结合颜色和MGD特征及MRF模型的场景文本分割

针对场景文本受到光照、复杂背景等因素影响而难以进行有效分割的问题,提出了一种融合颜色和最大梯度差(MGD,maximum gradient difference)特征及马尔科夫随机场(MRF,Markov random field)的场景文本分割方法。首先提取能够有效表达文本纹理特性的MGD特征,通过概率框架将其和颜色特征结合起来对观测图像进行建模;然后结合空间关系和邻域像素属性差异对传统势函数进行改进;最后建立场景文本分割的MRF模型,利用图割(graph cut)算法快速地求解该模型。实验结果表明,采用颜色和MGD特征相结合以及改进的势函数对分割结果具有较大地改善,尤其在光照不均匀及背景复杂情况下相比其他算法取得了较好的性能。

Segmentation of scene text image using color and MGD feature and MRF model

An effective algorithm for segmentation of scene text based on color, maximum gradient difference (MGD),and Markov random field (MRF) model is proposed to overcome the interfere nce of uneven illumination and clutter background to scene text segmentation.Firstly,in view of the compl exity of scene text,the effective texture feature MGD is extracted,and is combined with color feature to make a model for observed image through a probability framework.Secondly,potential function is improved by tak ing advantage of the spatial relationship and attribute difference between neighborhood pixels.Then,an MRF m odel is proposed for the problem of scene text segmentation,and graph cut method is adopted to solve the model quickly under the criterion of minimum energy.Experimental results demonstrate that the proposed method performs well and can achieve better segmentation results compared with other methods,especially in c ase of uneven light and complex background.