CN105243395A - Human body image comparison method and device - Google Patents
Human body image comparison method and device Download PDFInfo
- Publication number
- CN105243395A CN105243395A CN201510742746.4A CN201510742746A CN105243395A CN 105243395 A CN105243395 A CN 105243395A CN 201510742746 A CN201510742746 A CN 201510742746A CN 105243395 A CN105243395 A CN 105243395A
- Authority
- CN
- China
- Prior art keywords
- human body
- image
- region
- characteristic pattern
- body image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
The invention provides a human body image comparison method and device. The method comprises the following steps: obtaining a first human body image and a second human body image; dividing the first human body image into a plurality of image areas; selecting different amounts of image areas from the plurality of image areas to carry out stitching to obtain a plurality of human body subimages; and carrying out deep learning on the obtained human body subimages and the second human body image to obtain a characteristic differential chart of the first human body image and the second human body image; and carrying out the deep learning on the obtained characteristic differential chart to obtain a comparison result of the first human body image and the second human body image, wherein the comparison result comprises that the first human body image and the second human body image are similar and the first human body image and the second human body image are dissimilar. Human body image comparison accuracy can be improved.
Description
Technical field
The present invention relates to technical field of image processing, in particular to a kind of human body image comparison method and device.
Background technology
At present, in order to find the character image of particular persons (such as: suspect) and determine the whereabouts of particular persons from monitor video, Security Personnel just needs the figure and features feature according to particular persons, from the massive video monitoring image of video monitoring system shooting, determine the monitoring image with particular persons, and determine the whereabouts of particular persons according to the infield of the video monitoring system of shooting monitoring image.
In correlation technique, in order to find the monitoring image with particular persons from the video monitoring image of magnanimity, human body image in the human body image of particular persons and video monitoring image can be divided into upper, middle, and lower part respectively, then degree of depth study is carried out to the human body image after division, and go out to have the monitoring image of particular persons according to result comparison from massive video monitoring image of degree of depth study.
Finding in the process of the monitoring image with particular persons from the video monitoring image of magnanimity, the human body image shown in some video monitoring image only has upper half of human body or only has head, all the other human body parts are all blocked, so the human body image be blocked is divided into, in, after lower three parts, the human body parts just not having particular persons in the middle parts of images of human body image after dividing and/or lower part image may be caused, in the process of carrying out degree of depth study, so do not have the image of human body parts greatly can affect the comparison result of human body image, reduce the accuracy rate of human body image comparison.
Summary of the invention
In view of this, the object of the embodiment of the present invention is to provide a kind of human body image comparison method and device, to improve the accuracy rate of human body image comparison.
First aspect, embodiments provides a kind of human body image comparison method, comprising:
Obtain the first volume image and the second human body image;
Described the first volume image is divided into multiple image-region;
From described multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage;
Degree of depth study is carried out to the described multiple human body subimage obtained and described second human body image, obtains the feature difference figure of described the first volume image and described second human body image;
Carry out degree of depth study to the described feature difference figure obtained, obtain the comparison result of described the first volume image and described second human body image, it is dissimilar that described comparison result comprises phase Sihe.
In conjunction with first aspect, embodiments provide the first possible embodiment of first aspect, wherein, from described multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage and comprise:
From multiple described image-region, select k image-region to splice respectively, obtain multiple human body subimage, multiple described image-region is on average divided from top to bottom by described the first volume image and obtains;
Wherein,
m represents the image-region quantity that described the first volume image divides.
In conjunction with first aspect, embodiments provide the embodiment that the second of first aspect is possible, wherein, carry out degree of depth study to the described multiple human body subimage obtained and described second human body image, the feature difference figure obtaining described the first volume image and described second human body image comprises:
Degree of depth study is carried out to multiple described human body subimage and described second human body image, obtains multiple the first volume image characteristic pattern and the second human body image characteristic pattern;
According to the multiple described the first volume image characteristic pattern obtained and described second human body image characteristic pattern, obtain each the first volume image characteristic pattern in described multiple the first volume image characteristic pattern respectively with the feature difference figure of described second human body image characteristic pattern.
In conjunction with first aspect, embodiments provide the third possible embodiment of first aspect, wherein, process the described the first volume image characteristic pattern obtained and described second human body image characteristic pattern, each the first volume image characteristic pattern obtained in described multiple the first volume image characteristic pattern comprises with the feature difference figure of described second human body image characteristic pattern respectively:
With each pixel coordinate stored in the pixel coordinate set preset for eigenwert regional center, according to the eigenwert area size preset, ex-first lady's volume image characteristic pattern will be worked as respectively and described second human body image characteristic pattern is divided into multiple the First Eigenvalue region and multiple Second Eigenvalue region;
Profile maxima is obtained respectively from described multiple the First Eigenvalue region and described multiple Second Eigenvalue region;
Calculate respectively from the difference with the profile maxima obtained the First Eigenvalue region of same characteristic features value regional center and Second Eigenvalue region, obtain multiple feature difference;
With described multiple feature difference for pixel value, according to the feature difference figure size preset, generate the described feature difference figure working as ex-first lady's volume image characteristic pattern and described second human body image characteristic pattern.
In conjunction with first aspect, embodiments provide the 4th kind of possible embodiment of first aspect, wherein, carry out degree of depth study to the described feature difference figure obtained, the comparison result obtaining described the first volume image and described second human body image comprises:
Degree of depth study is carried out to each feature difference figure obtained, obtains the similar parameter of human body subimage corresponding to each described feature difference figure and described second human body image;
Determine the similarity of maximum similar parameter as described the first volume image and described second human body image;
When described similarity is more than or equal to the similarity threshold of setting, obtain the comparison result that described the first volume image is similar to described second human body image;
When described similarity is less than the similarity threshold of setting, obtain described the first volume image and the dissimilar comparison result of described second human body image.
Second aspect, embodiments provides a kind of human body image comparison device, comprising:
Acquisition module, for obtaining the first volume image and the second human body image;
Image divides module, for described the first volume image is divided into multiple image-region;
Image mosaic module, for selecting the image-region of varying number to splice from described multiple image-region, obtains multiple human body subimage;
Feature difference figure acquisition module, for carrying out degree of depth study to the described multiple human body subimage obtained and described second human body image, obtains the feature difference figure of described the first volume image and described second human body image;
Comparing module, for carrying out degree of depth study to the described feature difference figure obtained, obtains the comparison result of described the first volume image and described second human body image, and it is dissimilar that described comparison result comprises phase Sihe.
