Multispectral remotely sensed imagery segmentation based on first fundamental form

Abstract

Image segmentation is a valuable approach that performs an object-based rather than a pixel-based analysis of high-resolution remotely sensed imagery. An approach for segmenting the multispectral QuickBird image based on first fundamental form is proposed. The value of multispectral image at a given point can be regarded as N-dimensional vector, and the difference of image values can be defined from the theory of first fundamental form, which allows to access gradient information from all bands simultaneously. Thus it is used to fuse the gradient feature of all bands. Then, the image segmentation is implemented based on marker-controlled watershed transform. The experimental results show that the proposed approach gives a better solution of integrating multispectral information for the segmentation of remotely sensed imagery.