Domain adaptation-based object recognition apparatus and method

A domain adaptation-based object recognition apparatus comprising:

• a data collector processor configured to collect a face image;
• a generative model learning processor configured to learn a style of a gallery face image;
• a classification model learning processor configured to learn a classification model by using personal information requiring previous registration for performing face recognition and matching;
• and
• a recognition verifier processor configured to perform recognition on a real input face image by using the generative model and the classification model,
• wherein the data collector processor performs preprocessing on the face image by using feature point information and builds a gallery face image database,
• wherein the classification model learning processor performs error calculation on the basis of a face identification (ID) classification result based on preprocessing and feature extraction using the gallery face image database to allow a face ID classifier to repeatedly perform learning while simultaneously learning the style of the gallery face image and learning the generative model without losing ID information,
• wherein the generative model learning unit processor learns a face image generative model by using the gallery face image database and an external face image database, and
• wherein the generative model learning unit processor determines whether an input image is included in the gallery face image database, generates a new face image similar to a style of a learned gallery face image, and determines identification (ID) of the input image to learn the face image generative model.

A domain adaptation-based object recognition method comprising:

• step (a) of collecting an object image;
• step (b) of learning a style of a gallery image to learn a generative model;
• step (c) of learning a classification model by using personal information requiring previous registration for object recognition;
• and
• step (d) of recognizing an object included in an image by using the generative model and the classification model,
• wherein the step (a) comprises performing preprocessing on the object image by using feature point information and building a gallery image database,
• wherein the step (b) comprises learning the generative model for generating a new image or feature, determining whether an input image is included in the gallery face image database, matching a style of the gallery image, from an input image by using a gallery database and an external image database, and
• wherein the step (c) comprises performing preprocessing and feature extraction by using a gallery image database and performing error calculation on the basis of a face identification (ID) classification result to allow a face ID classifier to repeatedly perform learning while simultaneously learning the style of the gallery face image and learning the generative model without losing ID information.