In conjunction with second aspect, embodiments provide the first possible embodiment of second aspect, wherein, described image mosaic module is specifically for from multiple described image-region, k image-region is selected to splice respectively, obtain multiple human body subimage, multiple described image-region is on average divided from top to bottom by described the first volume image and obtains;
Wherein,
m represents the image-region quantity that described the first volume image divides.
In conjunction with second aspect, embodiments provide the embodiment that the second of second aspect is possible, wherein, described feature difference figure acquisition module comprises:
Degree of depth unit, for carrying out degree of depth study to multiple described human body subimage and described second human body image, obtains multiple the first volume image characteristic pattern and the second human body image characteristic pattern;
Feature difference figure acquiring unit, for according to the multiple described the first volume image characteristic pattern that obtains and described second human body image characteristic pattern, obtain each the first volume image characteristic pattern in described multiple the first volume image characteristic pattern respectively with the feature difference figure of described second human body image characteristic pattern.
In conjunction with second aspect, embodiments provide the third possible embodiment of second aspect, wherein, described feature difference figure acquiring unit comprises:
Region dividing subelement, for with each pixel coordinate of storing in the pixel coordinate set preset for eigenwert regional center, according to the eigenwert area size preset, ex-first lady's volume image characteristic pattern will be worked as respectively and described second human body image characteristic pattern is divided into multiple the First Eigenvalue region and multiple Second Eigenvalue region;
Profile maxima obtains subelement, for obtaining profile maxima respectively from described multiple the First Eigenvalue region and described multiple Second Eigenvalue region;
Feature difference computation subunit, for calculating respectively from the difference with the profile maxima obtained the First Eigenvalue region of same characteristic features value regional center and Second Eigenvalue region, obtains multiple feature difference;
Feature difference figure generates subelement, for described multiple feature difference for pixel value, according to the feature difference figure size preset, generate the described feature difference figure when ex-first lady's volume image characteristic pattern and described second human body image characteristic pattern.
In conjunction with second aspect, embodiments provide the 4th kind of possible embodiment of second aspect, wherein, described comparing module comprises:
Similar parameter computing unit, for carrying out degree of depth study to each feature difference figure obtained, obtains the similar parameter of human body subimage corresponding to each described feature difference figure and described second human body image;
Similarity determining unit, for determining the similarity of maximum similar parameter as described the first volume image and described second human body image;
First comparison result determining unit, for when described similarity is more than or equal to the similarity threshold of setting, obtains the comparison result that described the first volume image is similar to described second human body image;
Second comparison result determining unit, for when described similarity is less than the similarity threshold of setting, obtains described the first volume image and the dissimilar comparison result of described second human body image.
The human body image comparison method that the embodiment of the present invention provides and device, by being divided into multiple image-region by the first volume image, and according to the multiple image-regions after division, from multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage, then the multiple human body subimages by obtaining, compare with the second human body image respectively, obtain the result of comparison, with prior art, human body image is divided into, in, the human body image comparison process that the part that is blocked in each several part human body image after lower three parts can reduce human body image comparison result accuracy rate is compared, reduce in human body image the impact of part on comparison result that be blocked, improve the accuracy rate of human body image comparison.
For making above-mentioned purpose of the present invention, feature and advantage become apparent, preferred embodiment cited below particularly, and coordinate appended accompanying drawing, be described in detail below.
Accompanying drawing explanation
In order to be illustrated more clearly in the technical scheme of the embodiment of the present invention, be briefly described to the accompanying drawing used required in embodiment below, be to be understood that, the following drawings illustrate only some embodiment of the present invention, therefore the restriction to scope should be counted as, for those of ordinary skill in the art, under the prerequisite not paying creative work, other relevant accompanying drawings can also be obtained according to these accompanying drawings.
Fig. 1 shows the structural representation of a kind of implementation system involved by a kind of human body image comparison method that the embodiment of the present invention provides;
Fig. 2 shows the process flow diagram of a kind of human body image comparison method that the embodiment of the present invention 1 provides;
Fig. 3 shows the schematic diagram of characteristics of image figure in a kind of human body image comparison method that the embodiment of the present invention 1 provides;
Fig. 4 shows the schematic diagram of the another kind of human body image comparison method that the embodiment of the present invention 2 provides;
Fig. 5 shows the structural representation of a kind of human body image comparison device that the embodiment of the present invention 3 provides.
Embodiment
Below in conjunction with accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.The assembly of the embodiment of the present invention describing and illustrate in usual accompanying drawing herein can be arranged with various different configuration and design.Therefore, below to the detailed description of the embodiments of the invention provided in the accompanying drawings and the claimed scope of the present invention of not intended to be limiting, but selected embodiment of the present invention is only represented.Based on embodiments of the invention, the every other embodiment that those skilled in the art obtain under the prerequisite not making creative work, all belongs to the scope of protection of the invention.
Consider in relevant human body image comparison technology, finding in the process of the monitoring image with particular persons from the video monitoring image of magnanimity, the human body image shown in some video monitoring image only has upper half of human body or only has head, all the other human body parts are all blocked, so the human body image be blocked is divided into, in, after lower three parts, the human body parts just not having particular persons in the middle parts of images of human body image after dividing and/or lower part image may be caused, in the process of carrying out degree of depth study, so do not have the image of human body parts greatly can affect the comparison result of human body image, reduce the accuracy rate of human body image comparison.Based on this, embodiments provide a kind of human body image comparison method and device.
See Fig. 1, it illustrates the structural representation of a kind of implementation system involved by human body image comparison method that the embodiment of the present invention provides, this system comprises: human body image comparison equipment 10, and human body image comparison equipment 10 comprises: human body image comparison device 100 and carry out the image library 101 of data interaction with human body image comparison device 100.
Wherein, human body image comparison device 100, for obtaining the first volume image and the second human body image; The first volume image is divided into multiple image-region; From multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage; Degree of depth study is carried out to the multiple human body subimage obtained and the second human body image, obtains the feature difference figure of the first volume image and the second human body image; Degree of depth study is carried out to the feature difference figure obtained, obtain the comparison result of the first volume image and the second human body image, when all not identified the first volume image by the human body image in comparison determination image library 101, the first volume image is sent to image library 101; Image library 101, the first volume image sent for recipient's volume image comparison device 100 also stores.
