影像分割精度评价方法初探

Image segmentation, is the first step in object-based image analysis, the accuracy of image segmentation has decisive affects on the result of subsequent classification. At present, the accuracy of image segmentation evaluation methods can be divided into two kinds of supervised and unsupervised. The supervised method is take segmentation date compared with reference date for discrepancy analysis. This paper proposed a segmentation accuracy assessment method based vector-raster overlay analysis; this can avoid the errors in calculation caused by the date from different date sources. At the end of the paper calculated and compared four commonly used evaluation show that this method can more accurately calculate the area discrepancy between segmentation date and reference date, and improve the accuracy of the evaluation.