A low-complexity lossless algorithm for compression of multispectral data has been developed that takes into account pushbroom-type multispectral imagers' properties in order to make the file compression more effective. These types of imagers use a detector array to acquire data in spatial-spectral slices. Each detector element corresponds to a specific spectral band and cross-track position. Because the characteristics of detector elements generally vary somewhat from element to element, cross-track adjacent samples in a given spectral band are not as similar as they are in an instrument that uses the same detector for all samples in a given spectral band (e.g. in a whiskbroom-type instrument). Along-track adjacent samples will tend to be very similar.
Therefore, the fast lossless algorithm described in "Fast Lossless Compression of Multispectral-Image Data" (NPO-42517), NASA Tech Briefs, Vol. 30, No. 8 (August 2006), pages 26-28 was modified in two ways to increase the compression efficiency for these imagers. The first modification involves a local mean that was originally computed, for each sample, as the average sample value in a causal neighborhood of four adjacent samples within the spectral band. Local means are subtracted from sample values to produce the input quantities for the adaptive filtering algorithm. For data from pushbroom-type multispectral imagers, it is found that letting the local mean be equal to the previous sample in the same cross-track position (and in the same spectral band) gave significantly better results. Specifically, for all rows (along-track positions) except the first row in a segment, the local mean equals the previous sample in the same cross-track position. For the first row in a segment, no such sample is available, so the local mean equals the causal cross-track adjacent sample. In addition, the prediction neighborhood of the sample is changed. In the original algorithm, this neighborhood contains three samples from the same band as the given sample, and one sample from each of the three preceding spectral bands. Under this first modification, only the first three samples from the three preceding spectral bands are used.
For the second modification, crude calibration information is used as side information for the compressor. This modification exploits the observation that the variations in detector element characteristics are an inherent property of the detector, and there is little change in this variation over time. A key ingredient in this modification is a calibration array containing approximate, constant offsets for the detector elements. Each sample in a data set has the corresponding calibration value subtracted prior to starting compression, and the decompressor reverses this process after decompression. It is noted that the decompressor requires a copy of the calibration information. There are various ways in which the calibration array might be generated. In this instance, only a single calibration array was generated, but it appeared to work well with a large number of datasets from the same instrument.
When used with the first modification, the second modification primarily only affects how the first line of each segment is compressed. Nevertheless, the improvement for this line is significant enough to provide a noticeable, overall compression improvement.
This work was done by Matthew Klimesh of Caltech for NASA's Jet Propulsion Laboratory.
NPO-45473
This Brief includes a Technical Support Package (TSP).
Lossless, Multi-Spectral Data Compressor for Improved Compression for Pushbroom-Type Instruments
(reference NPO-45473) is currently available for download from the TSP library.
Don't have an account?
Overview
The document discusses advancements in lossless data compression for multispectral imagery, particularly focusing on pushbroom-type instruments. These instruments capture data in spatial-spectral slices using a detector array, where each element corresponds to a specific spectral band and cross-track position. The inherent variability in detector characteristics can lead to less similarity between cross-track adjacent samples compared to whisk-broom-type instruments, making traditional compression methods less effective.
To address these challenges, the document introduces two modifications to the original "fast lossless" compression algorithm, which employs predictive compression through adaptive filtering. Modification 1 alters the prediction neighborhood by using only three samples from preceding spectral bands, rather than including samples from the same band. This change improves the algorithm's performance by allowing the local mean to be based on the previous sample in the same cross-track position, enhancing compression effectiveness.
Modification 2 incorporates crude calibration information as side information for the compressor. It utilizes a calibration array containing approximate constant offsets for detector elements, which are subtracted from each sample before compression. This approach leverages the stability of detector characteristics over time, allowing for more effective compression. The decompressor requires access to this calibration information to reverse the process post-decompression.
The results of applying these modifications show significant improvements in compression efficiency. For three datasets tested, Modification 1 reduced compressed sizes by 15-25%, while the combination of Modifications 1 and 2 further decreased sizes by approximately 0.1 to 0.2 bits/sample. The document provides detailed results in a table comparing the original fast lossless compressor with the modified versions, demonstrating the effectiveness of the proposed changes.
Overall, the modifications not only enhance compression performance but also maintain a relatively low complexity in the algorithm. While Modification 1 slightly reduces complexity, Modification 2 requires additional storage for the calibration array, which may need occasional updates. The document concludes that both modifications can be integrated into hardware implementations without significant challenges, making them viable for real-world applications in aerospace and beyond.