Human body image comparison equipment 10 can adopt the server of existing disposable type or computing equipment to compare to human body image, repeats no longer one by one here.
Human body image comparison device 100 can adopt existing any central processing unit, microprocessor or programming device to compare to human body image, repeats no longer one by one here.
Image library 101 can adopt existing any large-capacity storage media to store human body image, repeats no longer one by one here.
Embodiment 1
See Fig. 2, the present embodiment provides a kind of human body image comparison method, comprises the following steps:
Step 200, obtain the first volume image and the second human body image.
The first volume image, be the image by the width that the input equipment of human body image comparison equipment is selected from the frame of video that monitor video is taken by Security Personnel with personage, the first volume image can be the image that suspect or missing crew etc. need the particular persons determined one's identity.
The input equipment of human body image comparison equipment can be that mouse, keyboard or Trackpad etc. can make Security Personnel from frame of video, select the external unit of any computing machine of character image, repeats no longer one by one here.
Second human body image is any character image stored in image library, can be the image that the whole body of a personage takes or picture etc. can characterize people's figure and features feature above the waist.
Comprise in the first volume image and the second human body image and only comprise a character image, so the process of the human body image comparison described in the present embodiment is the process of single another single human body image of human body image comparison.
Step 202, the first volume image is divided into multiple image-region.
By existing any image division methods, the first volume image is divided into the image-region of predetermined number, repeats no longer one by one here.Wherein, predetermined number can be greater than 3 natural number, so the first volume image generally can be divided into the image-region of multiple quantity such as 4,5,6 or 7 in the present embodiment.
Step 204, from multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage.
Human body subimage, comprise the different piece of human body in the first volume image respectively, such as after carrying out image-region splicing, the upper body portion that somebody's body subimage comprises the head of human body in the first volume image, somebody's body subimage comprises human body in the first volume image, and somebody's body subimage comprises whole parts of human body in the first volume image.
When part human body in the first volume image is blocked, when in such as the first volume image, the lower body portion of human body is blocked, the human body subimage that the multiple human body subimages so obtained comprise the upper body portion of human body in the first volume image is exactly comprise the minimum human body subimage of maximum, the non-human part of human body parts, thus by carrying out image ratio pair to this human body subimage and the second human body image, the impact of image in human body image comparison process of the part that is blocked can be reduced in the first volume image.
Step 206, degree of depth study is carried out to the multiple human body subimage obtained and the second human body image, obtain the feature difference figure of the first volume image and the second human body image.
Carry out degree of depth study to multiple human body subimage and the second human body image by having mutually isostructural two sub-convolutional neural networks, these two sub-convolutional neural networks are formed by identical primary image process arithmetic element, convolution algorithm unit and down-sampling arithmetic element.
Feature difference figure, be the image that can represent the first volume image and the second human body image similarity degree, each pixel value wherein in feature difference figure is more tending towards 0, and the similarity degree of the first volume image and the second human body image is higher.
Step 208, carry out degree of depth study to the feature difference figure obtained, obtain the comparison result of the first volume image and the second human body image, it is dissimilar that comparison result comprises phase Sihe.
Use, with the sub-convolutional neural networks above-mentioned human body subimage and the second human body image being carried out to the sub-convolutional neural networks different structure of degree of depth study, degree of depth study is carried out to feature difference figure.
Sub-convolutional neural networks feature difference figure being carried out to degree of depth study is made up of primary image process arithmetic element, convolution algorithm unit, down-sampling arithmetic element and softmax sorter.
In sum, the human body image comparison method that the present embodiment provides, by being divided into multiple image-region by the first volume image, and according to the multiple image-regions after division, from multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage, then the multiple human body subimages by obtaining, compare with the second human body image respectively, obtain the result of comparison, with prior art, human body image is divided into, in, the human body image comparison process that the part that is blocked in each several part human body image after lower three parts can reduce human body image comparison result accuracy rate is compared, reduce in human body image the impact of part on comparison result that be blocked, improve the accuracy rate of human body image comparison.
In order to the image-region after being divided by the first volume image obtains multiple human body subimage, particularly, from multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage and comprise the following steps:
From multiple image-region, select k image-region to splice respectively, obtain multiple human body subimage, multiple image-region is on average divided from top to bottom by the first volume image and obtains;
Wherein,
m represents the image-region quantity that the first volume image divides.
Spliced image-region by following example, the process obtaining multiple human body subimage conducts further description:
Arrange m=6, the first for input volume image is on average divided into 6 image-regions from top to bottom, these 6 image-regions are expressed as R
1, R
2, R
3, R
4, R
5and R
6.According to
determine from 6 image-regions, select 3 to 6 image-regions to form different human body subimages respectively.Wherein, by image-region R
1, R
2and R
3form human body subimage
(being namely upper 1/2 part of the first volume image), by region R
1, R
2, R
3and R
4form human body subimage
(being namely upper 2/3 part of the first volume image), by image-region R
1, R
2, R
3, R
4and R
5form human body subimage
(being namely upper 5/6 part of the first volume image), by image-region R
1, R
2, R
3, R
4, R
5and R
6form human body subimage
(being namely the first volume image); After so the first volume image being processed, obtain altogether the human body subimage of 4 different sizes
on average be divided into 6 tectonic images
cover local message and the global information of the first volume image, and each Zhang Renti subimage can comprise the upper body portion (in pedestrian retrieval, upper half of human body is larger to Influence on test result) of human body image in the first volume image, also can not affect the speed obtaining comparison result too much due to human body subimage quantity simultaneously.
In sum, by from multiple image-region, varying number is got respectively
image-region splice, thus obtain multiple human body subimage, can be blocked when to only have some people volume image in image by human body in the first volume image, can by comparing to the multiple human body subimages obtained, the part removing irrelevant human body in image as far as possible, on the impact of human body image comparison, improves the success ratio to human body image comparison.
In correlation technique, in the process of carrying out human body image comparison, degree of depth study is carried out again after needs divide the human body image obtained from image library, thus add the processing time of human body image comparison, in order to the processing time length of human body image comparison can be reduced, so carry out degree of depth study to the multiple human body subimage obtained and the second human body image, the feature difference figure obtaining the first volume image and the second human body image comprises the following steps 1 to step 2:
(1) degree of depth study is carried out to multiple human body subimage and the second human body image, obtain multiple the first volume image characteristic pattern and the second human body image characteristic pattern;
(2) according to the multiple the first volume image characteristic pattern obtained and the second human body image characteristic pattern, the feature difference figure of each the first volume image characteristic pattern respectively with the second human body image characteristic pattern in multiple the first volume image characteristic pattern is obtained.
