A Feature Analysis Approach for Object-Oriented Classification

A good feature analysis is prerequisite for successful object-oriented classification in image analysis.In most cases,feature analysis is performed empirically with the method of trial and error in which certain features and thresholds are tested.Limited by time,it is impossible to examine all features.Therefore,the selected thresholds are not statistically optimized,and no conclusion can be drawn as to the quality of the features.A new method known as SaT has been developed in this paper.With SaT,it is possible to extract not only characteristic features but also the associated thresholds for any number of object classes from a large number of object features available.The separability,the result of SaT analysis,serves as a comparative measure for classification quality at the same time.Conducted experiments have verified that this approach improves the precision and speed of object-oriented classification.