Evolved Multiresolution Analysis Transforms Improve Lossy Compression of Mars Images

Frank W. Moore and Brendan J. Babb


Image Compression, Wavelets, Evolution Strategies


This paper describes research that uses the CMA-ES evolution strategy to optimize multiresolution analysis (MRA) transforms that outperform wavelets for the lossy compression of images transmitted from Mars Exploration Rovers. Our approach first evolves wavelet and scaling numbers for a k-1-level transform that reduces both the information entropy (IE) of compressed images and the mean squared error (MSE) in their reconstruction. These numbers are then used to seed the first k-1 levels of a k-level MRA transform whose level-k wavelet and scaling numbers may then be evolved separately. This approach reduces the dimensionality of the search space, allowing us to optimize MRA transforms whose performance advantages over wavelets increase at higher levels of decomposition. Only after producing a highly optimized level-k transform was CMA-ES allowed to simultaneously evolve multiple MRA levels to achieve even greater MSE and/or IE reduction.

Important Links:

Go Back