Effective pictorial information capacity as an image quality metric

Abstract

The measurement of the MTF of JPEG 6b and similar compression systems has remained challenging due to their nonlinear and non-stationary nature. Previous work has shown that it is possible to estimate the effective MTF of the system by calculating an 'average' MTF using a noise based technique. This measurement essentially provides an approximation of the linear portion of the MTF and has been argued as being representative of the global pictorial effect of the compression system. This paper presents work that calculates an effective line spread function for the compression system by utilizing the derived MTFs for JPEG 6b. These LSFs are then combined with estimates of the noise in the compression system to yield an estimate for the Effective Pictorial Information Capacity (EPIC) of the system. Further modifications are made to the calculations to allow for the size and viewing distances of the images to yield a simple image quality metric. The quality metric is compared with previous data generated by Ford using Bartens' Square Root Integral with Noise and Jacobson and Topfers' Perceived Information Capacity. The metric is further tested against subjective results, derived using categorical scaling methods, for a number of scenes that are subjected to various amounts of photographic-type distortion. Despite its simplicity, EPIC is shown to provide correlation with results from subjective experimentation. Further improvements are also considered.