乘性噪声文本图像二值化的自适应演化模型

An adaptive evolutionary model for binarization of document image with multiplicative noise

针对乘性噪声文本图像二值化,提出一个以偏微分方程形式存在的新模型.模型包含数据保真项、二值归类项以及正则化项,通过引入边缘停止函数对数据保真项和二值化项进行加权处理,可使前两项根据图像自身特征自适应地对原图像进行保真或二值化,而正则化项用于抑制乘性噪声.模型采用显式有限差分法进行数值求解,并对DIBCO公开数据集中具有典型代表的文本图像进行大量测试.实验结果表明,模型对乘性噪声文本图像的二值化是有效的,同时与相关方法进行比较,总体上实现了最佳的二值化性能.

A new partial differential equation model is proposed for document image binarization with multiplicative noise.The model contains data fidelity term,binary classification term and regularization term,in which the former two terms are weighted by introduction of edge stop function,making them adaptive to the original image according to its own characteristics of the fidelity or binarization of the original image,while the regularization term is used to suppress multiplicative noise.The model is solved numerically by explicit finite difference method,and many typical document images in DIBCO public data set are tested.Experimental results show that the proposed model is effective for binarization of document images with multiplicative noise,and gives the best binarization performance on the whole when compared with other relevant methods.