中文摘要：

摘要在视频字幕识别过程中，由于字幕像素比较低，使得二值化后，汉字边缘出现不规整的锯齿。针对该类汉字的细化，提出基于模板匹配及曲线拟合的方法。用角度为0。、45。、90。、135。的线段作为模板，分别匹配汉字中“横”、“撇”、“竖”、“捺”笔划，实现笔划提取；用二次Bezier曲线拟合“撇”、“捺”笔划，实现笔划重绘。对比实验表明，该方法对视频字幕的细化是有效的，且结果可以更好地被识别。

英文摘要：

In recognition process of video caption, irregular sawtooth will appear at the edge of the Chinese characters after image binarisation due to the generally low pixels of the caption. In this paper, a method based on template matching and curve fitting is proposed to realise the thinning of this kind of Chinese characters. In order to realise strokes extraction, we use 0, 45, 90and 135 line segments as the templates to match four kinds of elementary strokes in terms of horizontal, left-falling, vertical and right-falling strokes in Chinese characters re- spectively. And the quadratic Bezier curve is used to fit the left-falling stroke and right-falling stroke and to realise the redrawing of the strokes. Comparative experiment illustrates that this method is effective in video caption thinning, and the effect can be reeognised more precisely.