|GeoImaging & GeoInformatics|
|Your location:||GIGI||>||Modules||>||Image Processing in the Spectral Domain|
|Introduction||Objectives||Getting Started||Essential Learning||Hungry Minds||Summary|
The classic view of a digital image is an array of pixels, each cell occupied by a Digital Number that relates in some way to the radiance reflected or emitted from a surface. Multispectral images consist of two or more layers, the DN of each layer being acquired by a sensor designed to measure radiance at a particular wavelength or waveband. Thus, perhaps not surprisingly, an early focus of digital image processing was the ability to process multispectral wavebands.
It is interesting to observe that, with the benefit of hindsight, we now realise that multispectral images often have low dimensionality and this means we have to think about how best to enhance the spectral information content and, perhaps more importantly, this makes it more difficult to extract information. This difficulty extends to all the major modelling approaches that have been utilised - namely biophysical modelling, thematic classification modelling and spectral mixture modelling.
The model-based approaches used by image analysts are all generic and similar techniques are used for studies of land, water and the atmosphere. Model-based approaches follow a common method. A model is created that relates a variable of interest to remote sensing data. These models can be operated in a forward 'descriptive' mode - forward modelling - or inverted, and the inverted model then used to estimate the variable of interest from a set of multispectral measurements. The main difference is that the approaches deliver information products of different data types. The output of a classification is a thematic map of nominal data values. That is if the model operates in a hard mode. If a soft classification model is applied then the output has ordinal characteristics. Ordinal data are also output from spectral mixture analysis, whilst information products derived from biophysical models are interval or ratio data.
Image processing in the spectral domain is a challenge but one that has attracted the attention of many researchers who have developed techniques that are used widely by image analysts.