Method, device and computer-readable medium for identifying feature of image

Patent metadata

Field	Value
Publication number	US-10297015-B2
Application number	US-201615359816-A
Country	US
Kind code	B2
Filing date	Nov 23, 2016
Priority date	Nov 25, 2015
Publication date	May 21, 2019
Grant date	May 21, 2019

Abstract

A method for identifying a feature of an image is provided. The method includes: segmenting an image into a plurality of blocks, wherein each block includes a plurality of cells; transforming pixels of each cell from a spatial domain to a frequency domain; and identifying a Histogram of Oriented Gradient (HOG) feature of the image in the frequency domain.

First claim

What is claimed is:

1. A method for identifying a feature of an image, comprising:
segmenting an image into a plurality of blocks, wherein each block includes a plurality of cells;
transforming pixels of each cell from a spatial domain to a frequency domain; and
identifying a Histogram of Oriented Gradient (HOG) feature of the image in the frequency domain,
wherein said identifying the HOG feature of the image in the frequency domain comprises:
calculating a gradient magnitude and a gradient direction of each cell in the frequency domain to obtain a descriptor of each cell;
obtaining a HOG feature of each block based on a descriptor of each block in the frequency domain; and
obtaining the HOG feature of the image based on the HOG feature of each block in the frequency domain,

wherein said obtaining the HOG feature of the image comprises:
cascading the HOG feature of each block into a matrix, wherein each column of the matrix represents the HOG feature of one of the plurality of blocks.

2. The method of claim 1 , wherein said transforming each cell from the spatial domain to the frequency domain comprises:
performing a Discrete Cosine Transform (DCT) operation on pixels of each cell.

3. The method of claim 1 , wherein said transforming each cell from the spatial domain to the frequency domain comprises:
performing a Discrete Fourier Transform (DFT) operation on pixels of each cell.

4. The method of claim 1 , wherein said obtaining the HOG feature of the image comprises:
rearranging the HOG feature of each block from an L×1 vector to an M×N matrix, wherein each block comprises M×N pixels, and L=M×N; and
obtaining the HOG feature of the image based on the rearranged HOG feature of each block and a position of the corresponding block in the image.

5. The method of claim 1 , wherein the gradient direction of each cell is divided into z portions, z is an integer, and the descriptor of each cell is a z-dimensional vector obtained based on gradient magnitude of pixels in each of the z portions.

6. The method of claim 1 , further comprising:
normalizing the image to obtain an image having a predefined magnitude.

7. The method of claim 1 , wherein each block overlaps with at least another block.

8. A device for identifying a feature of an image, comprising:
a processor; and
a memory for storing instructions executable by the processor;
wherein the processor is configured to:
segment an image into a plurality of blocks, wherein each block includes a plurality of cells;
transform pixels of each cell from a spatial domain to a frequency domain; and
identify a Histogram of Oriented Gradient (HOG) feature of the image in the frequency domain;
wherein in identifying the HOG feature of the image in the frequency domain, the processor is further configured to:
calculate a gradient magnitude and a gradient direction of each cell in the frequency domain to obtain a descriptor of each cell;
obtain a HOG feature of each block based on a descriptor of each block in the frequency domain; and
obtain the HOG feature of the image based on the HOG feature of each block in the frequency domain;
wherein in obtaining the HOG feature of the image, the processor is further configured to:
cascade the HOG feature of each block into a matrix, wherein each column of the matrix represents the HOG feature of one of the plurality of blocks.

9. The device of claim 8 , wherein the processor is further configured to perform a Discrete Cosine Transform (DCT) operation on pixels of each cell.

10. The device of claim 8 , wherein the processor is further configured to perform a Discrete Fourier Transform (DFT) operation on pixels of each cell.

11. The device of claim 8 , wherein the processor is further configured to:
rearrange the HOG feature of each block from an L×1 dimensional vector to an M×N matrix, wherein each block comprises M×N pixels, and L=M×N; and
obtain the HOG feature of the image based on the rearranged HOG feature of each block and a position of the corresponding block in the image.

12. The device of claim 8 , wherein the processor is further configured to:
divide the gradient direction of each cell into z portions, z being an integer; and
obtain the descriptor of each cell based on gradient magnitude of pixels in each of the z portions, the descriptor of each cell being a z-dimensional vector.

13. The device of claim 8 , the processor is further configured to:
normalize the image to obtain an image having a predefined magnitude.

14. The device of claim 8 , wherein each block overlaps with at least another block.

15. A non-transitory computer-readable storage medium having stored therein instructions that, when executed by a processor, cause the processor to perform a method for identifying a feature of an image, the method comprising:
segmenting an image into a plurality of blocks, wherein each block includes a plurality of cells;
transforming pixels of each cell from a spatial domain to a frequency domain; and
identifying a Histogram of Oriented Gradient (HOG) feature of the image in the frequency domain,
wherein said identifying the HOG feature of the image in the frequency domain comprises:
calculating a gradient magnitude and a gradient direction of each cell in the frequency domain to obtain a descriptor of each cell;
obtaining a HOG feature of each block based on a descriptor of each block in the frequency domain; and
obtaining the HOG feature of the image based on the HOG feature of each block in the frequency domain,

wherein said obtaining the HOG feature of the image comprises:
cascading the HOG feature of each block into a matrix, wherein each column of the matrix represents the HOG feature of one of the plurality of blocks.

Assignees

• Xiaomi Inc

Inventors

• Long Fei
• Chen Zhijun
• Zhang Tao

Classifications

• G06V10/50 Primary

  by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis · CPC title

• G06T5/40 Primary

  using histogram techniques · CPC title

• G06V10/431

  Frequency domain transformation; Autocorrelation · CPC title

• G06K9/6298

  Physics · mapped topic

• G06T5/10

  using non-spatial domain filtering · CPC title