Character recognition method for mapping skeleton information into image

The invention provides a character recognition method for mapping skeleton information into an image. The method relates to three processes of data preprocessing, feature extraction and classification. In the data pre-processing stage, Ki (three-dimensional) joint point coordinates are converted into defined joint point coordinate pixels through mathematical operation, the perspective problem of the coordinates is eliminated through the operation, data normalization is completed, and then a structure shaped like a Chinese character'da 'is designed to be used for layout of the joint point coordinate pixels. The generated articulation point coordinate image is used to preserve a natural connection relationship of the skeleton. In the feature extraction stage, a convolutional layer is utilized to extract a joint point coordinate image, and the operation can effectively extract static features in the eight directions of the skeleton. In the classification stage, a full-connection layer is used for character class