中文摘要：

为解决传统的基于Harris角点的图像文字检测算法易受非文字角点干扰,检测准确率低的问题,提出一种基于多尺度Harris图像文字检测算法。该算法在多个尺度下提取角点,并利用分块方法分析文字局部特征,有效剔除了非文字角点。使用多次迭代逐步剔除非文字区域角点,精确提取备选块中的文字角点;通过区域融合形成文字区域,用轮廓跟踪法标识文字区域。实验结果表明,该算法明显提高了图像/视频文字检测的稳定性和准确率。

英文摘要：

To deal with the conventional character detection method upon Harris corner detection is easily effected by non-character corners.A novel approach of image/video character detection is presented,which extracts character corner by multiscale Harris and eliminates the non-character corner by blocking.The experiments show that the discussed method could significantly achieves the stability of the image character detection,and improves the higher accuracy and more robust to noise.