# US 2007/0036398 A1 · Apparatus and method for partial component facial recognition

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2007/0036398 A1
- Document ID: US20070036398A1
- Status: Published application
- Kind: A1
- Application number: 11/464,145
- Filed: 2006-08-11
- Published: 2007-02-15
- Earliest claimed: 2005-08-12
- Assignees: InteliTrac, Inc.
- Inventors: Tianlong Chen
- Web page: https://patentgrep.com/patent/US20070036398A1

## CPC

- G06V 40/172: {Classification, e.g. identification}
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A method and system for identifying a human being or verifying two human beings by partial component(s) of their face which may include one or multiple of left eye(s), right eye(s), nose(s), mouth(s), left ear(s), or right ear(s). A gallery database for face recognition is constructed from a plurality of human face images by detecting and segmenting a plurality of partial face components from each of the human face images, creating a template for each of the plurality of partial face components, and storing the templates in the gallery database. Templates from a plurality of partial face components from a same human face image are linked with one ID in the gallery database. A probe human face image is identified from the gallery database by detecting and segmenting a plurality of partial face components from the probe human face image; creating a probe template for each of the partial face components from the probe human face image, comparing each of the probe templates against a category of templates in the gallery database to generate similarity scores between the probe templates and templates in the gallery database; generating a plurality of sub-lists of candidate images having partial face component templates with the highest similarity scores over a first preset threshold; generating for each candidate image from each sub-list a combined similarity score; and generating a final list of candidates from said candidates of combined similarity scores over a second preset threshold.

## Claims

### Claim 1

A method for constructing a gallery database for face recognition from a plurality of human face images, comprising the steps of:
detecting and segmenting a plurality of partial face components from each of said human face images;
creating a template for each of said plurality of partial face components;
storing said templates in said gallery database;
wherein templates from a plurality of partial face components from a same human face image will be linked with one ID in said gallery database.

### Claim 2 (depends on 1)

A method for constructing a gallery database for face recognition according to claim 1 wherein said plurality of partial face components comprises at least one of a right eye, a left eye, a nose, a mouth, a right ear, and a left ear of a human face.

### Claim 3 (depends on 1)

A method for constructing a gallery database for face recognition according to claim 1 wherein templates of a same kind from different human face images are grouped into one category in said gallery database.

### Claim 4 (depends on 1)

A method for constructing a gallery database for face recognition according to claim 1 wherein templates from left eyes are grouped into a first category and templates from right eyes are grouped into a second category.

### Claim 5 (depends on 4)

A method for constructing a gallery database for face recognition according to claim 4, wherein templates from noses are grouped into a third category and templates from mouths are grouped into a fourth category.

### Claim 6 (depends on 4)

A method for constructing a gallery database for face recognition according to claim 4, wherein templates from right ears are grouped into a third category and templates from left ears are grouped into a fourth category.

### Claim 7 (depends on 1)

A method for constructing a gallery database for face recognition according to claim 1 wherein said templates are only created only for said partial face components having good quality.

### Claim 8

A method for face identification of a probe human face image from a plurality of human face images in a gallery database having a plurality of templates, each template corresponding to a partial face component and said templates in said gallery database being grouped into categories, comprising the steps of:
detecting and segmenting a plurality of partial face components from said probe human face image;
creating a probe template for each of said partial face components from said probe human face image;
comparing each of said probe templates against a category of templates in said gallery database to generate similarity scores between said probe templates and templates in said gallery database;
generating a list of candidate images whose partial face component templates have similarity scores with partial face component templates of said human face probe image over a preset threshold.

### Claim 9 (depends on 8)

A method for face recognition according to claim 8 wherein said generating generates a list of candidate images having the highest similarity scores over said preset threshold.

### Claim 10

A method for face identification of a probe human face image from a plurality of human face images in a gallery database having a plurality of templates, each template corresponding to a partial face component and said templates in said gallery database being grouped into categories, comprising the steps of:
detecting and segmenting a plurality of partial face components from said probe human face image;
creating a probe template for each of said partial face components from said probe human face image;
comparing each of said probe templates against a category of templates in said gallery database to generate similarity scores between said probe templates and templates in said gallery database;
generating a plurality of sub-lists of candidate images having partial face component templates with the highest similarity scores over a first preset threshold;
generating for each candidate image from each said sub-list a combined similarity score; and
generating a final list of candidates having combined similarity scores over a second preset threshold.
