Introduction - Hyperspectral Image Fusion - page 9

Fig. 1.3 Schematic representation of the 3-D hyperspectral data cube and illustration of the spectral
signature

The first band of the hyperspectral data is stored as a stand alone image. A 2-D
image array is stored by collating the rows—which is a standard format for saving a
2-D data. This format is known as the band sequential (BSQ). If the hyperspectral
image is to be used for classification related activities, one needs to work with the
spectral signatures. In such a case, it is desirable to store the data as a contiguous
collection of 1-D spectral arrays. The band interleaved by pixel (BIP) structure stores
the hyperspectral image as a set of location-wise arrays for all the spatial locations
in the data. The BIP format facilitates faster reading and processing when individual
spectral signatures are to be dealt with. However, in order to observe any of the bands,
one needs to read the entire data, extract the necessary pixel values, and rearrange
them. A compromise between these two storage formats is provided by the Band
interleaved by line (BIL) format. In the BIL format, one stores the intensity values
of the first scanline of the first band, followed by the fist scanline of the second band,
and so on. After storing the first scanline values for all the bands in the data, one
proceeds towards the second scanline and the process continues till the entire data
gets stored. Table 1.3 summarizes the three formats. If the hyperspectral data contain
K bands of dimensions

pixels each, the structures for each storage format
can be described as given in Table 1.3 .

( X × Y )

Table 1.3 Storage formats for hyperspectral data
Storage format

Acronym Data structure

Band sequential BSQ X × Y × K

Band interleaved by pixel BIP K × X × Y

Band interleaved by line BIL X × K × Y