Image feature extraction method for person re-identification

Patent metadata

Field	Value
Publication number	US-11238274-B2
Application number	US-201716622586-A
Country	US
Kind code	B2
Filing date	Dec 27, 2017
Priority date	Jul 4, 2017
Publication date	Feb 1, 2022
Grant date	Feb 1, 2022

Abstract

An image feature extraction method for person re-identification includes performing person re-identification by means of aligned local descriptor extraction and graded global feature extraction; performing the aligned local descriptor extraction by processing an original image by affine transformation and performing a summation pooling operation on image block features of same regions to obtain an aligned local descriptor; reserving spatial information between inner blocks of the image for the aligned local descriptor; and performing the graded global feature extraction by grading a positioned pedestrian region block and solving a corresponding feature mean value to obtain a global feature. The method can resolve the problem of feature misalignment caused by posture changes of pedestrian, etc., and eliminate the effect of unrelated backgrounds on re-recognition, thus improving the precision and robustness of person re-identification.

First claim

What is claimed is:

1. An image feature extraction method for person re-identification, comprising:
performing person re-identification by means of aligned local descriptor extraction and graded global feature extraction;
performing the aligned local descriptor extraction by processing an image by affine transformation and performing a summation pooling operation on image block features of adjacent regions to obtain an aligned local descriptor;
reserving spatial information between inner blocks of the image for the aligned local descriptor; and
performing the graded global feature extraction by grading a positioned pedestrian region block and solving a corresponding feature mean value to obtain a global feature;
the method further comprising:
1) pre-processing the image to eliminate the effect of illumination on the image;
2) extracting the aligned local descriptor, comprising:
21) performing affine transformation on the image to obtain multiple images;
22) performing image block segmentation on all images generated after affine transformation, extracting features, and generating corresponding feature vector maps;
23) superimposing all the feature vector maps in space, performing a summation pooling operation on the image block features of the adjacent positions, and obtaining corresponding local descriptors; and
24) connecting all local descriptors in the image in order to obtain an aligned local descriptor;

3) extracting graded global feature, comprising:
31) using the foreground extraction method to perform significance detection on the image to obtain a corresponding significance image;
32) enhancing the contrast of significance image obtained;
33) locating image blocks of the pedestrian area;
34) classifying the image blocks of the pedestrian area to obtain multi-level image blocks of pedestrian area;
35) performing an average pooling operation on each level of the image blocks of the pedestrian area to obtain the feature of the level; and
36) connecting features of multi levels and obtaining graded global feature; and

4) calculating similarity of the images according to the aligned local descriptor extracted in Step 2) and the graded global feature extracted in Step 3) by using the metric learning method, thereby performing person re-identification.

2. An image feature extraction method according to claim 1 , wherein in Step 1), the image is preprocessed by using a multi-scale Retinex algorithm.

3. An image feature extraction method according to claim 1 , wherein in Step 21), an affine transformation is performed on the image, the method further comprising:
horizontally flipping a preprocessed image I to obtain image II; performing horizontal shear transformation on image I and image II respectively to generate corresponding image III and IV; and the horizontal shear transformation is as shown in Equation 1:

[

x
′

y
′

1

]

=

[

1

0

0

λ

1

0

0

0

1

]

⁡

[

ϰ

y

1

]

(
1
)

where λ is a shear transformation parameter.

4. An image feature extraction method according to claim 1 , wherein in Step 22), the sliding window method is used to obtain the image block features, and specifically extracting an RGB color histogram, an HSV color histogram, and statistical histogram of Scale-Invariant Local Three-valued Pattern (SILTP) for image blocks in the window, to obtain multiple feature vector maps,
wherein each feature map has M*N points, and each point corresponds to the feature vector of one image block, denoted as F i m,n .

5. An image feature extraction method according to claim 1 , wherein in Step 23), the four feature vector maps are superimposed by Equation 2, a summation pooling operation is performed on image block features of adjacent positions, and the corresponding local descriptor is obtained:

=

∑

i
=
1

4

⁢

(

F

m
,
n

i

+

F

m
+
1

Assignees

• Univ Peking Shenzhen Graduate School
• Univ Beijing

Inventors

• Wang Wenmin
• Zhang Yihao
• Wang Ronggang
• Li Ge
• Dong Shengfu
• Wang Zhenyu
• Li Ying
• Gao Wen

Classifications

• G06V40/103 Primary

  Static body considered as a whole, e.g. static pedestrian or occupant recognition · CPC title

• G06V10/50

  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

• G06V40/10

  Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G06F18/2163

  Partitioning the feature space · CPC title