一种最大压缩误差可控的高光谱图像压缩算法

A Hyperspectral Image Compression Algorithm of Maximum Compression Error Controllable

提出了一种可以实时控制压缩误差的高光谱图像有损压缩算法。该算法对高光谱图像矩阵进行奇异值分解得到奇异值矩阵和奇异向量矩阵;用部分奇异值及其对应的奇异向量重构图像;根据测量系统精度要求确定量化因子并对重构图像与原始图像间的光谱误差进行量化;采用预测编码和算术编码对用于图像重构的奇异值及其对应奇异向量进行无损压缩;设计了非零值编码算法完成对重构误差的无损压缩。对Luna Lake和Low Alti-tude图像的仿真结果为:最大相对误差分别控制在0.44%和0.33%时,压缩比为41.5:1和24.6:1,信噪比为50.4 dB和47.8 dB。

A novel algorithm for hyperspectral image compression is proposed in order to control compression error in the process of real - time image compression. In the first step, Singular Value Decomposition （SVD） of the original image matrix was computed and the singular values and singular vectors of the matrix were obtained; the image was then reconstructed with a smaller set of singular values and singular vectors. In the second step, the spectrum errors between the regional image and reconstructed image were calculated by subtracting the reconstructed image spectrum from the original image spectrum; the quantification for the spectrum errors could be obtained by dividing the maximum spectrum errors got from the first step with the acceptable error of the test system. Lastly, the singulars values and singular vectors for reconstructing image were compressed by lossless predictive coding and arithmetic coding, the quantified spectrum errors were also compressed by a novel lossless compression algorithm of non-zero element coding designed in this paper. The results of the simulation on the hyperspeetral images of Luna Lake and Low Altitude show that when the maximum relative errors are controlled to be 0.44% and 0.33% respectively, the compression ratios are. 41. 5 ： 1 and 24.6 ： 1, the SNRs are 50.4 dB and 47.8 dB.