We have presented a perception based image metric which can be used whenever two images should be compared. Human vision is taken into account using the contrast sensitivity function given by Mannos and Sakrison [MaSa74] as well as the CIE LUV color space which is almost perceptually uniform. The results are similar to those reported by Gaddipati et al. [GaMY97]. We find our metric more intuitive as it is computed in the original image space. The newly proposed approach overcomes some wavelet's drawbacks such as the fixed position of certain frequency levels, as well. There is also no need to transform the image in order to compute the metric and the color is treated more accurately due to the use of the CIE LUV color difference formula. The new method can be used for steering the rendering process when progressive methods are used, for evaluating sufficiency of appropriate levels in various LOD algorithms, or for an image query from large data bases. It can also show us if some differences will be visible from a certain distance range or not.
This is certainly not the ultimate metric, but rather one more solution that works fine, and can be easily understood. We will try to take into account some additional human vision properties in the future.