# US 2020/0005023 A1 · Facial image recognition using pseudo-images

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

- Number: US 2020/0005023 A1
- Document ID: US20200005023A1
- Status: Published application
- Kind: A1
- Application number: 16/453,545
- Filed: 2019-06-26
- Published: 2020-01-02
- Earliest claimed: 2018-07-02
- Assignees: Stowers Institute for Medical Research
- Applicants: Stowers Institute for Medical Research
- Inventors: Congrong (Ron) Yu; Rishabh Raj; Dar Wilbur Dahlen
- Web page: https://patentgrep.com/patent/US20200005023A1

## CPC

- G06V 10/7715: {Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods}
- G06V 40/169: {Holistic features and representations, i.e. based on the facial image taken as a whole}
- G06V 40/172: {Classification, e.g. identification}
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

This disclosure relates to the use of “pseudo-images” to perform image recognition, e.g., to perform facial image recognition. In an embodiment, the pseudo-image is obtained by starting with a real world image and, after optional preprocessing, subjecting the image to a non-linear transformation that converts the image into a pseudo-image. While real world objects (or, more generally, real world patterns) may be perceivable in the starting image, they cannot be perceived in the pseudo-image. Image recognition takes place by comparing the pseudo-image with a library of known pseudo-images, i.e., image recognition takes place in pseudo-image space without a return to real world space. In this way, robust image recognition is achieved even for imperfect real world images, such as, real world images that have been degraded by noise, poor illumination, uneven lighting, and/or occlusion, e.g., the presence of glasses, scarves, or the like in the case of facial images.

## Claims

### Claim 1

A method comprising:
(a) receiving an image in a computer system;
(b) using the computer system to perform a sparse, non-negative transformation of the image into a pseudo-image using a predetermined transformation matrix;
(c) using the computer system to compare the pseudo-image with a library of pseudo-images of known images; and
(d) using the computer system to output the results of the comparison of the pseudo-image with the library of pseudo-images of known images;
wherein the image has M components, the pseudo-image has K components, and K is greater than or equal to M.

### Claim 2

A method comprising:
(a) receiving an image in a computer system;
(b) using the computer system to perform a sparse, non-negative transformation of the image into a pseudo-image using a predetermined transformation matrix;
(c) using the computer system to compare the pseudo-image with a library of pseudo-images of known images; and
(d) using the computer system to output the results of the comparison of the pseudo-image with the library of pseudo-images of known images;
wherein the image has M components each of which has only one of two possible values.

### Claim 3 (depends on 1, 2)

The method of claim 1 or 2 wherein the computer system performs the sparse, non-negative transformation using at least one L2 norm.

### Claim 4 (depends on 1, 2)

The method of claim 1 or 2 wherein the predetermined transformation matrix is a matrix obtained by a method comprising using a computer system to perform a sparse, non-negative factorization of a matrix of training images.

### Claim 5 (depends on 4)

The method of claim 4 wherein the matrix of training images is an M×N matrix where N is greater than or equal to M.

### Claim 6 (depends on 4)

The method of claim 4 wherein the computer system performs the sparse, non-negative factorization using at least one Frobenius norm.

### Claim 7 (depends on 1, 2)

The method of claim 1 or 2 wherein the image of step (a) is a pre-processed image.

### Claim 8 (depends on 1, 2)

The method of claim 1 or 2 wherein the computer system performs the comparison of step (c) using at least one of a Euclidean distance and a cosine distance.

### Claim 9 (depends on 1)

The method of claim 1 wherein the image of step (a) comprises a human face.

### Claim 10

A method of performing computer-implemented image recognition comprising:
(a) providing to one or more computer processors a first-image having M components;
(b) providing to the one or more computer processors a predetermined transformation matrix, wherein:
(i) the predetermined transformation matrix is an M×K matrix in which the K columns constitute a set of K dictionary elements, and
(ii) the predetermined transformation matrix is constructed by a method comprising performing a sparse, non-negative factorization of an M×N matrix in which the N columns constitute a set of N training images, each training image having M components, the sparse, non-negative factorization employing at least one Frobenius norm;
(c) constructing, using the one or more computer processors, a pseudo-image for the first-image using the predetermined transformation matrix to perform a sparse, non-negative transformation of the first-image, said pseudo-image for the first-image consisting of K element weights, each element weight being for one of the K dictionary elements, the sparse, non-negative transformation employing at least one L2 norm;
(d) comparing, using the one or more computer processors, the pseudo-image for the first-image with a library of pseudo-images of known images using at least one of a Euclidean distance and a cosine distance; and
(e) outputting, using the one or more computer processors, the results of the comparison of the pseudo-image with the library of pseudo-images of known images;
wherein the M×N matrix has a rank R and K satisfies one or both of the following relationships:
(i) K is greater than or equal to M; and
(ii) K is greater than or equal to R.

