Zero-shot Learning Method For Body Constitution Recognition Based On Tongue Image

In the history of Chinese medicine,body constitution is regarded as a form of innate and acquired talent in the process of human life.It is a comprehensive expression of physiological function and psychological state.The type of body constitution is highly related to some diseases and even determines the tendency of the disease,so that it is of great importance for treatment in clinical medicine.As an important and unique part in traditional Chinese medicine,tongue diagnosis is one of the most widely used diagnostic methods in TCM.This diagnosis method provides a simple,direct and convenient diagnosis method for clinical medicine.And tongue diagnosis is also an important basis for identifying constitution in Chinese medicine.This is because the color and texture features of tongue coating image reflect the health status of the patient.With the development of technology,especially artificial intelligence technology,the use of computer to collect and analyze tongue images is gradually recognized as an effective way to assist Chinese medicine doctors.In this paper,object detection algorithm is applied to detect the tongue area from the whole image,and only the main area of tongue is retained.In this paper,zero-shot learning method is utilized to learn the mapping function from tongue image to the semantic properties of tongue to solve the problem that caused by extreme imbalance of sample distribution and improve the classification accuracy of tongue constitution recognition.This paper focuses on improving the performance of tongue constitution recognition by zero-shot learning method.Multi-branch attribute group learning method is proposed to learn similar attributes by group.This paper also proposes a method of discriminative latent feature learning to regulate the inter/intra class distances between latent attributes features.To evaluate the performance of the proposed method,experiments are conducted on the tongue images dataset.The experimental results illustrate the proposed method is effective to improve the accuracy of constitution recognition and achieve fast,accurate and stable constitution recognition by tongue images.