Uncertainty Measures for Assessing the Attribute Accuracy of Objected-Based Classification of Remotely Sensed Imagery

Traditional approaches for accuracy assessment are inadequate for object-based image classification map. We transform some measures to assess the accuracy of object based image classification result. The measures are based on error matrix theory and using polygons as sample units, because the pixel sample units are not suitable for assessing the accuracy of object-based classification result. Compared to pixel samples, we realize that the uniformity of polygon samples has changed. In order to make the indexes generating from error matrix reliable, we using the areas of polygon samples as the weight to establish the error matrix of object-based image classification map. We compare the result of two error matrixes setting up by the number of polygon samples and the sum of area of polygon samples. The error matrix using the sum of area of polygon sample is proved to be an intuitive, useful technique for reflecting the actual accuracy of object-based imagery classification result.