SPATIAL OBJECT RECOGNITION VIA INTEGRATION OF DISCRETE WAVELET DENOISING AND NONLINEAR SEGMENTATION

摘要

Spatial digital image analysis plays an important role in the information decision support systems, especially for regions frequently being affected by hurricanes and tropical storms. For the aerial and satellite imaging based pattern recognition, it is unavoidable that these images are affected by various uncertainties, like the atmosphere medium dispersing. Image denoising is thus necessary to remove noises and retain important signatures of digital images. The linear denoising approach is suitable for slowly varying noise cases. However, the spatial object recognition problem is essentially nonlinear. Being a nonlinear wavelet based technique, wavelet decomposition is effective to denoise blurring spatial images. The digital image can be split into four subbands, representing approximation (low frequency feature) and three details (high frequency features) in horizontal, vertical and diagonal directions. The proposed soft thresholding wavelet decomposition is simple and efficient for noise reduction. To further identify the individual targets, nonlinear K-means clustering based segmentation approach is proposed for image object recognition. The selected spatial images are taken across hurricane affected Louisiana areas. In addition to evaluate this integration approach via qualitative observation, quantitative measures are proposed on a basis of the information theory, where the discrete entropy, discrete energy and mutual information, are applied for the accurate decision support.