# US 2022/0027676 A1 · Image multiprocessing method for vision systems

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

- Number: US 2022/0027676 A1
- Document ID: US20220027676A1
- Status: Published application
- Kind: A1
- Application number: 17/295,771
- Filed: 2018-11-21
- Published: 2022-01-27
- Earliest claimed: 2018-11-21
- Assignees: Datalogic IP Tech S.R.L.
- Applicants: Datalogic IP Tech S.R.L.
- Inventors: Francesco D'ERCOLI; Francesco Paolo MUSCARIDOLA
- Web page: https://patentgrep.com/patent/US20220027676A1

## CPC

- G06F 18/2163: {Partitioning the feature space}
- G06T 1/20: Processor architectures; Processor configuration, e.g. pipelining
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/23: Clustering techniques
- G06F 18/24: Classification techniques
- G06V 10/955: {using specific electronic processors}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

The present disclosure relates to a method for image classes definition and to a method for image multiprocessing and related vision system, which implement said method for image classes definition. The latter comprising an image splitting operation for each image of M input images, the image splitting operation comprising the steps of: a) splitting the image into image portions; b) executing the algorithm onto each image portion with at least one processing unit; c) identifying the image portion associated with a maximum execution time of said algorithm; d) splitting said identified image portion into further image portions; e) checking if a stop criterion is met: e1) if the stop criterion is met, iterating steps a) to e) onto another one of the M input images; e2) if the stop criterion is not met, executing a predefined image processing algorithm onto each of the further image portions; identifying the image portion or further image portion associated with a maximum execution time of said algorithm; and iterating steps d) to e) on the so identified image portions portion or further image portion; wherein after executing steps a) to e) on all of the M input images, the method for image classes definition further comprises the steps of: f) identifying in an image space, for all the M input images, the positions of each split image portion/further image portion and defining clusters (A, B, C) based thereon; g) defining a set of Q image classes (A′, B′, C, AC) based on said clusters (A, B, C), each class (A′, B′, C, AC) being univocally associated with a splitting pattern representing in the image space a plurality of regions to be allocated to a corresponding plurality of processing units of the vision system for image multiprocessing.

## Claims

### Claim 1

A method for defining image classes definition for image multiprocessing according to a predefined image processing algorithm of a vision system, said method comprising:
an image splitting operation for each image of an integer number of input images, with the integer number being higher than 1, the image splitting operation including:
a) splitting the image into image portions;
b) executing the predefined image processing algorithm onto each image portion with at least one processing unit;
c) measuring an execution time of the predefined image processing algorithm onto each of the image portions and identifying the image portion among the image portions which is associated with a maximum execution time of the predefined image processing algorithm;
d) splitting the identified image portion into further image portions;
e) checking if a stop criterion is met:
e1) if the stop criterion is met, iterating steps a) to e) onto another one of the input images until the image splitting operation has been performed onto all the input images;
e2) if the stop criterion is not met,
executing the predefined image processing algorithm onto each of the further image portions;
measuring an execution time of the predefined image processing algorithm on each of the further image portions and identifying, among the image portions and the further image portions, the image portion or further image portion which is associated with a maximum execution time of the predefined image processing algorithm; and
iterating steps d) to e) on the so identified image portion or further image portion;
wherein after executing steps a) to e) on all of the input images, the method further including:
f) identifying in an image space, for all the input images, the positions of each split image portion and any split further image portion and defining clusters based on the identified positions;
g) defining a set of image classes based on the clusters, with an integer number of image classes being higher than 1, each class being univocally associated with a splitting pattern representing in the image space a plurality of regions to be allocated to a corresponding plurality of processing units of the vision system for image multiprocessing with the predefined image processing algorithm.

### Claim 2 (depends on 1)

The method according to claim 1, wherein in step g) a class is defined for each cluster.

### Claim 3 (depends on 1)

The method according to claim 1, wherein in step g) the set of image classes (A′, B′, C′, AC′) is defined taking into account a correlation of the input images with the defined clusters (A, B, C).

