Institute of Remote Sensing and GIS
Peking University, Beijing China, 100871
This paper presents a new approach to recognition of the road on high ground resolution image from satellite.
The traditional statistical approach of pattern recognition focuses on processing the discrete pixel and feature extraction for ground object in high spatial resolution images. However, this approach takes only feature of spectrum on discrete pixel rather than the shape and spatial construction formed by the adjacent pixels belonging to the same object. So, this paper tries to develop a new approach to recognition pf the road based on the integration of the statistical approach of pattern recognition and syntactic approach of pattern recognition.
It is imperative to analyse the shape features of road for recognizing it correctly. The shape of road differs from other objects on the image. For example, the buildings have many pixels, but their shapes are regular rectangle. Observed from character of configuration road in the city is a kind of linear ground objects, the extension of the road is long and its width is narrow. The road can be recognized by the composed shape of primary elements, so this is the basis of the analysis based on clustering of primary elements.
This indispensable condition in forming the road is together a certain number of primary elements along the line on the image. This means the distance of primary elements in the road width is comparatively small. Furthermore, clustering algorithm of the primary elements are used to classify and gather primary element from satellite imagery.
Basing on this view point. 6*6 window is created according to centered position by 2*2 window. At the same time, positions of contiguous primary elements are analysed by using 6*6 window. They can be considered as primary elements of road if contiguous primary elements lie at one end or two ends of horizontal line or upright line of the reticle in 6*6 window. Except this situation, contiguous primary element, which is at other positions such as at the top left corner, lower left corner, top right corner or lower right corner in 6*6 window, does not belong to road. Hereby, primary elements of other ground objects can be excluded.
Thus the road in the city on the image is identified by the analysis of shape feature in windows.