Method and apparatus for obtaining first and second images to detect first and second landmarks and face registration using 3D facial landmark

Patent metadata

Field	Value
Publication number	US-12536677-B2
Application number	US-202318361744-A
Country	US
Kind code	B2
Filing date	Jul 28, 2023
Priority date	Aug 2, 2022
Publication date	Jan 27, 2026
Grant date	Jan 27, 2026

Abstract

Provided is a face registration method performed by a face registration apparatus, the face registration method including obtaining first face image data and second face image data that are captured at different angles, sizes, or locations, detecting a first landmark for the first face image data and a second landmark for the second face image data, calculating the amount of movement and distance between the first and second landmarks and setting a plurality of reference landmarks on the basis of the amount of movement and the distance, and matching the first and second landmarks with each other by adjusting a position, angle, and size on the basis of the plurality of reference landmarks.

First claim

What is claimed is:

1 . A face registration method performed by a face registration apparatus, comprising:
(a) obtaining first face image data and second face image data that are captured at different angles, sizes, or locations; (b) detecting a first landmark for the first face image data and a second landmark for the second face image data; (c) calculating an amount of movement and distance between the first and second landmarks, and setting a plurality of reference landmarks on the basis of the amount of movement and the distance; and (d) matching the first and second landmarks with each other by adjusting a position, angle, and size on the basis of the plurality of reference landmarks.

2 . The face registration method of claim 1 , wherein (b) comprises detecting the first landmark for the first face image data and the second landmark for the second face image data by applying a deep learning algorithm to the first face image data and the second face image data, wherein the deep learning algorithm corresponds to an high-resolution network (HR-net), a convolutional neural network and conditional random field (CNN-CRF), or Mediapipe, and
the first and second landmarks each comprise a plurality of facial landmarks and a plurality of eye landmarks, wherein the plurality of facial landmarks and the plurality of eye landmarks are managed by assigning a series of numbers thereto, and the same number is assigned to a landmark corresponding to both the first and second landmarks.

3 . The face registration method of claim 1 , wherein (c) comprises:
dividing the first and second landmarks into an upper part and a lower part, and dividing the upper part into an upper side, a lower side, a right side, and a left side; calculating an amount of movement corresponding to a distance between a pair of corresponding landmarks on the upper part of the first and second landmarks in a three-dimensional (3D) space; and selecting a plurality of landmarks from each of the upper, lower, right and left sides of the upper part of the first and second landmarks, starting from a smallest amount of movement.

4 . The face registration method of claim 3 , wherein (c) comprises:
calculating a distance between landmarks included on the upper and lower sides of each of the first and second landmarks and a distance between landmarks included on the right and left sides of each of the first and second landmarks among the plurality of landmarks selected from among the first and second landmarks on the basis of the amount of movement; and determining a first reference landmark for the first landmark and a second reference landmark for the second landmark by setting, as an upper reference landmark and a lower reference landmark, a pair of landmarks spaced a largest distance from each other among the landmarks included on the upper side and the landmarks included on the lower side and setting, as a right reference landmark and a left reference landmark, a pair of landmarks spaced a largest distance from each other among the landmarks included on the right side and the landmarks included on the left side.

5 . The face registration method of claim 4 , wherein (d) further comprises:
moving coordinates of one of the reference landmarks among the first and second reference landmarks to coordinates of an origin; and adjusting positions of the first and second landmarks by moving coordinates of the first landmark by an amount of movement of the coordinates of the first reference landmark to the coordinates of the origin and moving coordinates of the second landmark by an amount of movement of the coordinates of the second reference landmark to the coordinates of the origin.

6 . The face registration method of claim 5 , wherein (d) further comprises:
generating a first connection line connecting the right and left reference landmarks for each of the first and second reference landmarks, and generating a second connection line by vertically connecting a center of the first connection line and the upper reference landmark; calculating an angle between the first and second connection lines and each axis of the 3D space; and adjusting angles of the first and second landmarks by rotating the first and second landmarks about each axis on each plane through transformation of Euler angles on the basis of the calculated angle.

7 . The face registration method of claim 5 , wherein (d) further comprises:
calculating a width and height of each of the first and second reference landmarks corresponding to a distance between the right and left reference landmarks and a distance between the upper and lower reference landmarks for each of the first and second reference landmarks; and adjusting a size of the second landmark according to a size of the first landmark on the basis of a ratio between the width of the first reference landmark and the width of the second reference landmark and a ratio between the height of the first reference landmark and the height of the second reference landmark.

8 . The face registration method of claim 1 , further comprising visualizing the first and second landmarks by setting colors thereof using a heat map based on a distance between a pair of corresponding landmarks in the first and second landmarks in a three-dimensional (3D) space before or after the matching of the first and second landmarks.

9 . A computer program stored in a non-transitory computer-readable storage medium, wherein, when instructions of the computer program are executed, the method of claim 1 is performed.

10 . A face registration apparatus comprising:
an image data obtainer configured to obtain first face image data and second face image data that are captured at different angles, sizes, or locations; a landmark detector configured to detect a first landmark for the first face image data and a second landmark for the second face image data; a reference landmark setter configured to calculate an amount of movement and distance between the first and second landmarks and set a plurality of reference landmarks on the basis of the amount of movement and the distance; and a landmark matching part configured to match the first and second landmarks with each other by adjusting a position, angle, and size on the basis of the plurality of reference landmarks.

Assignees

• Univ Kwangwoon Ind Acad Collab

Inventors

• Kwon Soonchul
• Lee Seunghyun
• Kim Hoijun
• Kim Jaeseung

Classifications

• G06V40/197

  Matching; Classification · CPC title

• G06V40/16

  Human faces, e.g. facial parts, sketches or expressions · CPC title

• G06V10/803

  of input or preprocessed data · CPC title

• G06V40/161

  Detection; Localisation; Normalisation · CPC title

• G06V10/7515

  Shifting the patterns to accommodate for positional errors · CPC title