Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Geography and Anthropology

First Advisor

Nina Lam


In this study we addressed fundamental characteristics of image analysis in remote sensing, enumerated unavoidable problems in spectral analysis, and highlighted the spatial structure and features that increase information amount and measurement accuracy. We addressed the relationship between scale and spatial structure and the difficulties in characterizing them in complex remotely sensed images. We suggested that it is necessary to employ multiscale analysis techniques for analyzing and extracting information from remotely sensed images. We developed a multiscale characterization software system based on an existing software called ICAMS (Image Characterization And Modeling System), and applied the system to various test data sets including both simulated and real remote sensing data in order to evaluate the performance of these methods. In particular, we analyzed the fractal and wavelet methods. For the fractal methods, the results from using a set of simulated surfaces suggested that the triangular prism surface area method was the best technique for estimating the fractal dimension of remote sensing images. Through examining Landsat TM images of four different land covers, we found that fractal dimension and energy signatures derived from wavelets can measure some interesting aspects of the spatial content of remote sensing data, such as spatial complexity, spatial frequency, and textural orientation. Forest areas displayed the highest fractal dimension values, followed by coastal, urban, and agriculture respectively. However, fractal dimension by itself is insufficient for accurate classification of TM images. Wavelet analysis is more accurate for characterizing spatial structures. A longer wavelet was shown to be more accurate in the representation and discrimination of land-cover classes than a similar function of shorter length, and the combination of energy signatures from multiple decomposition levels and multispectral bands led to better characterization results than a single resolution and single band decomposition. Significant improvements in classification accuracy were achieved by using fractal dimensions in conjunction with the energy signature. This study has shown that multiscale analysis techniques are very useful to complement spectral classification techniques to extract information from remotely sensed images.