Structural Similarity As a Prediction Metric in Lossy Image Set Compression

An automatic compression strategy proposed by Gergel et al. is a near-optimal lossy compression scheme for a given collection of similar images whose inter-image relationships are unknown. That algorithm uses the root mean square error (RMSE) as a measure of the similarity between two images. Since RMSE is a metric, provable guarantees on the quality of the decompressed images can be made. However, it is well known that the RMSE does not correspond well to the human visual system. Recently, Brunet et al. introduced a metric based on structural similarity (SSIM). In this work, we show that the application of a SSIM-based metric instead of RMSE in the lossy image set compression scheme give improvements on some types of image sets.