The Marr Approach in Edge Information Analysis for Multispectral Remote Sensed Imagery

This paper proposes an approach to edge analysis of multi-band image using Marr's method,on the basis of the introductions of Marr theory and the principles of edge detection proposed by Marr and Hildreth. As to Marr theory,this paper describes its frame structure with a few words in order to let the readers have a whole concept to it. Then the principles of edge detection proposed by Marr-Hildreth is analyzed by dividing into two parts,which are how to select the smooth filter and why the Laplacian operator is valid. They show that after using the Gauss smooth filter,the edge of the image can be detected well with the Laplacian operator. LOG operator is the integration of the two parts. On this base,the paper emphasizes on exploring the valid approaches for multi-band remote sensing image(TM image). Some ways are pointed out in the paper,such as weighted vector summation,optimal band selection,etc. But the approach used in the experiments is dependent upon the number of "cross-point" using LOG operator for each band respectively. In details,we can use LOG operator to process every band image to get "cross-point" edge,and then use a detection method to select the gray of the final edge image,which can describe the best information of the multi-band image,namely how to decide the most appropriate edge gray,which comes from every band. The results of the experiments include the edge image of each band with different σ and the edge image of the multi-band image. The experiments using TM image show the better effect of multi-band image analysis.