Uma Análise do uso de informacões multiescala no mapeamento da PSNR para pontuacão perceptual

Abstract

The prediction of visual quality is crucial in image and video systems. For this task, image quality metrics based on the mean squared error prevail in the field, due to their mathematical straightforwardness, even though they do not correlate well with the visual human perception. Latest achievements in the area support that the use of convolutional neural networks (CNN) to assess perceptual visual quality is a clear trend. Results in other applications, like blur detection and de-raining, indicate the combination of information from different scales improves the CNN performance. However, to the best of our knowledge, the best way to embody multi-scale information in visual quality characterization is still an open issue. Thus, in this work, we investigate the influence of using multi-scale information to predict the perceptual image quality. Specifically, we propose a single-stream dense network that estimates a spatially-varying parameter of a logistic function used to map values of a objective visual quality metric to subjective visual quality scores through the reference image. The proposed method achieved a reduction of 36.37% and 69.45% for the number of parameters and floating-point operations per second, respectively, and its performance is compared with a competing state-of-the-art approach by using a public image database.