In order to achieve image quality which is consistent with human eyes perception

a full reference image quality evaluation method based on visual saliency model is proposed. The proposed saliency model is divided into two layers: the first layer is that the gradient perception map extracted by just noticeable difference（ JND） model

the second layer is that the weight perception map extracted by superpixel segmentation algorithm.On the basis of fusion of two layers perception maps

the difference of feature is measured between perceptual maps of reference image and distorted image. Support vector regression（ SVR） is used to obtain objective quality fraction. The experimental results indicate that objective assessment of the proposed method is better than mainstream algorithm in image quality

and it is consistent with human eyes subjective assessment for different types of distortion and different databases.