Object oriented classification of coastal landform based on texton theory

This paper focus on classification of coastal landform. The study area is locate in QinZhou, Guangxi Zhuang Autonomous Region, China, using GF-1 image. In view of complex coastal landform, the current study firstly transforming RGB model to CIE LAB model, dividing image based on colour gradient, then conducting Gabor filtering and PCA transformation to develop texons and generating texton histogram. Finally, maximum likelihood classification is employed for classification. The study results demonstrate that classification accuracy has been improved due to the capacity of texton-based model in maintaining more image features, achieving variable decoupling, and reducing variable correlation.