基于决策树分析的激光打印文件黑色墨迹量化分析研究

摘要

ABSTRACT:ObjectiveTo explore an assay to classify short-term printing samples produced with different laser printers using ImageXaminer and quantitative analysis.Methods360 pages of short-term printing sample from 18 different laser printers were collected and 116 features, such as area, perimeter, gray scale, were scanned and measured by the ImageXaminer. Samples of different laser printers were input at the resolution of 800dpi and full-page measured by choosing the same-size target region. The corresponding mathematical model was established based on decision tree algorithm C4.5, and the measured values of 116 parameters were analyzed to ifnd out the appropriate ones which had a strong correlation to the classiifcation of different laser printers.ResultsShort-term printing samples of different laser printers could be correctly classified according to the comprehensive analysis of the following 11 parameters: median gray level threshold of 128, mode gray threshold of 80, Stat. Area Stedv threshold of 117 and the minimum area of 1 pixel, area coverage ratio threshold of 80, Stat. Area Stedv threshold of 160 and the minimum area of 1 pixel, median gray level threshold of 80, Stat. roundness Stdev threshold of 80 and the minimum area of 1 pixel, distribution symmetry threshold of 128, area coverage ratio threshold of 117, median gray level threshold of 160, Stat. roundness Average threshold of 117 and the minimum area of 1 pixel. These 11 parameters have quite strong correlation to the classiifcation of different laser printers.ConclusionsIt is feasible to make a distinction between short-term printing samples produced with different laser printers by using the quantitative analysis method of the ImageXaminer.%目的：研究利用图像扫描分析系统量化分析激光打印文件黑色墨迹的方法，优选可用于打印机种类区分的量化参数。方法使用图像扫描分析系统，采用每英寸800像素点的分辨率，对市场上9个主流品牌18种型号共18台激光打印机360页打印样本的面积、周长、灰度等特征进行整幅测量。通过决策树C4.5算法建立数学模型，对测量所得参数值进行数据分析。结果发现阈值分别设置为80、117、128、160时，测量所得的29项参数116组测量数值中阈值为128的中位灰度值，阈值为80的灰度众数值，阈值为117、最小测量面积为1的字块面积标准差，阈值为80的面积比值，阈值为160、最小测量面积为1的字块面积标准差，阈值为80的中位灰度值，阈值为80、最小测量面积为1的字块圆度标准差，阈值为128的对称分布值，阈值为117的面积比值，阈值为160的中位灰度值，阈值为117、最小测量面积为1的字块圆度平均值等11组参数组合能够将样本正确区分。结论不同种类激光打印机打印样本在上述11组参数所衡量的灰度、面积、圆度、打印墨迹数量等特征及特征的组合上存在差别，通过量化方法区分激光打印机打印样本具有可行性。