Stack objective image quality assessment:A practical evaluation method in super resolution task
Subjective assessments provide reliable evaluations that are established through significant amounts of manpower and material resources.Objective assessments based on mathematical models can be easily implemented,but they fail to capture the nuances of human judgement and therefore lack consistency with subjective assessments.Stack Objective Image Quality As-sessment(SOIQA)is a practical objective assessment,which combines existing classic image quality assessments based on deep learning to predict image quality close to subjective evaluation.SOIQA is applied to the task of image super-resolution(SR)and utilized PSNR,SSIM,VMAF,and NIQE assessments in SOIQA,as they are commonly used in SR tasks.To stack these assess-ments effectively,a model is designed encouraged by UNet model.Experimental results show that SOIQA can achieve better image quality evaluation performance with accuracy of 12%higher than the existing objective assessments in MOS consistency.
万方数据
