Share Email Print

Proceedings Paper

A region-based high spatial resolution remotely sensed imagery classification algorithm based on multiscale fusion and feature weighting
Author(s): Leiguang Wang; Tiancan Mei; Qianqin Qin
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

With the availability of high resolution multispectral imagery from sensors, it is possible to identify small-scale features in urban environment. Given attributes of image structure such as color, texture, have the character of highly scale dependency, a hierarchy segment fusion algorithm based on region deviation is proposed to extract more robust features and benefit single semantic level land cover classification. The fusion algorithm proposed is divided into in two successive sub-tasks: mean shift (MS) filtering based pre-segmentation and hierarchical segment optimization. Presegmentation is applied to get boundary- preserved and spectrally homogeneous initial regions, and then, a family of nested image partitions with ascending region areas is constructed by iteratively merging procedure. In every scale, regions of the corresponding critical size are evaluated according to potential region merge risk, which is measured by the region standard deviation change before and after a virtual merge. If a region measurement is larger than a specified change threshold, the region will be preserved to the next level and labeled as a candidate segment for following regionbased classification. Otherwise the segment will be merged to the next scale level. After fusing segments in different scales, a novel weighted minimum distance classifier is employed to get supervised classification result, in which every feature band's deviation is used to calculate its own weight. We show results for classification of a HR image over Washington DC Mall area taken by the HYDICE sensor. Different features combined with designed classifier have proved that fused segments provided a robust feature extraction and improve classification accuracy.

Paper Details

Date Published: 30 October 2009
PDF: 8 pages
Proc. SPIE 7494, MIPPR 2009: Multispectral Image Acquisition and Processing, 749411 (30 October 2009); doi: 10.1117/12.832824
Show Author Affiliations
Leiguang Wang, Southwest Forestry Univ. (China)
Wuhan Univ. (China)
Tiancan Mei, Wuhan Univ. (China)
Qianqin Qin, Wuhan Univ. (China)

Published in SPIE Proceedings Vol. 7494:
MIPPR 2009: Multispectral Image Acquisition and Processing
Faxiong Zhang; Faxiong Zhang, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?