Research On Height Prediction Method Based On Barefoot Footprint Image

Footprint image-based identification technology is a new type of biometric identification technology.Medical research shows that the human footprint is unique,which is determined by the nuances of human innate bones and the different living habits formed in the day after tomorrow.In addition,the footprint data collection is simple,hidden and does not require intentional cooperation.Therefore,more and more researchers pay attention to the study of footprint data.In this thesis,the footprint image is automatically marked,and the geometric features are extracted.The relationship between the geometric features and the height of the footprint image is studied using the traditional meaning regression analysis method and the convolutional neural network-based analysis method.The main contents of the thesis are as follows:1)A height regression analysis method based on the footprint characteristics of footprint images was proposed.A variety of features extracted from the plot were screened.At the same time,the fusion features were tested with scatter plots,statistics related theory,linear regression methods,and non-linear regression methods.The results show that the length of the foot GJ(GJ indicates the Euclidean distance between the most prominent point of the leading edge of the barefoot toe and the most prominent point of the trailing edge of the barefoot)is related to the height.Significant correlation,in addition to FJ(FJ represents the Euclidean distance from the most prominent point of the barefoot to the most prominent point of the trailing edge of the barefoot),BJ(BJ indicates the Euclidean distance from the most prominent point of the barefoot to the most prominent point of the trailing edge of the barefoot),BF(BF indicates the Euclidean distance from the most prominent point of the outer edge of the red-footed owl)and height have a strong correlation.CE(CE indicates the Euclidean distance from the most prominent point of the barefoot outer edge to the most prominent point of the barefoot inner edge)and height have a moderate degree of correlation.2)Regression prediction of height was performed using the improved convolutionalneural network.The algorithm first preprocesses the footprint image.Secondly,in order to realize the regression prediction of height,according to the characteristics of the actual problem,the regression layer is used instead of the softmax layer,and the last softmax layer improved by the convolutional neural network classification network is changed to the corresponding regression layer network.Finally,according to the obtained footprint image data set with the modified regression network,a good regression model was obtained after tuning,training,and testing.Experiments show that the algorithm has greatly improved the generalization and robustness compared with traditional height prediction methods,and also has a higher accuracy rate of height prediction.