基于不变矩理论的字符串图像识别方法研究

Study on the recognition of character based on projection moment invariability

针对字符串的特点,提出了采用基于旋转投影不变矩来识别字符的方法.先将出现偏斜的字符串进行旋转,使其与水平方向平行,然后将字符的灰度图像从上向下进行投影,对黑点像素进行累计求和,构成灰度累计分布图.该分布图具有平移和尺度放缩不变性的特点,因此可以将每个英文字母的灰度累计分布图作为一个几何形状,使用Hough变换对它们进行识别.文中对字符串自动找正、灰度累计分布图的生成、自动阈值选择技术以及Hough变换等关键技术进行了研究,并给出了26个英文字符的灰度累计分布图和投影不变矩值.经实际使用证实该方法的效果很好.

A character string recognition method using projection moment invariability is introduced. The character strings are first rotated to the horizontal direction and then projected in the vertical direction. The total numbers in each column is finally summed up to act as the amplitude of the projection histogram. It is invariant to translation, scaling and orientation and can be used to recognize the shape of characters. The key transfer techniques, including maximal square variance thresholding and character matching, are also analyzed. Experimental results indicate that the method is effective.