Hyperspectral影像压缩技术综述与应用

需积分: 10 144 浏览量更新于2024-07-19 收藏 45.15MB PDF 举报

《光谱数据压缩》是一本由Giovanni Motta、Francesco Rizzo和James A. Storer合编的专业书籍，着重于高级光谱数据（Hyperspectral Data）的压缩技术。本书涵盖了多个关键主题，反映了在信息技术领域对于高分辨率光谱图像处理与存储需求的日益增长。以下是一些章节中的核心知识点： 1. **无损预测性压缩**：介绍了一种利用预测编码策略来对光谱图像进行压缩的方法，这种方法通过分析数据的内在结构，如相邻像素之间的关系，实现不丢失任何信息的压缩过程。 2. **线性预测的光谱图像压缩**：探讨了基于线性模型的压缩方法，通过预测并消除冗余信息，有效地减小数据尺寸，同时保持图像质量。 3. **超紫外传感器数据的压缩**：针对特定的高光谱分辨率数据源，讨论了针对这类特殊类型数据设计的优化压缩方案，以适应其独特的特性。 4. **局部最优分割矢量量化**：介绍了一种将光谱数据划分为小块并使用矢量量化技术的方法，这种方法强调了在压缩过程中对局部信息的精确处理。 5. **基于Crisp/Fuzzy自适应DPCM的近无损压缩**：Crisp/Fuzzy是动态分段的编码方式，它能够根据图像细节的复杂程度调整压缩效率，提供接近无损的压缩效果。 6. **联合分类与压缩**：探讨了如何将图像分类任务与压缩技术相结合，以提升压缩效率的同时实现数据的有效分类，这对于环境科学和遥感应用尤为重要。 7. **Trellis编码量化在光谱图像中的应用**：Trellis编码是一种有效的数据压缩算法，它结合了前向和后向搜索策略，能更好地捕捉数据的统计特性，提高压缩比。 8. **三维小波变换压缩**：利用多尺度分析的三维小波变换方法，可以捕捉光谱数据在空间和频率域的特征，从而实现高效的压缩。 9. **光谱/空间光谱图像压缩**：这个章节可能讨论了同时考虑光谱和空间信息的综合压缩策略，旨在更全面地保留图像的多维特性。 10. **JPEG2000在地球科学数据压缩中的应用**：JPEG2000是一种广泛应用于图像和视频压缩的标准，对于地球科学中的海量数据，其高效编码性能和渐进传输特性具有重要意义。 11. **光谱图像压缩中的环形失真现象**：探讨了压缩过程中可能出现的一种问题——光谱 ringing（环形失真），即频谱边缘的失真，如何通过优化算法和技术避免或减少这种现象。《光谱数据压缩》是一本深入研究和实用指南，提供了丰富的理论和技术实践，对于从事遥感科学、信号处理、计算机视觉以及数据存储领域的研究人员和工程师来说，具有很高的参考价值。

Mark R. Pickering and Michael J. Ryan 7

2.3 A Suitable Distortion Measure

Since a general theoretical distortion measure is unlikely to be developed

for remote-sensing applications in the short term, a suitable compromise

distortion measure must be found. The answer will depend on the end-user

application which, as identified earlier, could be in a variety of forms. The

lossy compression of the image must not prejudice the use of, or conversely

favour, any one of the possible applications.

The distortion measure must also produce predictable results in the

performance of each remote-sensing application for reconstructed data with a

given level of distortion. By determining a particular remote-sensing

algorithm's performance on reconstructed data in the presence of the given

level of distortion, an application scientist should be able to infer the

performance of the algorithm on the original (undistorted) data.

This requirement is very important as the compressed data is of little

value unless the performance of a remote-sensing algorithm on the original

data can be inferred by observing the performance of the algorithm on data

distorted to any particular degree. The requirement also implies that the

performance of the algorithm gracefully degrades as the level of coding

distortion increases.

2.3.1 The Suitability of Mean Distortion Measures

Mean distortion measures are not suitable since they cannot be obtained

directly because a decision on a mean distortion level cannot be made on a

pixel-by-pixel basis. In video coding, it is normal to accommodate this

difficulty by encoding on a pixel-by-pixel basis using scalar quantization,

which can be predicted to result in a desired mean distortion across the

image. The other difficulty with a mean distortion measure is that the

specification of a mean level of distortion gives a general indication of the

distortion in the image but does not give any indication of the specific

magnitude of the errors in any particular spatial or spectral portion of the

image. Mean distortion measure therefore are of little assistance in the

application of

quantitative algorithm.

Additionally, in almost all quantitative applications, algorithms make

decisions based on local information such as pixel values. A mean distortion

measure provides an indication of the distribution of errors, but does not

assist the algorithm to make local decisions because it provides no

information as to the specific location and magnitudes of errors. Such

information can only really be provided by an absolute distortion measure

8 HYPERSPECTRAL DATA COMPRESSION

that indicates something quantitative about both the local and global

distortion in an image.

2.3.2 The Suitability of Absolute Distortion Measures

In addition to giving precise information about local distortion, absolute

distortion measures are able to be applied on a pixel-by-pixel basis as the

image is being encoded. Two possible absolute distortion measures are:

maximum absolute distortion dindpercentage maximum absolute distortion.

The maximum absolute distortion (MAD) measure [21] guarantees that

every pixel ^{x,y,t) in the reconstructed image is within a maximum

distance d of its original value B{x,y,t), where d is in absolute brightness

value terms. This MAD measure is inflexible, however, as it takes no

account of the original magnitude of B{x,y,t). Pixels with small original

values will therefore have a larger relative error in the reconstructed image

than will pixels that had large original values. A suitable distortion measure

must be relative to the original pixel intensity. The best way of achieving this

is to use a percentage absolute distortion measure.

The percentage maximum absolute distortion (PMAD) measure [21]

guarantees that every pixel ^{x,y,t) in the reconstructed image is within a

maximum distance of/?*100 percent to its original value B(x,j;,/). This

PMAD distortion measure appears to be the most suitable for application to

lossy compression for quantitative analysis. It gives a direct indication of the

loss of scientific value for a particular level of distortion and is able to be

implemented on a pixel-by-pixel basis. It also offers the best potential to

produce predictable results in a range of applications.

A similar measure to PMAD, the percentage maximum absolute error

(PMAE), was used by Motta et al [23].

Once a suitable distortion measure has been determined it then needs to

be incorporated into a compression algorithm that has been designed as a

complete system. A suitable architecture for such a system is described in the

following section.

A SUITABLE ARCHITECTURE FOR THE COMPRESSION OF

HYPERSPECTRAL DATA

This basic architecture for a hyperspectral compression system is shown

in Figure 1. This architecture consists of the following components; pre-

processing, compression, post-processing and classification.

剩余420页未读，继续阅读

aphrodite_aglai

粉丝: 4
资源: 35

Hyperspectral影像压缩技术综述与应用

DE405 JPL星历

高光谱/遥感图像常用数据集

Adaptive interference hyperspectral image compression with spectrum distortion control

Hyperspectral image compression using three-dimensional significance tree splitting

Deep Learning-Based Classification of Hyperspectral Data

Unmixing approach for hyperspectral data resolution enhancement using high resolution multispectral image

Recursive Dictionary-Based SimultaneousOrthogonal Matching Pursuit for Sparse Unmixing of Hyperspectral Data

Assessment of Zn and Pb pollution in Sheyang Estuary (China) using Hyperspectral data

Hyperspectral data analysis in R:R中高光谱数据的处理和基本分析-开源

generate_hyperspectral_data.zip_hyperspectral

最新资源