A Fundamental Examination of Recognition of Manual Alphabets on Color Image
Yuji Nagashima, Masanori Fujii, Kazuyuki Kanda*,
Mina Terauchi**, Hideyo Nagashima
Kogakuin University, 2665-1, Nakano-machi, Hachioji-shi, Tokyo, 192 Japan
* Chukyo University, 101, Tokodate, Kaizu-cho, Toyota-shi, 470-03 Japan
** The Polytechnic University, 4-1-1, Hashimoto-dai, Sagamihara-shi, Kanagawa, 229 Japan
Recognition of the handshape is the indispensable step to recognize the JSL (Japanese Sign Language) which uses hands as a part of articulators. We start from recognition of Manual Alphabets (MA below) as a basis to understand the handshape of a sign. In our method considering its application to sign recognition at bust shot size image, we succeeded in obtaining a hand region of the bust shot color image (512 x 512 pixels) by transforming RGB system into HVC (Hue, Value and Chroma) system, and by cutting out automatically a region of a hand out of a region in skin color. The finger region was extracted out by the Erosion-Dilation method from the hand region and the features such as finger vector were obtained. We detected a direction of gradient direction and a figure of a contacted extended finger and used them as the features, which constructed a decision tree and made the MA recognized.
This article is published under a Creative Commons Attribution-NoDerivatives 4.0 Internationa License.