LARS Tech Report Number
070982
Abstract
Classification of multispectral image data based on spectral information has been a common practice in the analysis of remote sensing data. However, the results produced by current classification algorithms necessarily contain residual inaccuracies and class ambiguity. By the use of other available sources of information, such as spatial, temporal and ancillary information, it is possible to reduce this class ambiguity and in the process improve the accuracy.
In this paper, the probabilistic and supervised relaxation techniques are adapted to the problem. The common probabilistic relaxation labeling algorithm (PRL), which in remote sensing pixel labeling usually converges toward accuracy deterioration, is modified. Experimental results show that the modified relaxation algorithm reduces the labeling error in the first few iterations, then converges to the achieved minimum error. Also a noniterative labeling algorithm which has a performance similar to that of the modified PRL is developed. Experimental results from Landsat and Skylab data are included.
Date of this Version
January 1982