Impact of Pooling Methods on Image Quality Metrics

Abstract

Image quality assessment using objective metrics is becoming more widespread, and an impressive number of image quality metrics have been proposed in the literature. An aspect that has received little attention compared to the design of these metrics is pooling. In pooling the quality values, usually from every pixel is reduced to fewer numbers, usually a single value, that represents overall quality. In this paper we investigate the impact of different pooling techniques on the performance of image quality metrics. We have tested different pooling methods with the SSIM and S-CIELAB image quality metrics in the CID:IQ database, and found that the pooling technique has a significant impact on their performance.