Face image de-identification apparatus and method
AssigneeElectronics and Telecommunications Research Institute
InventorsJong-Wook Han, Geonwoo Kim, So Hee Park, Seon Ho Oh
Life of this patent
Estimating the term…
Abstract
A face image de-identification apparatus and method are disclosed. The face image de-identification apparatus may include an emoji generator configured to generate a first emoji image including facial feature information corresponding to a face image using the face image stored in a database, and an image inserter configured to insert the first emoji image into the database by replacing the first emoji image with the face image.
Claims11
A face image de-identification apparatus comprising a processor configured to implement:
- an emoji generator configured to generate a first emoji image including facial feature information corresponding to a face image stored in a database;
- and
- an image inserter configured to insert the first emoji image into the database by replacing the face image with the first emoji image, wherein:
- the emoji generator comprises an emoji conversion deep learning model for generating the first emoji image using the face image,
- a second emoji image is generated by applying an emoji conversion deep learning network to a reference face image, and
- the emoji conversion deep learning model is constructed by learning the emoji conversion deep learning network so that a similarity degree between facial feature information corresponding to the second emoji image and facial feature information corresponding to the reference face image is higher than a predetermined threshold value.
A method for de-identifying a face image by a face image de-identification apparatus, the method comprising:
- constructing an emoji conversion deep learning model to generate a first emoji image including facial feature information corresponding to the face image;
- generating the first emoji image by applying the emoji conversion deep learning model to the face image;
- and
- replacing the face image with the first emoji image, where, the constructing the emoji conversion deep learning model comprises:
- generating a second emoji image by applying an emoji conversion deep learning network to a reference face image;
- extracting facial feature information corresponding to the second emoji image;
- extracting facial feature information corresponding to the reference face image;
- calculating a similarity degree between the facial feature information of the second emoji image and the facial feature information of the reference face image;
- and
- learning the emoji conversion deep learning network until the similarity degree becomes higher than a predetermined threshold value.
A face image de-identification apparatus for de-identifying a face image, the face image de-identification apparatus comprising a processor configured to implement:
- an emoji generator configured to generate an emoji image by applying an emoji conversion deep learning model to the face image where the emoji conversion deep learning model is preliminarily learned so that a similarity degree between facial feature information of the face image and facial feature information of the emoji image is higher than a predetermined threshold value;
- and
- an image inserter configured to insert the emoji image into a database by replacing the face image with the emoji image.
Dependencies are inferred from the claim text.