# US 2022/0067884 A1 · Method and system for designing an optical filter

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

- Number: US 2022/0067884 A1
- Document ID: US20220067884A1
- Status: Published application
- Kind: A1
- Application number: 17/007,789
- Filed: 2020-08-31
- Published: 2022-03-03
- Earliest claimed: 2020-08-31
- Assignees: Element Ai Inc.
- Applicants: ELEMENT AI Inc.
- Inventors: Philippe BEAUDOIN; Sherif ELSAYED-ALI
- Web page: https://patentgrep.com/patent/US20220067884A1

## CPC

- G06T 5/70: Denoising; Smoothing
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/00: Machine learning
- G06T 5/50: using two or more images, e.g. averaging or subtraction
- G06T 5/60: using machine learning, e.g. neural networks
- G06V 10/70: using pattern recognition or machine learning (optical pattern recognition or electronic computations therefor G06V10/88)
- G06T 2207/10024: Color image
- G06T 2207/20024: Filtering details
- G06T 2207/20076: Probabilistic image processing
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A method for designing an image filter, the method being executed by a machine learning algorithm (MLA), the method comprising: receiving unfiltered images, a desirable visual task and an undesirable visual task; for each unfiltered image, receiving a first label indicative of whether the desirable visual task is accomplishable and a second label indicative of whether the undesirable visual task is accomplishable; filtering the unfiltered image using a virtual filter model representative of the image filter; using the filtered images, training the MLA to perform the desirable visual task and prevent the undesirable visual task to be performed; using test images, determining a first efficiency for the MLA to perform the desirable task and a second efficiency for the MLA to prevent the undesirable task; adjusting a parameter of the virtual filter model based on the first and second efficiencies; and outputting the adjusted parameter.

## Claims

### Claim 1

A method for designing an image filter, the method being executed by a processor, the processor having access to a machine learning algorithm (MLA), the method comprising:
receiving at least one unfiltered image, a desirable visual task and an undesirable visual task being accomplishable from the at least one unfiltered image;
for each one of the at least one unfiltered image, receiving a first label indicative of whether the desirable visual task is accomplishable and a second label indicative of whether the undesirable visual task is accomplishable;
filtering the at least one unfiltered image using a virtual filter model representative of the image filter, thereby obtaining at least one filtered image;
using the at least one filtered image, training the MLA to perform the desirable visual task and prevent the undesirable visual task to be performed;
using a set of test images each having at least one of the first label and second label assigned thereto, determining a first efficiency for the MLA to perform the desirable task and a second efficiency for the MLA to prevent the undesirable task;
adjusting a value of at least one lens parameter of the virtual filter model based on the first and second efficiencies; and
outputting the value the at least one lens parameter.

### Claim 2 (depends on 1)

The method of claim 1, wherein said training the MLA is performed further using at least one of the at least one filtered image.

### Claim 3 (depends on 1)

The method of claim 1, wherein the set of test images comprises at least one of the unfiltered images.

### Claim 4 (depends on 1)

The method of claim 1, wherein said adjusting a value of at least one lens parameter further comprising:
determining a third efficiency for the MLA to perform the undesirable visual task;
adjusting at least one MLA parameter of the MLA based on the first and third efficiencies; and
outputting the trained MLA.

### Claim 5 (depends on 2)

The method of claim 2, further comprising, when the at least one unfiltered image comprises a set of unfiltered images, dividing the set of unfiltered images into a plurality of batches of unfiltered images, wherein said filtering the unfiltered images, said training the MLA, said determining the first and second efficiencies and said adjusting the value of the at least one lens parameter are performed successively for each one of the plurality of batches of unfiltered images.

### Claim 6 (depends on 1)

The method of claim 1, wherein said filtering the at least one unfiltered image comprises blurring the at least one unfiltered image.

### Claim 7 (depends on 1)

The method of claim 1, wherein the image filter is a lens and the virtual filter model is a virtual lens model.

### Claim 8 (depends on 1)

The method of claim 1, wherein said filtering the at least one unfiltered image comprises one of transforming the at least one unfiltered image in a colorspace and adding frames together.

### Claim 9 (depends on 1)

The method of claim 1, wherein the virtual filter model is differentiable.

### Claim 10 (depends on 1)

The method of claim 1, wherein the MLA comprises one of a classification machine learning algorithm and a regression machine learning algorithm.

### Claim 11

A system for designing an image filter, the system comprising:
a processor; and
a non-transitory storage medium operatively connected to the processor, the non-transitory storage medium comprising computer readable instructions;
the processor having access to a machine learning algorithm (MLA), the processor, upon executing the computer readable instructions, being configured for:
receiving at least one unfiltered image, a desirable visual task and an undesirable visual task being accomplishable from the at least one unfiltered image;
for each one of the at least one unfiltered image, receiving a first label indicative of whether the desirable visual task is accomplishable and a second label indicative of whether the undesirable visual task is accomplishable;
filtering the at least one unfiltered image using a virtual filter model representative of the image filter, thereby obtaining at least one filtered image;
using the at least one filtered image, training the MLA to perform the desirable visual task and prevent the undesirable visual task to be performed;
using a set of test images each having at least one of the first label and second label assigned thereto, determining a first efficiency for the MLA to perform the desirable task and a second efficiency for the MLA to prevent the undesirable task;
adjusting a value of at least one lens parameter of the virtual filter model based on the first and second efficiencies; and
outputting the value the at least one lens parameter.

### Claim 12 (depends on 11)

The system of claim 11, wherein the processor is configured for training the MLA further using at least one of the at least one filtered image.

### Claim 13 (depends on 11)

The system of claim 11, wherein the set of test images comprises at least one of the unfiltered images.

### Claim 14 (depends on 11)

The system of claim 11, wherein the processor is further configured for:
determining a third efficiency for the MLA to perform the undesirable visual task;
adjusting at least one MLA parameter of the MLA based on the first and third efficiencies; and
outputting the trained MLA.

### Claim 15 (depends on 12)

The system of claim 12, when the at least one unfiltered image comprises a set of unfiltered images, the processor is further configured for dividing the set of unfiltered images into a plurality of batches of unfiltered images, wherein said filtering the unfiltered images, said training the MLA, said determining the first and second efficiencies and said adjusting the value of the at least one lens parameter are performed successively by the processor for each one of the plurality of batches of unfiltered images.

### Claim 16 (depends on 11)

The system of claim 11, wherein said filtering the at least one unfiltered image comprises blurring the at least one unfiltered image.

### Claim 17 (depends on 11)

The system of claim 11, wherein the image filter is a lens and the virtual filter model is a virtual lens model.

### Claim 18 (depends on 11)

The system of claim 11, wherein said filtering the at least one unfiltered image comprises one of transforming the at least one unfiltered image in a colorspace and adding frames together.

### Claim 19 (depends on 11)

The system of claim 11, wherein the virtual filter model is differentiable.

### Claim 20 (depends on 11)

The system of claim 11, wherein the MLA comprises one of a classification machine learning algorithm and a regression machine learning algorithm.
