A study of Mixed-Mode Communication based on Word-Unit Processing and Graphic Commands

本文提出了一种利用模式识别技术对英文文本进行编码的算法。首先，利用加权传播和收缩操作将每一页文档分割成包含文本、表格、图形等特定内容的块。对于文本块，使用26个前缀的字典进行单词识别，338个词根和24个后缀，然后将识别出的单词编码成数字信号。将待识别字编码成ISO码作为信息交换字符，对表格块采用Hough变换提取直线，并通过图形命令传递直线属性。表格块中的符号和文本块一样被编码成数字信号，图形块中的直线再次通过Hough变换提取，并通过模式匹配识别出6种类型的符号：*、*。线和符号元素的属性也通过图形命令传输。在去除行和符号之后，剩余的图像通过具有行跳过扫描的MMR编码方案进行编码。跳过行的数目是从0、1或3中选择的，参考恢复的图像上的错误像素的数目。为了传送文档，编辑指示文档轮廓、布局对象、本文给出了英文文本和黑白文本传输的计算机模拟结果，白色图像显示，处理后的信号相对于传统传真信号的总压缩比为124.9。对于文本块，采用字单元编码比按字符编码的压缩比提高了1.91倍。

The encoding algorithm of English texts using pattern recognition technique is proposed.First, each page of document is segmented into blocks which contain specific kind of contents such as text, tables, graphs etc. by the weighted propagation and shrinking operation.For the text blocks, words identification is carried out using dictionaries of 26 prefixes, 338 roots and 24 suffixes and then the identified words are encoded into digital signals. Unidentified words are encoded into the ISO code for information interchange character.For the table blocks, straight lines are extracted by the Hough trans form and the attributes of lines are transmitted with the graphic commands. The remainders in the table blocks are encoded into digital signals just like as the text blocks.For the graphic blocks, straight lines are again extracted by the Hough transform and symbols of six types: *, *, *, *, *, * are identified by pattern matching. The attributes of lines and symbol elements are also transmitted by the graphic commands. After the lines and symbols are removed, the remaining picture is encoded by the MMR coding scheme with line-skipped scanning. The skipped lines are restored using the preceding and following lines.The number of skipped lines are selected out of 0,1 or 3 referring the number of error pixels on the restored picture.To transmit the documents, editing information indicating document profile, layout object, presentation function etc. is added to the encoded signals.The computer simulation results for transmission of English documents including texts and black-and-white images show that the overall compression ratio of the processed signal to the conventional facsimile signal is 124.9.With regard to the text blocks, by adopting the word-unit encoding, 1.91 times of compression ratio improvement is achieved compared with the encoding by characters.