基于语义约束的口型序列识别方法  

A MOUTH-SHAPE SEQUENCE RECOGNITION METHOD BASED ON SEMANTIC CONSTRAINTS

作　　者：肖庆阳[1] 张金[1] 左闯[1] 范娟婷[1] 梁碧玮[1] 邸硕临[1] 

机构地区：[1]南开大学信息技术科学学院,天津300071

出　　处：《计算机应用与软件》2012年第9期226-229,284,共5页Computer Applications and Software

基　　金：南开大学创新科研计划"百项工程"项目(BX9-313)

摘　　要：口型识别是唇语识别的重要环节,由于中文相似口型较多,口型识别率往往不高。针对这一问题,提出一种基于语义约束的口型序列识别方法。该方法针对口型序列,利用语义约束和序列组合,分别对单一口型不易区分的情况和词语间隔不明显的情况进行处理,实现音节序列的识别,从而完成信息的转换。通过实例对该方法进行了实验验证,表明该方法技术可行,并基于此提出一种基于口型序列的唇语中文输入方法。The mouth-shape recognition is an essential part of lip language recognition.Due to considerable similarities of mouth shapes in Chinese,usually the mouth-shape recognition rate is low.Aiming at this issue,we bring forward a new mouth-shape sequence recognition method based on semantic constraints.In light of mouth-shape sequence,the method employs semantic constrains and sequence combination to deal with respectively the situation of single mouth shape which is hard to be distinguished and the situation of indistinctive intervals between the words.In this way,the recognition of syllable sequence is accomplished and the message translation is achieved.Experimental validation on the method has been carried out through examples,it proves that the method is technically feasible.Moreover,on this basis a lip language input approach for Chinese is suggested.

关 键 词：图像处理 口型序列 语义约束 特征提取 输入方法 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]