手語辨識技術係由手部影像偵測與手語影像識別所組成。手部影像偵測包含: 如何將手掌影像轉正,及手腕與掌影像之切割分離,均為重要的手語影像前處理技術。本文首先呈現手掌影像自動轉正與手腕影像自動切除之手部影像前處理技術。若歪斜之手掌影像無法轉正,及手腕部位切割不正確,都會降低手語辨識之成功率。本文所提出之方法可自動地將歪斜之手掌影像轉正,並切除手掌以外的影像。根據此方法可提升手臂影像排除正確率平均達94%。 手語辨識技術主要在於辨識手指的變化。本文創立手語影像之斜線與橢圓編碼,並採用高斯混和機率分佈技術建立手語模型並進行辨識。斜線編碼的優點係可區別手指個數之變化,唯因手語建模樣本之遠近角度各有不同進而影響手語模型的辦識率。為改善此缺點,本論文進一步設計出橢圓編碼。此方法可更精準地區別手指在方位上之變化,且不會因手語樣本角度不同而影響辨識率。根據實驗,相較與斜線編碼,橢圓編碼在辨識正確率上平均可提昇84.6%。
Sign language identification and recognition technique is composed by the hand images detection and the hand gestures recognition. Hand images detection is locating the palm and fingers part from the sensed image, and rotating them to the appropriate hand posture, both are the important pre-processing for sign language identification and recognition. Lose those, the correctness rate of the sign language recognition algorithms will be dropped down to an unacceptable level. This paper presents a novel hand images detection processing algorithms, it rotates an oblique gesture to right position, and to delete the elbow and forearm parts from the sensed image. According to the newly hand images detection processing, the average recognition rate is improved about 94%. The major work in the hand gestures recognition is to identify the variance of the fingers. In this paper, two novel image coding techniques with slash lines and oval circles styles are presented to catch the palm and fingers’ gesture features. Following this coding, the mixture Gaussian probability distribution is used to modeling those gesture features. According to the final experiment, the recognition accuracy under the oval circles coding is improved 84.6% in average, there is based on the perform comparison between the slash lines and oval circles coding.