Predictive coding of hyperspectral images

Agnieszka C. Miguel, Richard E. Ladner, Eve A. Riskin, Scott Hauck, Dane K. Barney, Amanda R. Askew, Alexander Chang

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

13 Scopus citations

Abstract

Every day, NASA and other agencies collect large amounts of hyperspectral data. For example, one Airborne Visible InfraRed Imaging Spectrometer (AVIRIS) alone can produce data that require up to 16 Gbytes of storage per day. The hyperspectral images are used to identify, measure, and monitor constituents of the Earth's surface and atmosphere [1]. This huge amount of data presents a compression challenge. In this research, we propose algorithms to code the hyperspectral data. To reduce the bit rate required to code hyperspectral images, we use linear prediction between the bands. Each band, except the first one, is predicted by previously transmitted band. Once the prediction is formed, it is subtracted from the original band, and the residual (difference image) is compressed using a standard compression algorithm. To optimize the prediction algorithm we study several methods of ordering the bands for prediction. To rearrange the bands into a particular ordering, we define a measure of prediction quality, the prediction mean squared error. We compute the optimal ordering using this measure as well as two restricted orderings in which each band can be predicted by the best predictor among all of the previous or future bands in the standard band numbering. In addition, we define two simple orderings in which each band is predicted by its immediate previous or future neighbor. We use the prediction mean squared error to compare those orderings. The first proposed algorithm is lossless, that is the decompressed images are exact replicas of the original data. The difference images are encoded using bzip2 data compression algorithm [2]. We use bzip2 because it is a state-of-the-art open-source lossless data coding algorithm. We compare our results for five standard hyperspectral images with recently published results and conclude that our algorithm achieves comparable compression ratios. The second algorithm is lossy and therefore, the decompressed image is an approximation of the original image. In this case we encode the difference image using the Set Partitioning in Hierarchical Trees (SPIHT) algorithm [3], which is a wavelet-based lossy compression technique that codes images with both high compression ratio and high fidelity. SPIHT was originally designed as a sequential algorithm; however, with some modifications, it can be parallelized for implementation on field pro grammable gate arrays (FPGAs) [4] and therefore has great potential for applications where the compression is performed in hardware on the aircraft and satellite platforms. Note that we compress all bands to the same fidelity. To compute the exact difference between a band and its prediction, the encoder must have access to the decoded version of the band used for prediction; however, such a closed loop system requires a full implementation of the decoder at the transmitter, which increases its complexity. In this chapter we present a new prediction technique, bit planesynchronized closed loop prediction, that significantly reduces the complexity of the encoder [5]. Instead of requiring the encoder to fully reconstruct the compressed band from which the current band is predicted, the encoder and the decoder simply use the same integral number of full bit planes of the wavelet-coded difference image of the band used for prediction. This enables the transmitter to be less complex because, while it must still do an inverse wavelet transform, full decompression is avoided. The proposed prediction method is very promising in that for the same target fidelity, the average bit rate is only slightly higher than for traditional predictive coding. The chapter is organized as follows. In Section 2, we review related background material. In Sections 3 and 4, we describe our prediction methodology. The algorithm for lossless predictive coding of hyperspectral images is presented in Section 5. In Section 6, we introduce our new reduced complexity lossy encoder. Finally, we conclude in Section 7.

Original languageEnglish
Title of host publicationHyperspectral Data Compression
Pages197-231
Number of pages35
DOIs
StatePublished - 2006

Fingerprint

Dive into the research topics of 'Predictive coding of hyperspectral images'. Together they form a unique fingerprint.

Cite this