Particularly, step 1 comprises the following steps 11 to 12:
(11) from convolutional neural networks, degree of depth study is carried out to the human body subimage obtained by first, obtain multiple the first volume image characteristic pattern;
(12) carry out degree of depth study from convolutional neural networks to arbitrary second human body image selected from image library by second, obtain the second human body image characteristic pattern, second is identical from the structure of convolutional neural networks with first from convolutional neural networks.
Wherein, first identical primary image process arithmetic element, convolution algorithm unit and down-sampling arithmetic element is included from convolutional neural networks and second from convolutional neural networks.
Certainly, first can also use other can realize any arithmetic element of picture depth learning functionality from convolutional neural networks and second in convolutional neural networks, to substitute at least one arithmetic element in above-mentioned primary image process arithmetic element, convolution algorithm unit and down-sampling arithmetic element, repeat no longer one by one here.
In correlation technique, due to the difference of the posture of human body when shooting angle and shooting, same people can be made to seem in different images, and difference is very large, so in these cases, just effectively can not identify and comparison the human body in image, in order to effectively identify the same human body in different images with different gestures, step 2 comprises the following steps 21 to 24:
(21) with each pixel coordinate of storing in the pixel coordinate set preset for eigenwert regional center, according to the eigenwert area size preset, ex-first lady's volume image characteristic pattern will be worked as respectively and the second human body image characteristic pattern is divided into multiple the First Eigenvalue region and multiple Second Eigenvalue region;
(22) from multiple the First Eigenvalue region and multiple Second Eigenvalue region, profile maxima is obtained respectively;
(23) calculate respectively from the difference with the profile maxima obtained the First Eigenvalue region of same characteristic features value regional center and Second Eigenvalue region, obtain multiple feature difference;
(24) with multiple feature difference for pixel value, according to the feature difference figure size preset, generate the feature difference figure when ex-first lady's volume image characteristic pattern and the second human body image characteristic pattern.
By example below, the content that step 21 describes to step 22 is described further:
See Fig. 3, show a kind of form of expression of human body image characteristic pattern, the coordinate of the pixel shown in figure is the characteristic area center in multiple eigenwert region, and the size arranging each eigenwert region is the size of 3 × 3.Therefore, image according to Fig. 3, with coordinate (2,2), (4,2), (6,2), (8,2), (2,4), (4,4), (6,4), (8,4), (2,6), (4,6), (6,6), (8,6), (2,8), (4,8), (6,8), (8,8) are characteristic area center, size is the maximum eigenwert found out respectively in the characteristic area of 3x3 in human body image characteristic pattern in each characteristic area.
Wherein, the eigenwert area size that the First Eigenvalue region and Second Eigenvalue region are preset is n × n-pixel size, n ∈ { 3,5,7,9,11}.
Can be found out to step 24 by above-mentioned step 21, the feature difference figure of the first volume image characteristic pattern and the second human body image characteristic pattern can be obtained, when the pixel value of each pixel in feature difference figure is tending towards 0, illustrate that the similarity degree of the first volume image and the second human body image is higher, thus seem that difference is very greatly that the human body of same people effectively identifies in fact to causing in different photo due to the difference of posture of human body when shooting angle and shooting.
In sum, in the process of human body image comparison, directly degree of depth study is carried out to the second human body image, carrying out degree of depth study again without the need to being carried out dividing by the second human body image, calculated amount during image ratio pair can be reduced, improve the comparison speed of human body image.
In correlation technique, after the degree of depth learning outcome obtaining each several part in human body image to be detected and comparison people image respectively, must operate through the Fusion Features of more complicated, just can obtain the comparison result of human body image to be detected and comparison people image, in order to the comparison result of human body image can be obtained faster, carry out degree of depth study to the feature difference figure obtained, the comparison result obtaining the first volume image and the second human body image comprises the following steps 1 to step 4:
(1) degree of depth study is carried out to each feature difference figure obtained, obtain the similar parameter of human body subimage corresponding to each feature difference figure and the second human body image;
(2) similarity of maximum similar parameter as the first volume image and the second human body image is determined;
(3) when similarity is more than or equal to the similarity threshold of setting, obtain the comparison result that the first volume image is similar to the second human body image, similar comparison result is represented by numeral 1;
(4) when similarity is less than the similarity threshold of setting, obtain the first volume image and the dissimilar comparison result of the second human body image, dissimilar comparison result is represented by numeral 0.
Particularly, step 1 comprises: draw the similar parameter of often opening human body subimage corresponding to feature difference figure and the second human body image by the 3rd from convolutional neural networks.
3rd sub-convolutional neural networks is made up of primary image process arithmetic element, convolution algorithm unit, down-sampling arithmetic element and softmax sorter.
Certainly, 3rd can also use other can realize any arithmetic element of picture depth learning functionality in convolutional neural networks, to substitute at least one arithmetic element in above-mentioned primary image process arithmetic element, convolution algorithm unit, down-sampling arithmetic element and softmax sorter, repeat no longer one by one here.
In sum, by the degree of depth study and simple numeric ratio to operation, just can determine that whether the first volume image similar to the second human body image, accelerate the comparison speed of human body image.
Embodiment 2
See Fig. 3, the present embodiment provides another kind of human body image comparison method, comprises the following steps:
(1) inputting human body image A and human body image B, image A is current human body image to be retrieved, and image B is any human body image in image library;
(2) Region dividing: input picture A is on average divided into m region from top to bottom, this m region representation is R
jj={1,2,3 ..., m-1, m}, get front k
individual region R
j, j={1,2,3 ..., k-1, k} form subimage
can form so altogether
it is right to represent
round up) individual subimage
As m=6, obtain following Region dividing example:
Input human body image A is on average divided into 6 regions from top to bottom, and these 6 regions are expressed as R
1, R
2, R
3, R
4, R
5, R
6.By region R
1, R
2, R
3form subimage
(being namely upper 1/2 of input human body image A), by region R
1, R
2, R
3, R
4form subimage
(being namely upper 2/3 of input human body image A), by region R
1, R
2, R
3, R
4, R
5form subimage
(being namely upper 5/6 of input human body image A), by region R
1, R
2, R
3, R
4, R
5, R
6form subimage
(being namely input human body image A); Like this human body subimage that Region dividing obtains altogether 4 different sizes is carried out to input human body image A
on average be divided into 6 tectonic images
cover local message and the global information of input picture, and each image
the main contents (in pedestrian retrieval, upper half of human body is larger to Influence on test result) of input picture can be comprised, simultaneously also can not due to image
affect processing speed too much.
