Elimination of the Outliers from Aster Gdem Data

Digital Elevation Models (DEM) provided by stereo matching techniques often contain unwanted outliers due to the mismatched points. Detected outliers are candidate for wrong data that may otherwise adversely lead to model misspecification, biased parameter estimation and incorrect results (Maimon and Rokach, 2005). Therefore detection of outlying observations and elimination of them from the data is considered as one of the first steps towards obtaining a refined DEM. Several statistical and non-statistical methods are available for analyzing the point cloud and extracting the outliers. The statistical methods often assume that the data follow a Gaussian normal distribution and hence, they look for the observations which are represented doubtful based on mean and standard deviation. In Digital Elevation Models generated from ASTER data the outliers, i.e., artifacts and anomalies, are represented as the following types: residual clouds, and errors due to different stack numbers: it occurs when generating the final GDEM from variable number of individual ASTER DEMs. In this paper a segment-based algorithm is proposed for detecting and eliminating the outliers from ASTER GDEM images. Furthermore, a water mask is produced as an additional layer representing the lower quality of the pixels located on inland water bodies. In the segment-based algorithm, the potential regions are iteratively extracted and evaluated using geometrical feature descriptors to classify the outliers and non-outlier regions. After eliminating the location of the outliers in original ASTER GDEM, an interpolation procedure is employed to fill the gaps. The quality of the final product is compared to the SRTM data and quality parameters are measured.