Remote sensing images are used to estimate certain geophysical parameters or detect the presence or extent of geophysical phenomena. The amount of information extracted from remotely sensed data is dependent upon the type of sensor, the portion of the electromagnetic spectrum used, the quality of the data recorded, and certain physical limitations. There are other limiting factors involved in processing the data once they have been collected. In addition to noise added inherently by the sensor, image processing techniques also corrupt the image with noise in varying degrees. One of the measures to quantify information content is classification accuracy. Although the image information content reduces as the image is corrupted with noise, in certain applications there may not be as great a loss as one might expect. This can be attributed to the fact that although the value of a pixel may change as a result of corruption due to noise, the same pixel may in most cases, be correctly classified. Our research reveals that this loss in information is exponentially related to the variance of the added noise. The model is seen to be applicable for Landsat TM as well as multi-look and single-look SIR-C imagery. We observe that the relationship is independent of the type of noise (Gaussian, Gamma, or exponential). However, the rate of information loss increases with increasing correlation distance in the case of spatially correlated noise. The rate of information loss also increases with the number of classes chosen for classifying the scene. Using our mathematical model for information content as a function of noise variance, one can specify an “allowable” signal-to-noise ratio for a specific application.
ASJC Scopus subject areas
- Geography, Planning and Development
- Water Science and Technology