(3) by human body image
be input to sub-convolutional neural networks C1 and C2 with human body image B and carry out characteristic pattern extraction (sub-convolutional neural networks C1 and C2 has identical structure), output characteristic figure f and characteristic pattern g; Sub-convolutional neural networks C1 and C2 is by primary image process computing, and convolution algorithm and down-sampling computing are formed;
(4) calculate the feature difference figure of characteristic pattern f and characteristic pattern g: get the neighborhood window that size is n × n, with 2 pixels for step-length on characteristic pattern f from top to bottom, from left to right slip neighborhood window, asks for maximal value V in neighborhood window
f; With 2 pixels for step-length on characteristic pattern g from top to bottom, from left to right slip neighborhood window, asks for maximal value V in neighborhood window
g; Do differ from and ask for absolute value, be i.e. V to the difference of trying to achieve above
abs=abs (V
f-V
g); Until process all position feature values on characteristic pattern f and characteristic pattern g, as shown in Equation 1, concrete steps:
Get the neighborhood window that size is n × n, with 2 pixels for step-length on characteristic pattern f and characteristic pattern g from top to bottom, from left to right slip neighborhood window, perform step (a), (b), (c) respectively, until process all position feature values on characteristic pattern f and characteristic pattern g.
A () asks for the maximal value V of characteristic pattern f and characteristic pattern g in the n × n neighborhood centered by coordinate (x, y) respectively
fand V
g;
B () is to above-mentioned V
fand V
gdiffer from, i.e. V=V
f-V
g;
C () asks for absolute value to above-mentioned difference, i.e. V
abs=abs (V);
K(x,y)＝abs(max(N(f(x,y)))-max(N(g(x,y))))(1)
Wherein, the feature difference figure of K (x, y) denotation coordination position (x, y), N (f (x, y)) ∈ R
n × nn × n neighborhood of the eigenwert f (x, y) centered by coordinate (x, y), N (g (x, y)) ∈ R
n × nwith coordinate (x, y) the eigenwert g (x centered by, y) n × n neighborhood, max (N (f (x, y))) n × n neighborhood N (f (x is asked in expression, y) maximal value), n × n neighborhood N (g (x is asked in max (N (g (x, y))) expression, y) maximal value), abs (z) expression asks for absolute value to z, and in the present embodiment, the span of n is n={3, and 5,7, the representative value of 9,11}, n is n=5;
(5) feature difference figure K is inputted sub-convolutional neural networks C3 extract feature and classify (degree of depth learning network is a kind of self-learning networks, feature be self study out, be not artificial selection.); Sub-convolutional neural networks C3 is made up of primary image process computing, convolution algorithm, down-sampling computing and softmax sorter;
(6) Output rusults: output is a binary variable, 0 represents that input human body image A and human body image B is dissimilar, and 1 represents that input human body image A is similar to human body image B.
In sum, the human body image comparison method that the present embodiment provides, by being divided into multiple image-region by the first volume image, and according to the multiple image-regions after division, from multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage, then the multiple human body subimages by obtaining, compare with the second human body image respectively, obtain the result of comparison, with prior art, human body image is divided into, in, the human body image comparison process that the part that is blocked in each several part human body image after lower three parts can reduce human body image comparison result accuracy rate is compared, reduce in human body image the impact of part on comparison result that be blocked, improve the accuracy rate of human body image comparison.
Embodiment 3
See Fig. 5, the present embodiment provides a kind of human body image comparison device, for performing above-mentioned human body image comparison method, comprising: acquisition module 500, image divide module 502, image mosaic module 504, feature difference figure acquisition module 506 and comparing module 508.
Acquisition module 500, for obtaining the first volume image and the second human body image;
Image divides module 502, is connected, for the first volume image is divided into multiple image-region with acquisition module 500;
Image mosaic module 504, divides module 502 with image and is connected, and for selecting the image-region of varying number to splice from multiple image-region, obtains multiple human body subimage;
Feature difference figure acquisition module 506, is connected with image mosaic module 504, for carrying out degree of depth study to the multiple human body subimage obtained and the second human body image, obtains the feature difference figure of the first volume image and the second human body image;
Comparing module 508, is connected with feature difference figure acquisition module 506, for carrying out degree of depth study to the feature difference figure obtained, obtains the comparison result of the first volume image and the second human body image, and it is dissimilar that comparison result comprises phase Sihe.
In order to the image-region after being divided by the first volume image obtains multiple human body subimage, image mosaic module 504 is specifically for from multiple image-region, k image-region is selected to splice respectively, obtain multiple human body subimage, multiple image-region is on average divided from top to bottom by the first volume image and obtains;
Wherein,
m represents the image-region quantity that the first volume image divides.
In sum, by from multiple image-region, varying number is got respectively
image-region splice, thus obtain multiple human body subimage, can be blocked by human body in the first volume image, when only having some people volume image in image, can by comparing to the multiple human body subimages obtained, the part removing irrelevant human body in image as far as possible, on the impact of human body image comparison, improves the success ratio to human body image comparison.
In correlation technique, in the process of carrying out human body image comparison, degree of depth study is carried out again after needs divide the human body image obtained from image library, thus add the processing time of human body image comparison, in order to reduce the processing time length of human body image comparison, feature difference figure acquisition module 506 comprises:
Degree of depth unit, for carrying out degree of depth study to multiple human body subimage and the second human body image, obtains multiple the first volume image characteristic pattern and the second human body image characteristic pattern;
Feature difference figure acquiring unit, for according to the multiple the first volume image characteristic pattern obtained and the second human body image characteristic pattern, obtain the feature difference figure of each the first volume image characteristic pattern respectively with the second human body image characteristic pattern in multiple the first volume image characteristic pattern.
In sum, in the process of human body image comparison, directly degree of depth study is carried out to the second human body image, carrying out degree of depth study again without the need to being carried out dividing by the second human body image, calculated amount during image ratio pair can be reduced, improve the comparison speed of human body image.
