Compressing scientific data with control and minimization of the L -infinity metric under the JPEG 2000 framework

Aldo Lucero, University of Texas at El Paso

Abstract

The work presented here has as its main challenge the ability to control the maximum absolute error (MAE), which is also known as the L ∞ metric, when compressing scientific data sets using JPEG 2000 Part 2 and residual coding. The residuals are compressed using JPEG 2000 Part 1 by bypassing the wavelet transform, which means that only the embedded block coding with optimized truncation (EBCOT) is used. In addition to controlling the MAE, techniques that minimize the MAE when a bit rate is specified are also described and tested. The proposed algorithms use JPEG 2000 for both, lossy compression of the original data and lossless residual coding. Due to the use of these two compression steps, the overall compression (total bit rate) is a function of two variables, the lossy bit rate and the lossless residual bit rate. For the case of controlling the MAE, two methods that search for the lowest total bit rate based on information obtained at two pivot points (lossy bit rates) are proposed. The results obtained using these techniques are compared with those acquired utilizing an exhaustive search. It will be shown that the pivot-based algorithms yield similar results obtained by using the exhaustive search. For the case of minimizing the MAE when a bit rate is specified, two techniques are developed and tested. The use of the JPEG 2000 quality scalability property is analyzed with regards to solving this same problem. This technique requires some form of bit allocation for the case of 3D data sets. Analysis and results for the methods that minimize the MAE are also included. It will be demonstrated that JPEG 2000 can be used to control the maximum absolute error using model-based techniques that omit the use of an exhaustive search. The main idea behind this work is that the MAE can be controlled under the JPEG 2000 compression framework. Techniques and algorithms developed in order to accomplish this are discussed here.

Subject Area

Electrical engineering

Recommended Citation

Lucero, Aldo, "Compressing scientific data with control and minimization of the L -infinity metric under the JPEG 2000 framework" (2007). ETD Collection for University of Texas, El Paso. AAI3321160.
https://scholarworks.utep.edu/dissertations/AAI3321160

Share

COinS