TY - GEN
T1 - A static hand gesture recognition model based on the improved centroid watershed algorithm and a dual-channel CNN
AU - Dong, Xude
AU - Xu, Yuanping
AU - Xu, Zhijie
AU - Huang, Jian
AU - Lu, Jun
AU - Zhang, Chaolong
AU - Lu, Li
N1 - Publisher Copyright:
© 2018 Chinese Automation and Computing Society in the UK - CACSUK.
PY - 2018/9
Y1 - 2018/9
N2 - In order to achieve static hand gesture recognization within complex skin-like background regions in an effective and intelligent manner, this study proposed an integrated hand gesture recognition model based on the improved centroid watershed algorithm (ICWA) and a dual-channel convolutional neural network (DCCNN) structure. The effectiveness of this approach stemmed from more accurate segmentation of hand gestures from an original image by using the ICWA. The segmented image and the corresponding Local Binary Patterns (LBP) features extracted from the original image then serve as inputs for two channels of the devised DCCNN respectively for classification. The contributions of this study included an innovative method for reducing the image gradient difference while segmenting in the YCrCb color space, and the fusion of both Principal Component Analysis (PCA) for dimension reduction and a convexity detection process for identifying the secant line between the palm and arm. The devised DCCNN enables significant improvement on the static hand gesture classification accuracy by employing independent dual-convolution neural network framework for dealing with richer features at different scales. Tests and evaluations on benchmarking databases demonstrated that the devised models and techniques outperform classic methods with distinctive advantages when operating under challenging skin-like background conditions.
AB - In order to achieve static hand gesture recognization within complex skin-like background regions in an effective and intelligent manner, this study proposed an integrated hand gesture recognition model based on the improved centroid watershed algorithm (ICWA) and a dual-channel convolutional neural network (DCCNN) structure. The effectiveness of this approach stemmed from more accurate segmentation of hand gestures from an original image by using the ICWA. The segmented image and the corresponding Local Binary Patterns (LBP) features extracted from the original image then serve as inputs for two channels of the devised DCCNN respectively for classification. The contributions of this study included an innovative method for reducing the image gradient difference while segmenting in the YCrCb color space, and the fusion of both Principal Component Analysis (PCA) for dimension reduction and a convexity detection process for identifying the secant line between the palm and arm. The devised DCCNN enables significant improvement on the static hand gesture classification accuracy by employing independent dual-convolution neural network framework for dealing with richer features at different scales. Tests and evaluations on benchmarking databases demonstrated that the devised models and techniques outperform classic methods with distinctive advantages when operating under challenging skin-like background conditions.
KW - Centroid watershed algorithm
KW - Dual-channel convolution neural network
KW - Hand gesture recognition
KW - Hand gesture segmentation
UR - http://www.scopus.com/inward/record.url?scp=85069216991&partnerID=8YFLogxK
U2 - 10.23919/IConAC.2018.8749063
DO - 10.23919/IConAC.2018.8749063
M3 - Conference Proceeding
AN - SCOPUS:85069216991
T3 - ICAC 2018 - 2018 24th IEEE International Conference on Automation and Computing: Improving Productivity through Automation and Computing
BT - ICAC 2018 - 2018 24th IEEE International Conference on Automation and Computing
A2 - Ma, Xiandong
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 24th IEEE International Conference on Automation and Computing, ICAC 2018
Y2 - 6 September 2018 through 7 September 2018
ER -