In correlation technique, due to the difference of the posture of human body when shooting angle and shooting, same people can be made to seem in different images, and difference is very large, so in these cases, just effectively can not identify and comparison the human body in image, in order to effectively identify the same human body in different images with different gestures, feature difference figure acquiring unit comprises:
Region dividing subelement, for with each pixel coordinate of storing in the pixel coordinate set preset for eigenwert regional center, according to the eigenwert area size preset, ex-first lady's volume image characteristic pattern will be worked as respectively and the second human body image characteristic pattern is divided into multiple the First Eigenvalue region and multiple Second Eigenvalue region;
Profile maxima obtains subelement, for obtaining profile maxima respectively from multiple the First Eigenvalue region and multiple Second Eigenvalue region;
Feature difference computation subunit, for calculating respectively from the difference with the profile maxima obtained the First Eigenvalue region of same characteristic features value regional center and Second Eigenvalue region, obtains multiple feature difference;
Feature difference figure generates subelement, for multiple feature difference for pixel value, according to the feature difference figure size preset, generate the feature difference figure when ex-first lady's volume image characteristic pattern and the second human body image characteristic pattern.
In sum, the feature difference figure of the first volume image characteristic pattern and the second human body image characteristic pattern can be obtained, when the pixel value of each pixel in feature difference figure is tending towards 0, illustrate that the similarity degree of the first volume image and the second human body image is higher, thus seem that difference is very greatly that the human body of same people effectively identifies in fact to causing in different photo due to the difference of posture of human body when shooting angle and shooting.
In correlation technique, after the degree of depth learning outcome obtaining each several part in human body image to be detected and comparison people image respectively, must operate through the Fusion Features of more complicated, just can obtain the comparison result of human body image to be detected and comparison people image, in order to obtain the comparison result of human body image faster, comparing module 508 comprises:
Similar parameter computing unit, for carrying out degree of depth study to each feature difference figure obtained, obtains the similar parameter of human body subimage corresponding to each feature difference figure and the second human body image;
Similarity determining unit, for determining the similarity of maximum similar parameter as the first volume image and the second human body image;
First comparison result determining unit, during for being more than or equal to the similarity threshold of setting when similarity, obtains the comparison result that the first volume image is similar to the second human body image;
Second comparison result determining unit, during for being less than the similarity threshold of setting when similarity, obtains the first volume image and the dissimilar comparison result of the second human body image.
In sum, by the degree of depth study and simple numeric ratio to operation, just can determine that whether the first volume image similar to the second human body image, accelerate the comparison speed of human body image.
In sum, the human body image comparison device that the present embodiment provides, by being divided into multiple image-region by the first volume image, and according to the multiple image-regions after division, from multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage, then the multiple human body subimages by obtaining, compare with the second human body image respectively, obtain the result of comparison, with prior art, human body image is divided into, in, the human body image comparison process that the part that is blocked in each several part human body image after lower three parts can reduce human body image comparison result accuracy rate is compared, reduce in human body image the impact of part on comparison result that be blocked, improve the accuracy rate of human body image comparison.
The computer program carrying out human body image comparison method that the embodiment of the present invention provides, comprise the computer-readable recording medium storing program code, the instruction that described program code comprises can be used for performing the method described in previous methods embodiment, specific implementation see embodiment of the method, can not repeat them here.
Those skilled in the art can be well understood to, and for convenience and simplicity of description, the specific works process of the system of foregoing description, device and unit, with reference to the corresponding process in preceding method embodiment, can not repeat them here.
In several embodiments that the application provides, should be understood that disclosed system, apparatus and method can realize by another way.Device embodiment described above is only schematic, such as, the division of described unit, be only a kind of logic function to divide, actual can have other dividing mode when realizing, again such as, multiple unit or assembly can in conjunction with or another system can be integrated into, or some features can be ignored, or do not perform.Another point, shown or discussed coupling each other or direct-coupling or communication connection can be by some communication interfaces, and the indirect coupling of device or unit or communication connection can be electrical, machinery or other form.
The described unit illustrated as separating component or can may not be and physically separates, and the parts as unit display can be or may not be physical location, namely can be positioned at a place, or also can be distributed in multiple network element.Some or all of unit wherein can be selected according to the actual needs to realize the object of the present embodiment scheme.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, also can be that the independent physics of unit exists, also can two or more unit in a unit integrated.
If described function using the form of SFU software functional unit realize and as independently production marketing or use time, can be stored in a computer read/write memory medium.Based on such understanding, the part of the part that technical scheme of the present invention contributes to prior art in essence in other words or this technical scheme can embody with the form of software product, this computer software product is stored in a storage medium, comprising some instructions in order to make a computer equipment (can be personal computer, server, or the network equipment etc.) perform all or part of step of method described in each embodiment of the present invention.And aforesaid storage medium comprises: USB flash disk, portable hard drive, ROM (read-only memory) (ROM, Read-OnlyMemory), random access memory (RAM, RandomAccessMemory), magnetic disc or CD etc. various can be program code stored medium.
The above; be only the specific embodiment of the present invention, but protection scope of the present invention is not limited thereto, is anyly familiar with those skilled in the art in the technical scope that the present invention discloses; change can be expected easily or replace, all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should described be as the criterion with the protection domain of claim.
Claims (10)
1. a human body image comparison method, is characterized in that, comprising:
Obtain the first volume image and the second human body image;
Described the first volume image is divided into multiple image-region;
From described multiple image-region, select the image-region of varying number to splice, obtain multiple human body subimage;
Degree of depth study is carried out to the described multiple human body subimage obtained and described second human body image, obtains the feature difference figure of described the first volume image and described second human body image;
Carry out degree of depth study to the described feature difference figure obtained, obtain the comparison result of described the first volume image and described second human body image, it is dissimilar that described comparison result comprises phase Sihe.
2. method according to claim 1, is characterized in that, from described multiple image-region, select the image-region of varying number to splice, and obtains multiple human body subimage and comprises:
From multiple described image-region, select k image-region to splice respectively, obtain multiple human body subimage, multiple described image-region is on average divided from top to bottom by described the first volume image and obtains;
Wherein,
m represents the image-region quantity that described the first volume image divides.
