Multi-resolution Remotely Sensed Image Segmentation in Combination with Spectral and Shape Features
WANG Min,WAN Qiming,ZHANG Daqian,ZHANG Qingfeng
2010-01-01
Geo-information Science
Abstract:The redundant image details of high resolution remotely sensed imagery may cause more distinct spectral diffusion and noise distribution.It sometimes makes traditional pixel-based image analyzing methods inapplicable.Nowadays,object-oriented image analyses have become a hot research topic in the field of remotely sensed image processing and interpretation.It is because the latter has stronger capability to deal with noise,and it can introduce more abundant image features and expertise knowledge in analyses.The first and most important step of object-oriented image analysis is image segmentation,which segments an image into many visual homogenous parcels.Based on these parcels,which are objects,not pixels,more features can be extracted which facilitate the succeeding image interpretation.In this study,we designed and implemented a multi-resolution image segmentation method combining spectral and shape features with reference to the basic ideas of eCognition,a famous object-oriented image analyzing software package.The algorithm includes the following steps.(1) The initial segmentation parcels,called the 'sub feature units',are obtained with rainfalling watershed algorithm,which is of very fast speed and fair segmentation precision.(2) A fast region merging technique is used to merge these sub feature units in a hierarchy way.Based on the sub feature units,a merge cost function integrating spectral heterogeneity and shape heterogeneity is designed to guide the merging of parcels.The use of the shape features is to make the merged parcels more regular in shapes.A scale parameter is used to control the merging process,which stops a merge when the minimal merging cost of parcels exceeds its power.A multi-resolution segmentation can be implemented by setting different scale parameters,for smaller scales mean less cost while merging which create smaller parcels,and vice versa.Several experiments on high spatial resolution remotely sensed imagery were carried out to validate our method.In most cases,the multi-resolution segmentation produced highly visually homogeneous image objects with arbitrary resolutions on different types of data.