Abstract: Text location is the premise and foundation of text extraction in images. In order to overcome the complex background and the effect of illumination, a coarse-to-fine text location algorithm is proposed. The algorithm firstly uses connected-component analysis for coarsely locating on the edge image, and then extracts histogram of oriented gradient feature and modified local binary patterns feature to classify the candidate regions, removes the false text to achieve accurate location. Experimental results indicate that this algorithm can effectively reduce the influence of non-uniform illumination and complex background, accurately locate the text area in scene image.

Key words: text location, connected-component analysis, histogram of oriented gradient feature, local binary patterns feature

摘要： 文本定位是图像中文本提取的前提与基础。针对场景图像中背景复杂和光照影响，提出一种由粗略到精确的文本定位算法。该算法首先在边缘图像上利用连通区域分析进行粗略定位得到文本候选区域，然后提取候选区域的方向梯度直方图特征和改进的局部二值模式特征进行分类，去除虚假文本达到精确定位。仿真实验结果表明，该算法能够有效地降低背景复杂与光照不均的影响，在场景图像中准确地定位文本区域。

关键词: 文本定位, 连通区域分析, 方向梯度直方图特征, 局部二值模式特征