Fusion Recognition of Color Images Quantified by Multi-feature of Electronic Handwriting Based on CNN Attention Mechanism
In order to extract effective features of electronic handwriting to achieve recognition tasks,it is proposed that various features of electronic handwriting are quantified by color to form the dot matrix y-x representation map of handwriting dat,x/y/p-t representation map of coordinate value and pressure value of p with time changing,8 direction feature map of handwriting,pen wielding velocity feature map and pen pressure velocity feature map.Then a 5-layer convolutional neural network(CNN)is constructed,and a 2-dimensional attention mechanism algorithm is proposed to capture cross-channel feature interac-tion information to increase the degree of feature aggregation.Finally,the multi-feature maps of electronic handwriting are used simultaneously as the input data of the CNN network,and the fusion recognition is carried out at different network layers.The experimental results show that the proposed method can signif-icantly accelerate the network convergence speed and improve the recognition accuracy and robustness.The recognition accuracy rate for 12 people can reach more than 95％,and the accuracy rate for 50 peo-ple can reach more than 92％.The visual graphs of multi-features of electronic handwriting can be used to assist handwriting identification,and the fusion recognition method can avoid the risk of personal priva-cy disclosure.
NETL
NSTL
万方数据
