A general evaluation method for segmentation algorithm based on experimental design methodology

Many techniques for segmenting images have been developed over the past decade in automatic target recognition (ATR) area. This paper reports an evaluation method for segmentation component in ATR system on the basis of experimental design methodology. A set of image metrics is developed to allow an objective evaluation. Response surface modeling approach is employed to analysis the performance of segmentation approaches on selected common data. The result of the test shows that only 27 frames tested images can result in good and reliable performance models, and the relative performance to be consistent with intuitive expectation.