THE EVALUATION AND OPTIM IZATION OF THE ENDM EMBERS EXTRACTED FROM PIXEL PURITY INDEX (PPI)

摘要

The extraction of sub-pixel in form ation plays a major role in accurate estimation of land cover due to the mixed pixels problem. In this regard, Linear Spectral Mixture Model (LSMM) could be considered as a common method in estimating the fractions of different classes with in a pixel. The accuracy of this model is significantly related to the identification of pure pixels. Thus, different methods have been developed for identifying pure pixels. Pixel Purity Index (PPI) is known as a prominent geometric index in this area. Large number of selected pixels, their poor quality and also unclassified nature of the output pixels could be considered as major lim itations of this algo rithm. In this study, clustering of image pixels based on the pure pixels'mask is used in order to clustering of pixels obtained from the PPI Likewise, thresholding on the PPI results is proposed to choose the most pure pixels. Then, the accuracy of the initial pure pixels obtained from the PPI and also the selected pixels based on the thresholding is evaluated by determining the number of under/over-shoot pixels and the RMSE of LSM M results. According to the results, the number of under/over-shoot pixels is decreased considerably by increasing the threshold value applied on the PPI image and is determined less than 5 percent of total image pixels.