SEMIAUTOMATIC SEGMENTATION OF HIGH RESOLUTION IMAGERY WITH TEXTURE SEED REGION GROWING

High spatial resolution satellite imagery has become an important source of information for mapping and a great number of related applications. Region based segmentation of high resolution imagery is now considered a more suitable method than traditional per pixel classification techniques. Region growing is a classical method in image segmentation due to its simplicity and effectiveness in making using of spatial information among pixels. On the other hand, the automatic and optimal selection of the seeds of growing has been a key in the context. In order to take great advantage of human vision's capability of object recognition, this paper presents a semiautomatic segmentation scheme by which seed regions provided by human operator grow to their boundary separating the seed object and its background. The algorithm 'learns' texture measurement from the seed region and tries to expand the seed region till the grown region has maximal difference of texture property with the background while the in-class texture property is still consistent. We used a local binary pattern based texture measurement and tested the approach with a number of high resolution images to extract residential, forestry and different land coverage. The result shows its potential of practical utilization in analysis of high resolution imagery.