Reduced complexity wavelet-based predictive coding of hyperspectral images for FPGA implementation

Abstract

We present an algorithm for lossy compression of hyperspectral images for implementation on field programmable gate arrays (FPGA). To greatly reduce the bit rate required to code images, we use linear prediction between the bands to exploit the large amount of inter-band correlation. The prediction residual is compressed using the Set Partitioning in Hierarchical Trees algorithm. To reduce the complexity of the predictive encoder, we propose a bit plane-synchronized closed loop predictor that does not require full decompression of a previous band at the encoder. The new technique achieves almost the same compression ratio as standard closed loop predictive coding and has a simpler on-board implementation.

An efficient macroblock-based diverse and flexible prediction modes selection for hyperspectral images coding

“…So, the best compression scheme can be determined by choosing the minimum average prediction MSE over all the 224-band images from all of the orderings. From the experimental results, the performance of the best reverse ordering is actually very close to the optimal ordering and the running time of the best reverse ordering is much smaller than the latter, so the best reverse ordering is used in [23]. The reordering algorithm could provide an improvement in compression ratio.…”

“…To take advantage of the correlation between the bands and to follow the fact that some bands do not perform well as predictors, while other bands are very easily predicted, the best ordering and the prediction based on the best ordering are employed in [23], in which the compression scheme is measured by the prediction MSE,99B i ÀB i À 99 2 =n, where B i and B i are the original and the predicted band i, respectively, and n is the number of the pixels in a single band. Generally, the higher the prediction MSE, the more bits are needed to compress the difference.…”

“…The reason for doing so is to abide by the facts that the correlations between the macroblock and the ones in the corresponding reference band images are not always identical. Table 1 lists the average prediction MSE over all the pre-quantized 224 bands of three hyperspectral images, using 3 different band orderings prediction in [23] and our macroblock-based inter-band prediction method, respectively. As can be seen, the minimum MSE can be obtained by our method.…”

“…Exploiting the strong correlation, a linear prediction between bands was proposed in [23] to greatly reduce the bit rate required to code images, in which several band orderings are compared and the best reverse ordering is used for all the simulations. However,in the complexity of the reordering there is O(m 2 ), where m is the number of bands of the dataset.…”

Lossless Predictive Compression of Hyperspectral Images

“…Wavelet decomposition and predictive coding can be combined in two ways. The prediction error residuals can be encoded using a wavelet coder [9], or the decomposition of the images can be used as a preprocessing step prior to prediction.…”