复杂光照条件和文字—背景的交融是自然场景图像中文字检测的主要难点。为解决该问题，提出了基于LHBP(local Haar binary pattern)多尺度向性滤波的文字检测算法。该算法首先采用对光强变化不敏感并具文字特征显式描述特点的LHBP模式的纹理描述算子；并在LHBP模式上采用多尺度向性滤波器MDF(multi-scale directional filtering)来确定候选文字区域；最后使用基于LHBP直方图的支持向量机法精确定位文字区域。实验结果表明，与其他主流算法相比，该算法能够去除复杂光照条件和文字—背景交融的影响，具有更好的性能。

The main difficulties for text detection are due to complicated illumination condition and text-background weak contrast in natural scene images. In order to resolve them, this paper presents a local Haar binary pattern(LHBP) based algorithm for text detection. Firstly, LHBP texture description operator is presented. It is insensitive to illumination variance and can effectively describe the text feature. Secondly, the Multi-scale Directional Filtering based on LHBP is proposed for fast filtering to obtain candidate text regions. Finally, the LHBP-histogram-based SVM is presented to refine the text location. Comparing with state-of-the-art algorithms, the experiment results demonstrate the robustness of the proposed method with better accuracy.