is used to represent the color similarity and the gradient similarity is for the structure similarity.The specific content of this study is shown in the following:Based on the fact that,the retouched images have some improvement in chroma compared with reference images,the chroma is proved to have a good correction effect on traditional images,so a quality evaluation method for retouched images is proposed by combining the traditional metrics and the chroma.At first,the chroma maps of both reference and retouched images are extracted and the similarity of two chromas is computed.Then the chroma similarity model is integrated into traditional quality evaluation methods by linear weighted way to obtain the final objective quality score of the retouched image.The experiments in retouched image database demonstrate that chroma indeed has a good correction effect on traditional image quality methods when they are used to evaluate the quality of retouched images.A quality assessment method for retouched images is proposed by combining chroma similarity and gradient similarity based on the improvement of retouched images in chroma and structure,namely gradient.The chroma and gradient maps of both reference and retouched images are first extracted and the similarity features are measured.Then these two features are used as the input and the predicting model for retouched images is obtained using the Support Vector Regression(SVR)model.At last,the trained model is adopted to predict the quality scores of retouched images.The experimental results show that the predicted scores have good consistency with subjective quality scores.What is more,the trained model has better performance for the quality assessment of retouched images than traditional image quality metrics and the present quality methods designed for retouched images.