Data Generation and Its Validity Inspection of Signer-Independent Sign Language

Abstract

This paper proposes the combination of sign language linguistics with human kinematics to generate and detect the data of SISL (signer independent sign language) according to the characteristics of gesture sign language (GSL). An improved Mean-Shift algorithm is applied to the generation of hand shape data channels without losing the linguistic features of GSL, and then the key hand shape phonetic notation is used to detect the effectiveness of data. In order to enrich the kinematic characteristics of GSL, an improved genetic algorithm is applied to the generation of movement related data channels. Moreover, Labannotation is adopted to inspect the effectiveness of data. Finally, an experimental inspection framework is established based on an original sample to make the proposed detection method adapt to multi-classes data inspection of linguistics. Experimental results show that the proposed method for the generation and detection of SISL data is effective and feasible.