3. method according to claim 1, is characterized in that, carry out degree of depth study to the described multiple human body subimage obtained and described second human body image, the feature difference figure obtaining described the first volume image and described second human body image comprises:
Degree of depth study is carried out to multiple described human body subimage and described second human body image, obtains multiple the first volume image characteristic pattern and the second human body image characteristic pattern;
According to the multiple described the first volume image characteristic pattern obtained and described second human body image characteristic pattern, obtain each the first volume image characteristic pattern in described multiple the first volume image characteristic pattern respectively with the feature difference figure of described second human body image characteristic pattern.
4. method according to claim 3, it is characterized in that, process the described the first volume image characteristic pattern obtained and described second human body image characteristic pattern, each the first volume image characteristic pattern obtained in described multiple the first volume image characteristic pattern comprises with the feature difference figure of described second human body image characteristic pattern respectively:
With each pixel coordinate stored in the pixel coordinate set preset for eigenwert regional center, according to the eigenwert area size preset, ex-first lady's volume image characteristic pattern will be worked as respectively and described second human body image characteristic pattern is divided into multiple the First Eigenvalue region and multiple Second Eigenvalue region;
Profile maxima is obtained respectively from described multiple the First Eigenvalue region and described multiple Second Eigenvalue region;
Calculate respectively from the difference with the profile maxima obtained the First Eigenvalue region of same characteristic features value regional center and Second Eigenvalue region, obtain multiple feature difference;
With described multiple feature difference for pixel value, according to the feature difference figure size preset, generate the described feature difference figure working as ex-first lady's volume image characteristic pattern and described second human body image characteristic pattern.
5. method according to claim 1, is characterized in that, carry out degree of depth study to the described feature difference figure obtained, the comparison result obtaining described the first volume image and described second human body image comprises:
Degree of depth study is carried out to each feature difference figure obtained, obtains the similar parameter of human body subimage corresponding to each described feature difference figure and described second human body image;
Determine the similarity of maximum similar parameter as described the first volume image and described second human body image;
When described similarity is more than or equal to the similarity threshold of setting, obtain the comparison result that described the first volume image is similar to described second human body image;
When described similarity is less than the similarity threshold of setting, obtain described the first volume image and the dissimilar comparison result of described second human body image.
6. a human body image comparison device, is characterized in that, comprising:
Acquisition module, for obtaining the first volume image and the second human body image;
Image divides module, for described the first volume image is divided into multiple image-region;
Image mosaic module, for selecting the image-region of varying number to splice from described multiple image-region, obtains multiple human body subimage;
Feature difference figure acquisition module, for carrying out degree of depth study to the described multiple human body subimage obtained and described second human body image, obtains the feature difference figure of described the first volume image and described second human body image;
Comparing module, for carrying out degree of depth study to the described feature difference figure obtained, obtains the comparison result of described the first volume image and described second human body image, and it is dissimilar that described comparison result comprises phase Sihe.
7. device according to claim 6, it is characterized in that, described image mosaic module is specifically for from multiple described image-region, k image-region is selected to splice respectively, obtain multiple human body subimage, multiple described image-region is on average divided from top to bottom by described the first volume image and obtains;
Wherein,
m represents the image-region quantity that described the first volume image divides.
8. device according to claim 6, is characterized in that, described feature difference figure acquisition module comprises:
Degree of depth unit, for carrying out degree of depth study to multiple described human body subimage and described second human body image, obtains multiple the first volume image characteristic pattern and the second human body image characteristic pattern;
Feature difference figure acquiring unit, for according to the multiple described the first volume image characteristic pattern that obtains and described second human body image characteristic pattern, obtain each the first volume image characteristic pattern in described multiple the first volume image characteristic pattern respectively with the feature difference figure of described second human body image characteristic pattern.
9. device according to claim 8, is characterized in that, described feature difference figure acquiring unit comprises:
Region dividing subelement, for with each pixel coordinate of storing in the pixel coordinate set preset for eigenwert regional center, according to the eigenwert area size preset, ex-first lady's volume image characteristic pattern will be worked as respectively and described second human body image characteristic pattern is divided into multiple the First Eigenvalue region and multiple Second Eigenvalue region;
Profile maxima obtains subelement, for obtaining profile maxima respectively from described multiple the First Eigenvalue region and described multiple Second Eigenvalue region;
Feature difference computation subunit, for calculating respectively from the difference with the profile maxima obtained the First Eigenvalue region of same characteristic features value regional center and Second Eigenvalue region, obtains multiple feature difference;
Feature difference figure generates subelement, for described multiple feature difference for pixel value, according to the feature difference figure size preset, generate the described feature difference figure when ex-first lady's volume image characteristic pattern and described second human body image characteristic pattern.
