Research On Non-Reference Image Quality Evaluation Method Based On Saliency Region

Image quality evaluation occupies an important position in digital image processing technology,which is divided into subjective image quality evaluation and objective image quality evaluation.Subjective image quality evaluation requires a lot of manpower and can not be embedded in a real-time application system.Therefore,it is necessary to study objective image quality evaluation methods that can be applied to a variety of occasions.Objective image quality evaluation methods are divided into three types: full reference,partial reference and no reference.In most cases,the reference image is diffrcult to obtain,so the non-reference image quality evaluation method has a wider range of application scenarios.BRISQUE(Blindl Referenceless Image Spatial Quality Evaluator)has the advantages of lower technical complexity and faster implementation process.However,because this method ignores the characteristics of the human visual system when evaluating image quality,there is a certain deviation between the result and the subjective evaluation value.Therefore,this paper proposes to integrate the characteristics of visual attention mechanism into the BRISQUE algorithm,and conducts the following research:(1)Because the BRISQUE algorithm does not consider the impact of the salient area of the image on the overall image quality,and the human eye pays more attention to the content in the salient area of the image when viewing the image.The degradation of these areas will have a greater impact on the overall image quality.Therefore,this paper proposes a non-reference image quality evaluation method based on salient regions.This method combines the saliency region extraction algorithm GLGOV(Gestalt Laws Guided Optimization and Visual Attention)to partition the image,extract the natural scene statistical features of the image from the saliency region and the non-saliency region,and perform weighted fusion.The image quality evaluation model is established by training SVM with the fused features.Experiments were performed on the LIVE and CSIQ image databases respectively,and the results show that the method proposed in this paper has better evaluation performance.(2)Aiming at the problem of extracting only the statistical features of the natural scene of the image in the BRISQUE method,this paper proposes an improved non-reference image quality evaluation method based on the perceptual features and the statistical features of the natural scene.Considering that the perceptual characteristics can better reflect the relationship between the image quality and the subjective perception of the human eye.Based on the statistical features of natural scenes based on salient regions,image information entropy and image fourth-order moments are added as image perception features to improve algorithm performance.The performance of the method given in this article has been verified in the LIVE and CSIQ image databases.The experimental results show that compared with the traditional non-reference image quality evaluation method,the method in this article is closer to the subjective perception of the human eye and has achieved better results.