Efficient lossy ultraspectral data compression: hybrid predictive transform coding algorithm.

• Online Access: http://hdl.handle.net/2047/d20003364

This project proposes a new lossy ultraspectral image compression algorithm, hybrid predictive transform (HPT) coding algorithm. It is based on the predictive coding and the transform coding algorithms. The HPT algorithm consists of four stages: preprocessing, prediction, discrete cosine transformation, and compression protocol. Since the attributes of the ultraspectral data change obviously in the spectral direction, the HPT employs the two predictors, that are used in the partition differential plus-code modulation (PDPCM) and low complexity lossless compression for two-dimensional (LOCO-2D) algorithms, to remove the correlation. When the threshold of spectral correlation is 0.9, the HPT algorithm divides the ultraspectral data into two groups-- the data with a low spectral correlation and the data with a high spectral correlation--in the preprocessing stage. The data with a high spectral correlation uses the PDPCM predictor to remove spectral correlation, and then information is converted into transform domain using a two-dimensional discrete cosine transform (2D-DCT). Since 2D-DCT implements two separable cosine transforms in horzentrol and vertical directions, it concentrates information on a number of low frequency coefficients. In order to increase the compression ratio, the HPT algorithm designs a compression protocol in each image to discard non-essential information. Each compression protocol, requiring 255 bits, is comprised of three components: removing parameter, essentiality sequence of index sets and corresponding quantization sequence. The parameter regulates the number of information discarded and the essentiality sequence evaluates the essentiality of each index set. Moreover, the HPT algorithm is an adaptive algorithm, where the compression protocol is adapted to the error of the reconstructed image by using a threshold to control the error generated in compression process. In order to ignore the error accumulation generated in prediction process, the HPT adopts a switching filter to restart the compression model when the error is too large. The HPT algorithm provides an excellent compression performance in lossy compression, that achieves compression ratio between 9:1 and 35:1 with average SNR ranging from 25 dB to 32 dB.