10. device according to claim 6, is characterized in that, described comparing module comprises:
Similar parameter computing unit, for carrying out degree of depth study to each feature difference figure obtained, obtains the similar parameter of human body subimage corresponding to each described feature difference figure and described second human body image;
Similarity determining unit, for determining the similarity of maximum similar parameter as described the first volume image and described second human body image;
First comparison result determining unit, for when described similarity is more than or equal to the similarity threshold of setting, obtains the comparison result that described the first volume image is similar to described second human body image;
Second comparison result determining unit, for when described similarity is less than the similarity threshold of setting, obtains described the first volume image and the dissimilar comparison result of described second human body image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510742746.4A CN105243395B (en) | 2015-11-04 | 2015-11-04 | A kind of human body image comparison method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510742746.4A CN105243395B (en) | 2015-11-04 | 2015-11-04 | A kind of human body image comparison method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105243395A true CN105243395A (en) | 2016-01-13 |
| CN105243395B CN105243395B (en) | 2018-10-19 |
Family
ID=55041036
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510742746.4A Active CN105243395B (en) | 2015-11-04 | 2015-11-04 | A kind of human body image comparison method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105243395B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106599860A (en) * | 2016-12-20 | 2017-04-26 | 北京小米移动软件有限公司 | Human face detection method and device |
| CN106711865A (en) * | 2017-02-21 | 2017-05-24 | 国网山东省电力公司邹城市供电公司 | Cable digging tool |
| CN106707109A (en) * | 2017-02-21 | 2017-05-24 | 国网山东省电力公司邹城市供电公司 | Underground cable detection system |
| CN106815590A (en) * | 2017-02-21 | 2017-06-09 | 国网山东省电力公司邹城市供电公司 | A kind of cable trough and cable management system |
| CN106875385A (en) * | 2017-02-09 | 2017-06-20 | 广州中国科学院软件应用技术研究所 | A kind of high robust region intrusion detection algorithm |
| CN106982741A (en) * | 2017-04-06 | 2017-07-28 | 南京三宝弘正视觉科技有限公司 | A kind of pet supervisory-controlled robot and system |
| CN107009373A (en) * | 2017-04-06 | 2017-08-04 | 南京三宝弘正视觉科技有限公司 | A kind of solitary's supervisory-controlled robot and system |
| CN107014420A (en) * | 2017-04-14 | 2017-08-04 | 南京三宝弘正视觉科技有限公司 | A kind of sensor detection electronic pen and system |
| CN107402947A (en) * | 2017-03-29 | 2017-11-28 | 北京粉笔未来科技有限公司 | Picture retrieval method for establishing model and device, picture retrieval method and device |
| CN107545221A (en) * | 2016-06-28 | 2018-01-05 | 北京京东尚科信息技术有限公司 | Baby kicks quilt recognition methods, system and device |
| CN107798285A (en) * | 2016-08-31 | 2018-03-13 | 富士施乐株式会社 | Image processing apparatus and image processing method |
| CN108509961A (en) * | 2017-02-27 | 2018-09-07 | 北京旷视科技有限公司 | Image processing method and device |
| CN112308102A (en) * | 2019-08-01 | 2021-02-02 | 北京易真学思教育科技有限公司 | Image similarity calculation method, calculation device, and storage medium |
| CN112712489A (en) * | 2020-12-31 | 2021-04-27 | 北京澎思科技有限公司 | Method, system and computer readable storage medium for image processing |
-
2015
- 2015-11-04 CN CN201510742746.4A patent/CN105243395B/en active Active
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107545221A (en) * | 2016-06-28 | 2018-01-05 | 北京京东尚科信息技术有限公司 | Baby kicks quilt recognition methods, system and device |
| CN107798285A (en) * | 2016-08-31 | 2018-03-13 | 富士施乐株式会社 | Image processing apparatus and image processing method |
| CN106599860B (en) * | 2016-12-20 | 2019-11-26 | 北京小米移动软件有限公司 | A kind of method and apparatus of Face datection |
| CN106599860A (en) * | 2016-12-20 | 2017-04-26 | 北京小米移动软件有限公司 | Human face detection method and device |
| CN106875385A (en) * | 2017-02-09 | 2017-06-20 | 广州中国科学院软件应用技术研究所 | A kind of high robust region intrusion detection algorithm |
| CN106815590A (en) * | 2017-02-21 | 2017-06-09 | 国网山东省电力公司邹城市供电公司 | A kind of cable trough and cable management system |
| CN106707109A (en) * | 2017-02-21 | 2017-05-24 | 国网山东省电力公司邹城市供电公司 | Underground cable detection system |
| CN106711865A (en) * | 2017-02-21 | 2017-05-24 | 国网山东省电力公司邹城市供电公司 | Cable digging tool |
| CN108509961A (en) * | 2017-02-27 | 2018-09-07 | 北京旷视科技有限公司 | Image processing method and device |
| CN107402947A (en) * | 2017-03-29 | 2017-11-28 | 北京粉笔未来科技有限公司 | Picture retrieval method for establishing model and device, picture retrieval method and device |
| CN107009373A (en) * | 2017-04-06 | 2017-08-04 | 南京三宝弘正视觉科技有限公司 | A kind of solitary's supervisory-controlled robot and system |
| CN106982741A (en) * | 2017-04-06 | 2017-07-28 | 南京三宝弘正视觉科技有限公司 | A kind of pet supervisory-controlled robot and system |
| CN107014420A (en) * | 2017-04-14 | 2017-08-04 | 南京三宝弘正视觉科技有限公司 | A kind of sensor detection electronic pen and system |
| CN112308102A (en) * | 2019-08-01 | 2021-02-02 | 北京易真学思教育科技有限公司 | Image similarity calculation method, calculation device, and storage medium |
| CN112712489A (en) * | 2020-12-31 | 2021-04-27 | 北京澎思科技有限公司 | Method, system and computer readable storage medium for image processing |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105243395B (en) | 2018-10-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105243395A (en) | Human body image comparison method and device | |
| Wang et al. | Adaptive fusion for RGB-D salient object detection | |
| US9396415B2 (en) | Neural network image representation | |
| US9418458B2 (en) | Graph image representation from convolutional neural networks | |
| CN107169463B (en) | Method for detecting human face, device, computer equipment and storage medium | |
| CN105593901B (en) | Training data generation device, method, and program, and crowd state recognition device, method, and program | |
| US20190279028A1 (en) | Method and Apparatus for Object Re-identification | |
| CN107844753A (en) | Pedestrian in video image recognition methods, device, storage medium and processor again | |
| US10325184B2 (en) | Depth-value classification using forests | |
| EP1631933A1 (en) | Human figure contour outlining in images | |
| US20160098842A1 (en) | Method and system for unsupervised image segmentation using a trained quality metric | |
| US20140270479A1 (en) | Systems and methods for parameter estimation of images | |
| Palou et al. | Monocular depth ordering using T-junctions and convexity occlusion cues | |
| Yigitbasi et al. | Edge detection using artificial bee colony algorithm (ABC) | |
| US20170061644A1 (en) | Image analyzer, image analysis method, computer program product, and image analysis system | |
| KR102111667B1 (en) | Apparatus of generating 2d image data set for deep learning through 3d design drawing data processing and system for searching simillar design drawing data based on deep learning using the same | |
| KR20190126857A (en) | Detect and Represent Objects in Images | |
| EP3885979A1 (en) | Method and apparatus for detecting liveness based on phase difference | |
| Goudelis et al. | Fall detection using history triple features | |
| CN108234821A (en) | Detect the methods, devices and systems of the action in video | |
| CN115048969A (en) | Visual analysis system for evaluating, understanding and improving deep neural networks | |
| EP3748460A1 (en) | Search system, search method, and program | |
| CN112734747A (en) | Target detection method and device, electronic equipment and storage medium | |
| Geetha et al. | An improved method for segmentation of point cloud using minimum spanning tree | |
| Merentitis et al. | Automatic fusion and classification of hyperspectral and LiDAR data using random forests |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PP01 | Preservation of patent right | ||
| PP01 | Preservation of patent right |
Effective date of registration: 20220726 Granted publication date: 20181019 |