# US 2008/0187220 A1 · Device and method for fast computation of region based image features

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

- Number: US 2008/0187220 A1
- Document ID: US20080187220A1
- Status: Published application
- Kind: A1
- Application number: 11/987,785
- Filed: 2007-12-04
- Published: 2008-08-07
- Earliest claimed: 2006-12-04
- Assignees: Lockheed Martin Corporation
- Inventors: Gianfranco Doretto; Xiaogang Wang; Thomas Sebastian; Jens Rittscher; Peter H. Tu
- Web page: https://patentgrep.com/patent/US20080187220A1

## CPC

- G06V 10/446: {using Haar-like filters, e.g. using integral image techniques}
- G06V 10/507: {Summing image-intensity values; Histogram projection analysis}
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A device and method for efficient computation of statistical information, such as a mean, co-variance, or histogram of the image pixels, over discrete image regions. The computation employs integral computations to determine the statistical information over image regions of arbitrary shape, including irregular polygonal shaped regions. The integral computations are simplified by categorizing corner points of boundaries of image regions. The computation can be applied to calculate descriptors or signatures of persons or objects within an image. The computation also has a low computational cost enabling fast calculation of image statistics.

## Claims

### Claim 1

An image statistic computation device for computing region-based image statistics of an inputted image, the device comprising:
a propagation device that is configured to produce a cumulative image representation by propagating an aggregate function through image data from the inputted image, where each data point in the cumulative image representation includes cumulative image information that is based on a value of previously propagated adjacent points as well as a value of the data point;
a target region recognition device that is configured to identify at least one generalized rectangular region from the cumulative image representation;
a corner analyzing device that is configured to identify corners of the identified generalized rectangular regions, and characterize each corner as one of a predefined corner type; and
a computation device that is configured to compute statistical information over the identified generalized rectangular regions based on the type of each corner and the cumulative image information at each corner.

### Claim 2 (depends on 1)

The image statistic computation device according to claim 1, wherein the corner analyzing device assigns each corner of the generalized rectangular region a value based on the type of corner.

### Claim 3 (depends on 2)

The image statistic computation device according to claim 2, wherein the computation device computes statistical information over the at least one generalized rectangular region by summing, for each corner of the at least one generalized rectangular region, the product of the value assigned to each corner and the cumulative image information at each corner.

### Claim 4 (depends on 1)

The image statistic computation device according to claim 1, wherein the corner analyzing device characterizes each corner as one of ten predefined corner types.

### Claim 5 (depends on 2)

The image statistic computation device according to claim 2, wherein the corner analyzing device assigns each corner of the generalized rectangular region a value selected from the set {0, ±1, ±2}.

### Claim 6 (depends on 1)

The image statistic computation device according to claim 1, wherein the corner analyzing device characterizes each corner as one of eight predefined corner types.

### Claim 7 (depends on 2)

The image statistic computation device according to claim 2, wherein the corner analyzing device assigns each corner of the generalized rectangular region a value selected from the set {0, ±1}.

### Claim 8 (depends on 1)

The image statistic computation device according to claim 1, wherein the aggregate function is an integral image function.

### Claim 9 (depends on 1)

The image statistic computation device according to claim 1, wherein the aggregate function is an integral histogram function.

### Claim 10 (depends on 1)

The image statistic computation device according to claim 1, wherein the propagation device propagates the aggregate function through the image data on a pixel-by-pixel basis.

### Claim 11 (depends on 10)

The image statistic computation device according to claim 10, wherein the inputted image is a two dimensional image and the propagation device propagates the aggregate function by scanning the image data in a direction from left-to-right.

### Claim 12 (depends on 1)

The image statistic computation device according to claim 1, wherein the image statistic computation device is configured to calculate region-based image statistics from a two-dimensional image.

### Claim 13 (depends on 1)

The image statistic computation device according to claim 1, wherein the image statistic computation device is configured to calculate region-based image statistics from a three-dimensional image.

### Claim 14 (depends on 1)

The image statistic computation device according to claim 1, wherein the target region recognition device is configured to identify a generalized rectangular region having an L-shape from the cumulative image representation.

### Claim 15 (depends on 1)

The image statistic computation device according to claim 1, wherein the target region recognition device is configured to identify a generalized rectangular region from the cumulative image representation, where the region is defined by pixels that have a common attribute.

### Claim 16 (depends on 1)

The image statistic computation device according to claim 1, further comprising:
a pre-processing device that is configured to convert RGB image data from the inputted image into other data formats.

### Claim 17 (depends on 1)

The image statistic computation device according to claim 1, further comprising:
a pre-processing device that is configured to convert image data from the inputted image into an appearance labeled image where each data point is assigned a label based on an appearance attribute.

### Claim 18 (depends on 1)

The image statistic computation device according to claim 1, further comprising:
a pre-processing device that is configured to convert image data from the inputted image into a shape labeled image where each data point is assigned a label based on a shape attribute.

### Claim 19

A method for computing statistical information over a region of an image, the method comprising:
receiving image data of an inputted image;
producing a cumulative image representation by propagating an aggregate function through the image data, where each data point in the cumulative image representation includes cumulative image information that is based on a value of previously propagated adjacent points as well as a value of the data point;
identifying a generalized rectangular region in the cumulative image representation;
identifying corners of the generalized rectangular region;
characterizing each corner as one of a predefined corner type;
assigning a value to each corner based on the type of each corner; and
computing statistical information over the identified generalized rectangular region based on the value assigned to each corner and the cumulative image information at each corner.

### Claim 20 (depends on 19)

The method according to claim 19, further comprising:
summing, for each corner of the identified generalized rectangular region, the product of the value assigned to each corner and the cumulative image information at each corner.

### Claim 21

A method for calculating an identifying descriptor for a person or object in an image, the method including:
receiving image data from an inputted image;
calculating an appearance labeled image from the image data;
producing a cumulative image representation of the appearance labeled image by propagating an aggregate function through the appearance labeled image, where each data point in the cumulative image representation includes cumulative image information that is based on a value of previously propagated adjacent points as well as a value of the data point;
identifying a first generalized rectangular region in the cumulative image representation;
identifying corners of the first generalized rectangular region, and characterizing each corner as one of a predefined type;
calculating a first image statistic over the first generalized rectangular region based on the type of each corner and the cumulative image information at each corner.

### Claim 22 (depends on 19)

The method for calculating an identifying descriptor according to claim 19, further comprising:
calculating a shape labeled image from the image data;
identifying a second generalized rectangular region in the shape labeled image;
identifying corners of the second generalized rectangular region, and characterizing each corner as one of a predefined type;
identifying data points of the cumulative image representation that correspond to the corners of the second generalized rectangular region;
calculating a second image statistic over the second generalized rectangular region based on the type of each corner of the second generalized rectangular region and the first image statistic calculated at each data point of the cumulative image representation that corresponds to a corner of the second generalized rectangular region.