### Claim 11

A method of preparing a predetermined transformation matrix for use in image recognition comprising:
(a) providing a set of N training images to a computer system, each training image having M components;
(b) using the computer system to produce a predetermined transformation matrix by performing a sparse, non-negative factorization of an M×N matrix in which each of the N columns of the matrix constitutes one of the training images, said sparse, non-negative factorization employing at least one Frobenius norm; and
(c) storing the predetermined transformation matrix in a non-transitory computer readable medium;
wherein:
(i) the predetermined transformation matrix is a M×K matrix;
(ii) the M×N matrix has a rank R; and
(iii) K satisfies one or both of the following relationships:
(A) K is greater than or equal to M; and
(B) K is greater than or equal to R.

### Claim 12 (depends on 11)

The method of claim 11 wherein a set of pseudo-images for the N training images is produced in step (b) and the method further comprises using the computer system to store at least some of those pseudo-images in a non-transitory computer readable medium as at least part of a pseudo-image library.

### Claim 13 (depends on 11)

The method of claim 11 further comprising distributing the predetermined transformation matrix as an article of commerce.

### Claim 14 (depends on 1, 10)

The method of claim 1, 10, or 11 wherein K is greater than M.

### Claim 15

A method for preparing a predetermined transformation matrix for use in image recognition from a prior predetermined transformation matrix comprising:
(a) providing a prior predetermined transformation matrix Φ0 to a computer system, said prior predetermined transformation matrix having been obtained using a set of N training images;
(b) providing a set A0 of pseudo-images for the N training images to the computer system;
(c) providing a set Y of N′ training images to the computer system, where N′ is greater than or equal to one and at least one member of said set is a training image that is not part of the set of N training images;
(d) using the computer system to produce a predetermined transformation matrix using a concatenation of Y with the matrix product Φ0A0, where the Φ0A0 matrix product serves as a proxy for the set of N training images; and
(e) storing the predetermined transformation matrix of step (d) in a non-transitory computer readable medium.

### Claim 16 (depends on 15)

The method of claim 15 wherein a set of pseudo-images for the N′ training images is produced in step (d) and the method further comprises using the computer system to store at least some of those pseudo-images in a non-transitory computer readable medium as at least part of a pseudo-image library.

### Claim 17 (depends on 15)

The method of claim 15 further comprising distributing the predetermined transformation matrix of step (d) as an article of commerce.

### Claim 18

A method of preparing or augmenting a library of pseudo-images for use in image recognition comprising:
(a) providing a set of known images to a computer system;
(b) using the computer system to perform sparse, non-negative transformations of the known images into pseudo-images using a predetermined transformation matrix; and
(c) using the computer system to store at least some of the pseudo-images in a non-transitory computer readable medium as at least part of a pseudo-image library.

### Claim 19 (depends on 12, 16)

The method of claim 12, 16, or 18 further comprising distributing the pseudo-image library as an article of commerce.

### Claim 20 (depends on 11, 15)

A non-transitory computer readable medium having a predetermined transformation matrix prepared by the method of claim 11 or 15 stored therein.

### Claim 21 (depends on 12, 16)

A non-transitory computer readable medium having a pseudo-image library prepared at least in part by the method of claim 12, 16, or 18 stored therein.

### Claim 22

A non-transitory computer readable medium comprising a library of pseudo-images of known images for comparison with a pseudo-image for an unknown image wherein the pseudo-images of known images are obtained by a method comprising performing sparse, non-negative transformations of the known images into pseudo-images using a predetermined transformation matrix.

### Claim 23

A non-transitory computer readable medium with instructions stored therein capable of being executed by a computer processor to perform the steps of:
(a) transforming an image into a pseudo-image;
(b) comparing the pseudo-image with a library of pseudo-images of known images; and
(c) outputting the results of the comparison of the pseudo-image with the library of pseudo-images of known images;
wherein the transformation of step (a) is a sparse, non-negative transformation using a predetermined transformation matrix.

### Claim 24 (depends on 23)

A computer system comprising the non-transitory computer-readable medium of claim 23 and a computer processor for executing the instructions stored therein.

### Claim 25

A system comprising:
a computer processor;
at least one computer memory (e.g., a RAM);
at least one computer storage device (e.g., a hard drive, a flash drive, and/or the cloud);
a computer interface that receives an image and stores the image in the at least one computer memory; and
a computer program capable of being executed by the computer processor to generate a pseudo-image for the received image and store the pseudo-image in the at least one computer storage device;
wherein the computer program is capable of generating the pseudo-image by a method comprising performing a sparse, non-negative transformation of the image using a predetermined transformation matrix.

### Claim 26 (depends on 25)

The system of claim 25 wherein the computer program is capable of comparing the pseudo-image with a library of pseudo-images and outputting a result of the comparison.

### Claim 27 (depends on 25, 26)

The system of claim 25 or 26 wherein the computer program is capable of including the pseudo-image in a library of pseudo-images.

### Claim 28 (depends on 25)

The system of claim 25 further comprising an image capture device capable of providing an image to the computer interface.
