Extraction And Application Of Wavelet-domain Fractal Texture Features In High-resolution Remote Sensing Images

Image texture features, exuberant in high resolution remotely sensed images, as one ofimportant assisted characteristics of spectral features, play an important role in image analysisand automatic recognition of land surface features. The extraction, classification andsegmentation of texture features from remotely sensed image are important research topics in thefield of image processing, with very broad research prospects.The paper proposes a wavelet domain fractal texture features extraction approach, andimplements it using the high resolution remotely sensed image of Fuzhou City. Main researchcontents and results are as follows:(1) Computation and comparison of fractal texture of remote sensing images. Fractaldimensions of images have been calculted using usual methods such as triangular prism method,fractal Brownian motion method, DBC box-counting method and multi-fractal method. After thecomparison of various fractal dimensions calculating methods with the variance and computationtime, a conclusion has been drawn that DBC box-counting method and multi-fractal calculationmethod are the most suitable for fractal texture calculation, with high accuracy in land typedistinction and high computation efficiency.(2) Computation and analysis of wavelet-field fractal texture features. One reasonableapproach of calculating the wavelet domain fractal texture feature has been proposed. Bycomparing methods combined fractal and wavelet theory, the overall combination approach hasbeen proved more reasonable. After the experiment using the overall processing approach, theresults shows that these methods are more conductive for the extraction and analysis ofmulti-scale fractal texture features: the first and second level rough information image, the firstand second level mean value images of three directions’ detailed information computed by theDBC gap feature and multi-fractal methods.(3) Application of wavelet-field fractal texture features on building extraction. Usingspectral and texture characters, an approach of buildings extraction from high resolutionremotely sensed image has been proposed. Effects of window size and decomposition level onwavelet domain fractal texture features have been analyzed. Using Quickbird image of Fuzhoucity, the classification results and building extraction accuracy have been improved based onboth the supervised classification results from spectral features and texture images.