人工神经网络的直线图素识别质量判别器

Using Neural Network to Verify Recognition Quality of A Line

讨论矢量化直线图素识别质量的判别方法 ,采用BP网络 ,网络输入为反映直线图素识别质量的特征因子 ,网络输出为从斜率、线宽及直线端点的定位精确性等方面对直线图素识别质量的评价 ,BP网络经训练成为直线图素质量判别器 由于输入到网络中的各特征因子均是与线宽的相对比值 ,因此 ,该方法对扫描分辨率的影响不敏感 。

The evaluation process can be divided into two parts: the first part is to extract factors reflecting the quality of a line, such as slope, width, location of end points; the second part is to put these factors into an evaluator for analyzing. A BP neural network with a hidden layer was used. After training the BP network became a quality evaluator for lines and output the evaluated results on slope, width and end point's location of a line. This approach is not sensitive to scanning resolution and easy to detect poor quality lines. The method can be readily extended to verify other types of graphic elements.