### Claim 4 (depends on 1)

The method according to claim 1,
wherein in step g) at least one of the classes (A′, B′, C′, AC′) is defined based on an aggregation of more clusters (A, B, C).

### Claim 5 (depends on 4)

The method according to claim 4, wherein the aggregation of clusters is taken into account only if a percentage of the input images is correlated with it.

### Claim 6 (depends on 1)

The method according to claim 1, wherein in step g) the plurality of regions of the splitting pattern are defined in the image space by splitting the image space more densely in correspondence of the position of the cluster(s) on the basis of which the corresponding class is defined, and by splitting sparsely the rest of the image space.

### Claim 7 (depends on 1)

The method according to claim 1, wherein in step f), the positions of each split image portion and any split further image portion are marked in the image space by using suitable markers.

### Claim 8 (depends on 7)

The method according to claim 7, wherein the clusters are clusters of the suitable markers identified in the image space.

### Claim 9 (depends on 1)

The method according to claim 1, wherein the stop criterion checks whether a maximum number Rmax of iterations of step d) is reached.

### Claim 10 (depends on 1)

The method according to claim 1, wherein the stop criterion checks whether the last measured maximum execution time is lower than a predefined threshold value.

### Claim 11 (depends on 1)

The method according to claim 1, wherein the splitting of step a) is performed so that the image portions are equal to each other.

### Claim 12 (depends on 1)

The method according to claim 1, wherein the splitting of step d) is performed so that the further image portions are equal to each other.

### Claim 13 (depends on 1)

The method according to claim 1, wherein at least one of the image portions obtained in step a) or the further image portions obtained in step d) have a same aspect ratio of the starting image.

### Claim 14

(canceled)

### Claim 15 (depends on 1)

The method according to claim 1, wherein at each iteration of step d), a same integer number of the further image portions is obtained, with integer number higher than 1.

### Claim 16 (depends on 1)

The method according to claim 1, wherein the image portions obtained in step a) are in a same integer number as the further image portions obtained in step d), with integer number higher than 1.

### Claim 17 (depends on 1)

The method according to claim 1, wherein at least one of the image portions or the further image portions are in a number which is a power of two.

### Claim 18

(canceled)

### Claim 19 (depends on 1)

The method according to claim 1, wherein the input images are training images.

### Claim 20

A method for image multiprocessing in a vision system according to a predefined image processing algorithm, comprising:
receiving an image to be processed;
associating with the image a class among a predefined set of image classes, the class having a splitting pattern univocally associated therewith;
splitting the image according to the splitting pattern; and
executing the predefined image processing algorithm separately onto all portions resulting from the splitting by a corresponding plurality of processing units of the vision system,
wherein the predefined set of image classes and the splitting patterns univocally associated therewith are defined through a method of defining image classes executed with the predefined image processing algorithm.

### Claim 21 (depends on 20)

The method according to claim 20, wherein the method of image classes definition is executed:
wholly offline with training images; or
partially offline with training images and partially online with the same images to be processed by the vision system; or
wholly online with the same images to be processed by the vision system.

### Claim 22 (depends on 1)

A vision system for image multiprocessing according to a predefined image processing algorithm, comprising:
a memory for storing a set of image classes and corresponding splitting patterns univocally associated with the classes, with the number integer images being higher than 1, as defined through a method of image classes definition according to claim 1, executed with the predefined image processing algorithm; and
processing units adapted to:
receiving an image to be processed;
associating with the image a class among a predefined set of image classes, the class having a splitting pattern univocally associated therewith;
splitting the image according to the splitting pattern; and
executing the predefined image processing algorithm separately onto all portions resulting from the splitting by a corresponding plurality of processing units of the vision system,
wherein the predefined set of image classes and the splitting patterns univocally associated therewith are defined through a method of defining image classes, executed with the predefined image processing algorithm.

### Claim 23

(canceled)

### Claim 24

(canceled)

### Claim 25

(canceled)
