High Order Textural Classification of Two SAR ERS Images on Mount Cameroon

Abstract

Many researchers have demonstrated that textural data increase the precision of a classification when they are combined with level of grey information. However, the calculation of textural parameters of order two is often too long in a computer. The problem is more complex when one must compute higher order textural parameters, which however can considerably improve the precision of a classification. This work is based on statistical methods of order two and three for the calculation of textural parameters [Akono et al., 2003]. In this work, we suggest a new method of calculation of textural parameters, which is more general, not limiting itself only on order two or three, but which goes until an order n > 2, but still remaining applicable to orders two and three. The principle of this study consists in reducing the calculation of a n‐summation of type () generally utilized in the evaluation of textural parameters, to a simple summation of type . ψ is a function depending on the matrix frequency (co‐occurrence matrix in the case of order two parameters) of the n‐uplets of levels of grey in an image window. After this transformation, a considerable gain in time and memory space of the computer has been obtained. This method produces exactly the same result as the classical method, but with a spatial and a temporal gain growing with the order of the textural parameter.

This method has been developed and tested successively at orders two, three, four and five with two synthetic aperture radar (SAR) images of satellites ERS‐1 and ERS‐2, of size 986 × 1390 pixels, on a computer with a random access memory (RAM) of 352 megabyte and a Pentium processor of 2,4 gigahertz of speed. The obtained results show that, for the evaluation of the “summation” textural parameter of order two, considering an image window of size 7 × 7, the new calculation method is ten times faster than the classical method of computing co‐occurrence matrix of levels of grey. At order three, the new method is nineteen times faster and at order four, it is a twelve thousands times faster. The developed methodology has allowed making a comparative classification study of the ground's occupation on the volcanic area of mount Cameroon, between the years 1998 and 2003. In fact the “summation” textural parameter at orders two, three, four and five has been applied on the two SAR images of the study region. The comparison of the ground occupation areas on the study site has been done from a non supervised classification realized on the textural image. The increasing of the area of the flows of lava due to the 1999 and 2000 eruptions has been put in evidence.