1 DESIGN OF FAST ALGORITHMS FOR PIXEL PURITY INDEX FOR ENDMEMBER EXTRACTION IN HYPERSPECTRAL IMAGERY

This paper presents new fast algorithms to improve the widely used endmember extraction algorithm, called pixel purity index (PPI) and also modify a revised version of PPI, called block of skewers (BOS) method. Since both the PPI and the BOS method suffer from same drawbacks in implementation, our proposed algorithms are designed to mitigate these dilemmas by introducing a newly developed concept, called virtual dimensionality (VD) and a process that can automatically determine a set of endmembers without appealing for parameters used in the PPI and BOS method. As a result, two modified PPI and BOSPPI algorithms are developed. Unlike the PPI and BOS method, the two new algorithms are iterative and provide stopping rules to automatically extract endmembers without manually selecting endmembers. The experimental study shows that the MPPI performs better than the PPI not only in computational complexity, but also in stability and performance, and the MBOSPPI could perform better than the BOS method if the blocks of skewers are properly designed. 1