Patent attributes
An object of the present invention is to provide an image processing method and an image processing apparatus that make it possible to efficiently learn images having different identities. In learning and recognition using a hierarchical network, it is known that, based on experiences, a layer near the input functions as a feature extractor for extracting a feature that is necessary for recognition, and a layer near the output performs recognition by combining extracted features. Thus, performing learning by setting a higher learning rate to a layer near the input side of the hierarchical network than a learning rate in a layer near the output side in second learning processing as in an aspect of the present invention corresponds to mainly relearning (adjusting) a feature extraction portion in data sets having different identities. Accordingly, the difference between data sets can be absorbed, and learning can be performed more efficiently than in the case of simply performing transfer learning.
