Hyperspectral image compression based on adaptive recursive bidirection prediction/JPEG

Abstract

This paper proposes a new recursive bidirection prediction (RBP)/JPEG method for hyperspectral image compression. The compression is achieved by using this novel RBP scheme for spectral decorrelation followed by the standard JPEG algorithm for coding the resulting decorrelated residual images. The key feature of the new method is that it presents a recursive algorithm to reduce the interband correlation. This algorithm is easy to implement, overcomes the drawbacks of one-sided prediction and coding noise accumulation in DPCM, and also overcomes the disadvantages of computation complexity in the KLT and VQ-based methods. In addition, by combining the RBP scheme with the standard JPEG algorithm, both spectral and spatial correlations inherent in hyperspectral images are exploited. The high compression performances are maintained. In order to evaluate the effectiveness of the proposed method, the computer simulations are conducted on AVIRIS data. The experimental results indicate that the new method provides significant improvement over conventional bidirection prediction (BP) and DPCM for hyperspectral image compression in both objective and subjective measures. While the average compression ratio ( CR) is increase compared with BP and DPCM, the average signal-to-noise ratio ( SNR) of the reconstructed images is also increases. The effect of compression on classification is negligible compared with the results of the original image classification.