The current trend of using ROC style observer performance studies to evaluate image compression schemes is inefficient and has greatly limited our ability to certify particular techniques for use on radiographic images. The goal of this project is to develop more efficient automated techniques for evaluating the efficacy of image compression, but which are independent of any particular compression method. Two approaches will be pursued. First, we will develop fidelity measures which can be applied to the difference between an original and compressed version of an image. These will be based on not only the magnitude of the difference image but also on the structure in this image, and should overcome the limitations of simpler measures such as mean square error. We will base the design of these measures on plausible hypotheses about what factors are important rather than on psycho-visual theory directly. Although these measures cannot be expected to be highly correlated to observer performance, for small levels of image degradation they should be useful for placing limits on the expected loss of diagnostic performance. Secondly, a digital phantom will be designed which can be used to evaluate any compression technique and compression ratio. This phantom would be evaluated with the above measures. Use of a phantom would be more efficient since it could be done once for each compression technique and would avoid the necessity of evaluating a measure for each image. Finally, we will attempt to apply both methods to analyze the databases of images currently being used in our observer performance studies on image compression and on image resolution. As part of this we will correlate the physical measures being developed to ROC results from these studies.
