Multispectral Image Coding

Multispectral images are a particular class of images that require specialized coding algorithms. In multispectral images, the same spatial region is captured multiple times using different imaging modalities. These modalities often consist of measurements at different optical wavelengths, but the same term is sometimes used when the separate image planes are captured from completely different imaging systems. Multispectral images are three-dimensional data sets in which the third (spectral) dimension is qualitatively different from the other two. Because of this, a straightforward extension of two-dimensional image compression algorithms is generally not appropriate. Also, unlike most two-dimensional images, multispectral data sets are often not meant to be viewed by humans. Remotely sensed multispectral images, for example, often undergo electronic computer analysis. Most multispectral compression algorithms assume that the multispectral data can be represented as a two-dimensional image with vector-valued pixels. Each pixel then consists of one sample from each image plane (spectral band). This chapter presents some of the current methods for both lossless and lossy coding of multispectral images. Effective methods for multispectral compression exploit the redundancy across spectral bands while also incorporating more conventional image coding methods based on spatial dependencies of the image data. Importantly, spatial and spectral redundancy differ fundamentally in that spectral redundancies generally depend on the specific choices and ordering of bands and are not subject to the normal assumptions of stationarity used in the spatial dimension.