• 电话: 0531-88771456
  • 传真: 0531-80992585
  • 手机: 13698644688
  • 邮箱: 371162023@qq.com
  • 地址: 济南市历城区工业北路161-1号
        华政印务
新闻中心当前位置:首页-新闻中心
    中文字的识别两种方式
    来源: 2016-12-22 15:26:33
    1.米线结构模式识别是早期汉字识别研究的主要方法。其主要出发点是汉字的组成结构。从汉字的构成上讲,汉字是由笔划(点横竖撇捺等)、偏旁部首构成的;还可以认为汉字是由更小的结构基元构成的。
    1 line structure pattern recognition is the main method of early Chinese character recognition research. Its main starting point is the composition structure of Chinese characters. From the Chinese characters constitutes, Chinese characters by stroke (little did leave Na etc.), the radicals form; can also think Chinese characters is composed of smaller base structure element.
    由这些结构基元及其相互关系完全可以精确地对汉字加以描述,就像一篇文章由单字、词、短语和句子按语法规律所组成一样。所以这种方法也叫句法模式识别。
    From these structural elements and their relations can be accurately described as the Chinese characters, an article by words, words, phrases and sentences in the grammar rules as. So this method is also called syntactic pattern recognition.
    (1)利用变换特征的方法。对字符图象进行二进制变换(如Walsh, Hardama变换)或更复杂的变换(如Karhunen-Loeve, Fourier,Cosine,Slant变换等),变换后的特征的维数大大降低。但是这些变换不是旋转不变的,因此对于倾斜变形的字符的识别会有较大的偏差。
    (1) method of using transform features. Character image binary transform (such as Walsh, Hardama transform) or more complex transformation (such as Karhunen-Loeve, Fourier, Cosine, Slant transform, etc.), the dimensionality of the transformed features greatly reduced. But these transformations are not rotation invariant, so there is a large deviation in the recognition of the characters of tilt deformation.
    二进制变换的计算虽然简单,但变换后的特征没有明显的物理意义。K-L变换虽然从最小均方误差角度来说是最佳的,但是运算量太大,难以实用。总之,变换特征的运算复杂度较高,且有一定弱点。
    Although the calculation of binary transform is simple, the transformed feature has no obvious physical meaning. Although the K-L transform is optimal from the minimum mean square error angle, the computation is too large to be practical. In a word, the computational complexity of transform feature is high, and has some weakness.
    (2) 模板匹配。模板匹配并不需要特征提取过程。字符的图象直接作为特征,与字典中的模板相比,相似度最高的模板类即为识别结果。这种方法简单易行,可以并行处理;但是一个模板只能识别同样大小、同种字体的字符,对于倾斜、笔划变粗变细均无良好的适应能力。
    (2) template matching. Template matching does not require feature extraction. Compared with the template in the dictionary, the most similar template class is the recognition result. This method is simple and can be processed in parallel; but a template can only identify the same size, same font characters for tilt, strokes thicken fine had no good adaptability.
    本文来自:济南印刷厂更多的内容请点击:http://www.jinanyinshuachang.com我们将会为您提问的问题提供一个满意的服务,欢迎您的来电!
    This article from: Ji'nan printing plant more content please click: http://www.jinanyinshuachang.com we will ask questions for you to provide a satisfactory service, you are welcome to call!