Granted Patent B2

US 12,738,094 · App. 18/434,093 · Granted Sep 15, 2026

Inventors: Ho-Sub Yoon (Daejeon, KR); Jae-Hong Kim (Daejeon, KR); Jae-Yoon Jang (Daejeon, KR)

Assignee: Electronics and Telecommunications Research Institute

G06V40/172G06V40/171

Abstract

Claims (32)

1 . A facial recognition method based on masking comprising:

performing detection and normalization of a face area including face landmarks in an input image;

separating a face portion other than a portion covered by an obstacle in the detected face area by performing face segmentation that separates an uncovered face portion;

generating a mask corresponding to the separated uncovered face portion;

applying the mask to a registered face image in a face image registration database;

extracting a first set of features from a whole face area of both the input image and the registered face image, and extracting a second set of features from a masking region corresponding to the uncovered face portion of both the input image and the registered face image; and

determining whether the input image and the registered face image represent an identical person by calculating a total similarity based on a combination of a first similarity derived from the first set of features and a second similarity derived from the second set of features, wherein the first similarity and the second similarity are combined using weights determined by a ratio of the masking region to the whole face area.

2 . The facial recognition method of claim 1 , wherein generating the mask corresponding to the separated face portion comprises:

determining that there is a covered portion when the face landmarks are not present in the separated face portion or when a size of a detected face element is smaller than a preset size.

3 . The facial recognition method of claim 2 , wherein the face landmarks include center positions of both eyes, a center position of nose, and both corner positions of mouth.

4 . The facial recognition method of claim 1 , wherein, among the weights determined by the ratio of the masking region to the whole face area, a weight applied to the second similarity increases as the ratio of the masking region to the whole face area decreases.

5 . The facial recognition method of claim 1 , wherein the weights comprise a first weight applied to the first similarity and a second weight applied to the second similarity, and wherein a sum of the first weight and the second weight is one.

6 . The facial recognition method of claim 1 , wherein the face image registration database stores both the registered face image and a pre-calculated feature vector corresponding to the whole face area of the registered face image.

7 . The facial recognition method of claim 1 , further comprising:

determining whether the input image and the registered face image represent based solely on the first similarity when there is no portion covered by an obstacle in the separated face portion.

8 . A facial recognition apparatus based on masking comprising:

a memory configured to store one or more instructions; and

a processor configured to execute,

the one or more instructions to:

perform detection and normalization of a face area including face landmarks in an input image;

separates a face portion other than a portion covered by an obstacle in the normalized face area by performing face segmentation that separates an uncovered face portion;

generate a mask corresponding to the separated uncovered face portion;

applying the mask to a registered face image in a face image registration database;

extract a first set of features from a whole face area of both the input image and the registered face image, and extracting a second set of features from a masking region corresponding to the uncovered face portion of both the input image and the registered face image; and

determine whether the input image and the registered face image represent is an identical person by calculating a total similarity based on a combination of a first similarity derived from the first set of features and a second similarity derived from the second set of features, wherein the first similarity and the second similarity are combined using weights determined by a ratio of the masking region to the whole face area.

9 . The facial recognition apparatus of claim 8 , wherein the processor is further configured to execute the one or more instructions to:

determine that there is a covered portion when the face landmarks are not present in the separated face portion or when a size of a detected face element is smaller than a preset size.

10 . The facial recognition apparatus of claim 9 , wherein the face landmarks include center positions of both eyes, a center position of nose, and both corner positions of mouth.

11 . The facial recognition apparatus of claim 8 , wherein, among the weights determined by the ratio of the masking region to the whole face area, a weight applied to the second similarity increases as the ratio of the masking region to the whole face area decreases.

12 . The facial recognition apparatus of claim 8 , wherein the weights comprise a first weight applied to the first similarity and a second weight applied to the second similarity, and wherein a sum of the first weight and the second weight is one.

13 . The facial recognition apparatus of claim 8 , wherein the face image registration database stores both the registered face image and a pre-calculated feature vector corresponding to the whole face area of the registered face image.

14 . The facial recognition apparatus of claim 8 , wherein the program is configured to determine whether the input image and the registered face image represent based solely on the first similarity when there is no portion covered by an obstacle in the separated face portion.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (20)