Method and apparatus for generating de-identified training set using backpropagation of convolutional neural network

摘要

A method and apparatus for generating an unidentified training set using backpropagation of a convolutional neural network are provided. The generating device generates a second label set from a neural network model already trained based on the first input image set and the first label set, and arbitrarily configures a second input image set for outputting a recognition result of the learned neural network model. Then, the second input image set is updated by performing backpropagation using the learned neural network model. Then, when the second label set and the updated second input image set satisfy the set condition, the backpropagation is terminated and the updated second input image set and the second label set are used as the de-identified training set.