视觉乐谱图像动态多阈值二值化方法  

Dynamic Multi-threshold Binarization Method for Visual Music Images

作　　者：杨芳[1,2] 滕桂法[1] 田学东[2] 

机构地区：[1]河北农业大学机电工程学院,保定071001 [2]河北大学计算机科学与技术学院,保定071000

出　　处：《计算机科学》2016年第1期310-314,共5页Computer Science

基　　金：国家自然科学基金(61375075);河北省自然科学基金(F2011201106;F2013201134)资助

摘　　要：针对视觉乐谱图像在不规则区域内光照不均匀的二值化问题,提出了一种对前景内容动态分块二值化的方法,以降低背景噪声对二值化效果的影响,提高分块多阈值方法的有效性。该方法以像素灰度差分为基础,以滑动窗口为处理对象,根据窗口内差分变化率确定前景区域位置;再根据前景区域背景像素灰度分布情况将窗口分类,从而划分出不规则光照区域;最后在每个区域内分别采用Otsu阈值算法进行二值化。在实验中详细讨论了窗口大小和差分变化率阈值设置对前景区域提取正确率的影响,结果表明所提出的方法对乐谱图像前景内容的查全率高,能划分出不规则形状的范围,可以有效处理光照不均匀情况,并且可以用于一般文档的二值化。A dynamic binarization method was proposed for solving the uneven illumination problems in irregular regions of visual music images,to reduce the background influence and improve the effectiveness of multi-threshold method.This method is based on pixel grayscale difference,and slide window is the segment unit.The foreground is detected by the difference variance ratio in each window,and the irregular illumination regions are decided by window background gray estimation.The Otsu＇s binarization threshold is employed in each region in this algorithm.Foreground detection accuracy is decided by window size and variance ratio threshold.Those two parameters were discussed in detail in experiment.The experimental results show that this method can divide foreground into irregular illumination regions with a high foreground recall ratio.And it is effective to solve the uneven illumination binarization problem of music images.This method can also be applied to the binarization of general document images.

关 键 词：二值化 乐谱图像 差分 前景区域检测 动态阈值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]