A COMPARISON OF THE PERFORMANCE OF PIXEL-BASED AND OBJECT-BASED CLASSIFICATIONS OVER IMAGES WITH VARIOUS SPATIAL RESOLUTIONS

摘要

In the last two decades, advances in computer technology, earth observation sensors and GIS science, led to the development of "Object-based Image Analysis (OBIA)" as an alternative to the traditional pixel-based image analysis method. It was recognized that traditional pixel-based image analysis is limited because of the following reasons: image pixels are not true geographical objects and the pixel topology is limited; pixel based image analysis largely neglects the spatial photo-interpretive elements such as texture, context, and shape; the increased variability implicit within high spatial resolution imagery confuses traditional pixel-based classifiers resulting in lower classification accuracies (Hay and Castilla 2006). Different from pixel-based method, OBIA works on (homogeneous) objects produced by image segmentation and more elements can be used in the classification. As an object is a group of pixels, object characteristics such as mean value, standard deviation, ratio, etc can be calculated; besides there are shape and texture features of the objects available which can be used to differentiate land cover classes with similar spectral information. These extra types of information give OBIA the potential to produce land cover thematic maps with higher accuracies than those produced by traditional pixel-based method. In this paper, we look at the performance of OBIA with different spatial resolution satellite images; comparing the classification results with those produced by the pixel-based method, we intend to find out how spatial resolution of satellite images influences the performance of OBIA. Experiment results showed that with the two sets of images of four different spatial resolutions, object based image analysis obtained higher accuracies than those of the pixel based one; with the increasing of the spatial resolution, the difference in accuracy values between object based and pixel based is decreasing. This paper showed that the object-based image analysis has advantage over the pixel-based one, and in accuracy rating, the advantage was better represented by higher spatial resolution satellite images.