The Study Of Filtering And Edge Detection For Hyperspectral Images

Based on the concepts of multi-dimensional co-correlation coefficients and density of co-correlation coefficients between pixels of remote sensed multi-spectral image. Using the analysis for neighboring pixels, one can define the minimum co-correlation coefficients and maximum co-correlation coefficients, the difference between those two for gray level vector of pixels. The density of co-correlation coefficients is the statistics for co-correlation coefficients within whole image. The applications of those concepts in simulative images show that the edge information can be extracted with the analysis using those concepts. The minimum co-correlation coefficients and the difference of co-correlation coefficients can provide evidently the information of position and strength of edges. That implies the analysis with co-correlation coefficients could be used for edge extraction of multi-spectral image. The further experiments have been performed using MAIS images. The different results with different value ranges of minimum co-correlation coefficients and the difference of co-correlation coefficients indicate that one should select the rational threshold of co-correlation coefficients for edge extractions.The energy edge can also be extracted by using the energy relationship of gray level vectors.This study introduces the method of frequency filtering for hyper-spectral (or multi-spectral) images and its application in hyper-spectral image edge detection. The main task of method is to extract a set of different frequency information from original multi-spectral data, and to dissolve mixed hyper-spectral signal to different components in a certain degree, in terms of hyper-spectral filtering.In order to test the applied efficiency of this method in the field of edge detection, the experiments have been done separately in hyper-spectral model and real multi-spectral remote sensing images. The main parameters of method have been fully analyzed: such as filter operators, multi-dimensional correlation coefficients, profile of continuous low frequency ecges, profile of continuous high frequency edges, changed degree of frequency edges, thresholds of displaying edges. The experiment results show the capability of hyper-spectral filtering method in analyzing hyper-spectral remote sensing data in general and hyper-spectral edge detection in particular.The studied results showed the differences between frequency edge maps by using methods of space filtering and spectral filtering.The method of standardizing frequency-correlation showed the ability to choose the most appropriate filtered parameters to analyze a parts of stable frequencies and unstable frequencies in original hyperspectral signal.