380 likes | 482 Views
Explore evolving wavelet-like transforms for improved image compression and reconstruction under quantization. Achieve 23% average quality enhancement using Genetic Algorithms. Ideal for JPEG 2000, digital cameras, video, and MP3s. Results show significant improvements over standard wavelet methods.
E N D
Evolving Multiresolution Analysis Transforms for Improved Image Compression and Reconstruction under Quantization Brendan J. Babb, Frank Moore, and Pat Marshall University of Alaska, Anchorage and AFIT CIISP 2007
Results • We were able to improve image quality on average by 23% over a known wavelet transform with quantization using a Genetic Algorithm to evolve forward and reverse transforms. • For 3 level MRA the improvement is 11% over the standard wavelet.
Overview • Why I might care? • Wavelet image compression and quantization • Evolving wavelet like transforms • Results • Future Research • Questions
Applications • JPEG 2000 • FBI Fingerprints database – 200 million cards – 2000 Terabytes of data • Web • Digital Cameras • Video • MP3s
Wavelet Compression Compressor Original Image Forward Wavelet Transform Quantizer Encoder 10011… Decompressor Lossy Image Inverse Wavelet Transform Dequantizer Decoder
Quantization • Quantization of 64 • Y value is 300 • 300/64 = 4.6875 = 4 • Dequantization multiplies 4 * 64 = 256 • 17 times smaller file size
Mean Squared Error (MSE) • The common method for comparing the quality of a reproduced image is Mean Squared Error • The average of the square of the difference between the desired response and the actual system output (the error) • Must consider file size
Genetic Algorithms • Optimization techniques inspired by Darwinian evolution
Previous Research • Dr. Moore published papers on 1-D signals and images, evolving the Inverse transform • 90% improvement on 1-D and 5 – 9 % improvement on images over Wavelets
Specifics • Matlab code modified from Michael Peterson’s code based on Dr. Moore’s code. • Forward and Reverse at the same time • Start with a population of real coefficients from a known Wavelet • Daubechies 4 ( 8 forward and 8 reverse) • MR Levels 1 through 3 • Parallel operation on Supercomputer
Genetic Operators • Initial Population • Fitness • Selection • Mutation • Cross-over
Fitness Function • Restrain File Size • A * MSE ratio + B * File Size ratio • Good MSE but bigger files or vice versa • Penalize for bigger file size or bigger MSE with if statement combinations
GA Parameters • Population size: 500 to 10000 • Generations: 500 to 2000 • Elite Survival Count: 2 • Parental Selection: stochastic uniform • Crossover: Heuristic • Mutation: varies by experiment • Population initialization: Random factor times the original Wavelet • Crossover to Mutation ratio: 0.7 (unless noted)
Resulting images • 23% MSE improvement for the same filesize for Fruits.bmp that generalizes • 40% MSE improvement for Zelda image
1 Level Runs Run #1 Run #2 Run #3
Evolved Coeffs SetMRA LevelValues (% Change Relative to D4 Wavelet) h1 (Lo_D) 1 -0.1278, 0.2274, 0.8456, 0.4664 (-1.24%, +1.47%, +1.09%, -3.44%) 2 -0.1274, 0.2289, 0.8446, 0.4661 (-1.55%, +2.14%, +0.97%, -3.50%) 3 -0.1278, 0.2281, 0.8455, 0.4670 (-1.24%, +1.78%, +1.08%, -3.31%) g1 (Hi_D) 1 0.4791, 0.8474, -0.2347, -0.1278 (-0.81%, +1.30%, +4.73%, -1.24%) 2 -0.4894, 0.8447, -0.2317, -0.1279 (+1.33%, +0.98%, +3.39%, -1.16%) 3 -0.4901, 0.8462, -0.2291, -0.1288 (+1.47%, +1.16%, +2.23%, -0.46%) h2 (Lo_R) 1 0.4811, 0.8152, 0.2274, -0.1069 (-0.39%, -2.55%, +1.47%, -17.39%) 2 0.4805, 0.8159, 0.2279, -0.1093 (-0.52%, -2.46%, +1.70%, -15.53%) 3 0.4820, 0.8172, 0.2278, -0.1097 (-0.21%, -2.31%, +1.65%, -15.22%) g2 (Hi_R) 1 -0.2008, 0.0274, 0.5960, -0.1472 (+55.18%, -87.78%, -28.75%, -69.52%) 2 -0.1618, -0.1105, 0.6870, -0.3201 (+25.04%, -50.69%, -17.87%, -33.73%) 3 -0.1572, -0.1495, 0.7861, -0.4033 (+21.48%, -33.29%, -6.03%, -16.50%)
Summary • Forward and Inverse Transforms evolved from Wavelets have better image quality than the Wavelet under quantization and multiple levels • Improves image quality with the same amount of file size • Training images exist which generalize well across other images
Recent Research • Increased Information Entropy results in 60% improvement for Zelda • Evolving for fingerprint images results in 16% improvement over FBI standard for 80 images (Humie) • Training over 4 images and using Differential Evolution • Evolved Fingerprint wavelet does poorly on standard test images
Future Research • Evolving different shape wavelets • Mathematically analyze • Use of different operators and techniques • What makes a good representative training image • Improve on JPEG 2000 wavelets • Custom wavelets for other applications
Fitness Logic • If (SE ratio > 1) and (IE ratio > 1) • then fitness = (SE ratio)^4 +(IE ratio)^4 • else if (SE ratio > 1) • then fitness = (SE ratio)^4 + IE ratio • else if (IE ratio > 1) • then fitness = SE ratio + (IE ratio)^4 • else • fitness = (SE ratio)^2 + IE • fitness = fitness *1000