MULTISPECTRAL DATA COMPRESSION USING STAGGERED DETECTOR ARRAYS (LANDSAT, REMOTE SENSING).

Abstract

A multispectral image data compression scheme has been investigated in which a scene is imaged onto a detector array whose elements vary in spectral sensitivity. The elements are staggered such that the scene is undersampled within any single spectral band, but is sufficiently sampled by the total array. Compression thus results from transmitting only one spectral component of a scene at any given array coordinate. The pixels of the mosaic array may then be directly transmitted via PCM or undergo further compression (e.g. DPCM). The scheme has the advantages of attaining moderate compression without compression hardware at the transmitter, high compression with low-order DPCM processing, and a choice of reconstruction algorithms suitable to the application at hand. Efficient spatial interpolators such as parametric cubic convolution may be employed to fill in the missing pixels in each spectral band in cases where high resolution is not a requirement. However, high-resolution reconstructions are achieved by a space-variant minimum-mean-square spectral regression estimation of the missing pixels of each band from the adjacent samples of other bands. In this case, reconstruction accuracy is determined by the local spectral correlations between bands, the estimates of which include the effects of interband contrast reversal. Digital simulations have been performed on three-band aerial and four-band Landsat multispectral images. Spectral regressions of mosaic array data can provide reconstruction errors comparable to second-order DPCM processing and lower than common intraband interpolators at data rates of approximately 2 bits per pixel. When the mosaic data is itself DPCM-coded, the radiometric accuracy of spectral regression is superior to direct DPCM for equivalent bit rates.