Tooth-marked Tongue Recognition based on Wavelet Transform and Feature Fusion

Currently, convolutional neural networks are widely used in Chinese medicine tooth-scarred tongue recognition, but convolutional neural networks lack representation of local details. To extract the texture features of the dentate tongue more comprehensively, a non-downsampling wavelet transforms combined with AR-LGC operator is proposed as a multi-feature fusion method for the recognition of dentate tongue in Chinese medicine. The non-downsampled wavelet transform is used to decompose each input image at multiple scales, and the AR-LGC coding algorithm is used to retain key texture information in the decomposed sub-map and remove redundant information. The inverse wavelet transform is then used to reconstruct the processed decomposed subgraphs to obtain a new fused image, and finally, the fused image is fed into the convolutional neural network ResNet34 for training and classification. The experimental results show that the proposed method can achieve better recognition accuracy than other convolutional neural network methods for identifying dentate tongues.