基于数据生成的手语识别自适应方法

An adaptive Chinese sign language recognition method based on data generating

采用自适应技术来解决非特定人手语识别问题,提出了一种基于数据生成的手语识别自适应方法。首先,对非特定人模型的均值向量进行自动聚类,寻找聚类中心生成手语词码本,然后,根据此码本选择词根子集,该子集能覆盖手语词码本的所有码字,继而,使用新用户的词根子集数据通过遗传算法生成其它词根的数据,最后,结合词根子集的真实数据和其它词根的生成数据,利用最大似然线性回归(MLLR)和最大后验概率(MAP)算法对非特定人模型进行自适应。实验结果表明,该方法既能够降低所需要的自适应数据量,又能够在非特定人模型基础上取得识别正确率的大幅提高。

This paper proposes an adaptive Chinese sign language recognition method based on data generating to solve the signer-independent （SI） problem using the adaptation technology. The method is described as below. First, the SI models＇ means are automatic clustered, and then the cluster centers are taken as codebooks of all etyma. A subset of all etyma is selected to cover all codewords. The data of the subset for a new signer are used to generate the data of other etyma with the genetic algorithm. By utilizing the original data and generated data with the algorithms of Maximum Likelihood Linear Regression （MLLR） ＆ Maximum a Posteriori （MAP）, SI models＇ means are adapted to the new signer. The experimental resuhs show that this method can both decrease the adaptation data and increase the recognition rate compared with the SI models.