Statistical and Visual Analysis of the Cross Talk Correction and the ECW Compression Effects on Feature Extraction of Aster Data

Chuyen Nguyen


With increasing numbers of ASTER applications and explorations, ASTER imagery processing faces versatile challenges. Multispectral images obtained in shortwave infrared (SWIR) regions are very important for numbers of remote sensing applications, including geology, chemistry, and earth vegetation surface assessment (Rowan & Mars, 2002). Systematic sensor errors such as crosstalk (scattering) among sensors is a critical problem of image quality. Crosstalk correction is used to rectify and enhance the radiometric quality of ASTER imagery. This thesis shows significant results of crosstalk correction in the SWIR data (Bands 5, 6, and 7). The purpose of this research is to comprehend the qualitative and quantitative natures of changed pixel values in order to achieve effective analysis for feature extraction and pattern recognition. The correction techniques lead to unique changes in individual brightness values, in relationships among brightness values, and in the entire brightness range. The crosstalk correction changes each brightness value pixel very differently. It affects each ratio image in a very different way. It is important to visualize pixel by pixel differences and quantify the amount of change. The research shows that the further the distance from Band 4 is, the more scattered the area of crosstalk influence is, and the weaker the crosstalk influences. For large a dataset, it can be critical for display, storage, processing, and internet transmission. ASTER data is a very large dataset, 84.24 MB, as determined by 5736 × 5133 pixels. The choices of compression algorithm, compression ratio, compression option, can result in lost or changed information. This research goal is to visualize pixels by pixel differences and quantify the amount of changed values. It found that ECW

completely changes the VNIR data set. Bands 1, 2, and 3 are dynamically changed by the ECW compression. None of the brightness value pixels remains the same. ECW significantly reduces the image contrast. It causes brightness values to be diffused or scattered about the mean of VNIR imagery data. It totally alters the correlation among bands which causes errors in the feature extraction results. The fact that ECW compression converts all null pixels to non-null pixels changes VNIR data into a completely different data set.