CN106845550A - A kind of image-recognizing method based on multi-template - Google Patents
A kind of image-recognizing method based on multi-template Download PDFInfo
- Publication number
- CN106845550A CN106845550A CN201710056968.XA CN201710056968A CN106845550A CN 106845550 A CN106845550 A CN 106845550A CN 201710056968 A CN201710056968 A CN 201710056968A CN 106845550 A CN106845550 A CN 106845550A
- Authority
- CN
- China
- Prior art keywords
- image
- region
- template
- axis direction
- templates
- 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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image-recognizing method, including：Receive image to be identified；Convolutional calculation is carried out to described image simultaneously using two templates, so as to each region for obtaining described image category score corresponding with described two templates；Judge whether include recognized object in image-region based on the category score.
Description
Technical field
The present invention relates to image processing field, more particularly to a kind of image-recognizing method based on multi-template.
Background technology
Convolutional neural networks (Convolutional Neural Network, CNN) are a kind of feedforward neural networks, with biography
The BP neural network of system is compared, and has the advantages that recognition efficiency is high, it is good to rotate scaling consistency, in Digital Image Processing and people
The every field such as face identification are widely used.
The application principle of traditional convolution neural network model is：First, the attribute design convolution god according to image to be entered
Through network template framework, designed convolutional neural networks template framework is sandwich construction, including 1 input layer, in input layer
Afterwards, several convolutional layers and several down-sampled layers are placed with various orders, are finally output layer.Wherein, input layer is used
In reception original image；Each convolutional layer includes the characteristic pattern of multiple same sizes, also, each characteristic pattern pixel, correspondence
The pixel set of some characteristic pattern respective window positions specified in preceding layer；Each down-sampled layer includes multiple same sizes
Characteristic pattern, also, down-sampled layer every characteristic pattern, corresponding to a characteristic pattern of preceding layer convolutional layer, the spy of down-sampled layer
Levy sample area of the image element corresponding to preceding layer individual features figure.A certain node layer and previous node layer and latter node layer it
Between by side be connected with each other.
After building and obtaining the above-mentioned convolutional neural networks template with particular network architecture, when needing to recognize a certain picture
When, it is necessary to be trained to above-mentioned convolutional neural networks template, training process is：Initialize the ginseng of convolutional neural networks template
Number is random value, including：The weighted value on side and the value of convolution kernel etc.；Then, training sample is input into convolutional neural networks mould
Plate, " stimulates " convolutional neural networks template repeatedly, constantly value of the weighted value on adjustment side and convolution kernel etc., until training
To the convolutional neural networks template that can recognize that the picture.In subsequent applications, it is only necessary to be analysed to picture or other samples are defeated
Enter in the convolutional neural networks template for training, you can reach the purpose of classification and Intelligent Recognition.
In order to from complex scene separate and identify each object, it is necessary to be traveled through to image using substantial amounts of template
Convolutional calculation, its is computationally intensive, and the calculating time is long, is difficult to realize real-time object identification.
The content of the invention
For in the prior art, separated from complex scene and identify each object, it is necessary to use substantial amounts of template pair
Image carries out traversal convolutional calculation, calculates the problem of overlong time, and the present invention proposes a kind of image-recognizing method, including：Connect
Receive image to be identified；Convolutional calculation is carried out to described image simultaneously using two templates, to obtain each area of described image
Domain category score corresponding with described two templates；Judge whether include what is recognized in image-region based on the category score
Object.
Further, the method is additionally included in using two templates while before carrying out convolutional calculation to described image, sentencing
Described two template sizes that break are identical, if described two template sizes are different, methods described terminates.
Further, described two templates are symmetrical template.
Further, described two templates are assembled for training by specific program and mass data and got.
Further, carrying out convolutional calculation to described image simultaneously using two templates includes：According to specific rule, make institute
State two templates and travel through whole image pixel by pixel.
Further, according to specific rule, described two templates is traveled through whole image pixel by pixel includes：
A) original position with described image takes the image being equal to template size with y-axis direction along the x-axis direction as starting point
Region；
B described image region) is carried out into convolutional calculation with described two templates respectively, obtain described image region respectively with
The corresponding category score of two templates；
C) starting point coordinate is made to add 1 along the x-axis direction, taken with y-axis direction along the x-axis direction with template size based on starting point coordinate etc.
Same image-region；
D) judge whether taken image-region exceeds the scope of image along the x-axis direction, if the image-region for being taken is along x
Direction of principal axis without departing from image scope, then return to step B), and repeat step B) to D), until the image-region for being taken is along x-axis
Direction then proceeds to step E beyond the scope of image)；
E the x values of starting point coordinate) are set to starting position coordinates value and y values is increased by 1, based on starting point coordinate along the x-axis direction
The image-region being equal to template size is taken with y-axis direction；
F) judge whether taken image-region exceeds the scope of image along the y-axis direction, if the image-region for being taken is along y
Direction of principal axis without departing from image scope, then return to step B), and repeat step B) to E), until the image-region for being taken is along y-axis
Scope of the direction beyond image.
Further, CPU will carry out the pixel matrix of the image-region of convolutional calculation from image reading,
And image-region and two convolutional calculations of template are completed by graphics processing unit under the control of the CPU.
Further, image-region includes the corresponding value of the value and image-region of template every with the convolutional calculation of template
It is multiplied, the value for obtaining is sued for peace again, category score finally giving and as the image-region.
Further, when the category score of a certain image-region is more than specific threshold, then judge to be wrapped in the image-region
Containing the object for being recognized；When the category score of a certain image-region is less than or equal to specific threshold, then the image-region is judged
In do not include recognized object.
Further, by counting training pictures after all positive sample target score descending sorts before 95% position
Put the specific threshold of the fraction as template.
According to the scheme that the present invention is provided, while carrying out convolution using two templates, the reading times of data are reduced, it is right
For the whole calculating of image, substantial amounts of data read time can be saved, significantly improve recognition speed.
Brief description of the drawings
For the above and other advantages and features of each embodiment that the present invention is furture elucidated, refer to the attached drawing is presented
The more specifically description of various embodiments of the present invention.It is appreciated that these accompanying drawings only describe exemplary embodiments of the invention, therefore
It is restriction on its scope to be not to be regarded as.In the accompanying drawings, in order to cheer and bright, identical or corresponding part will use identical or class
As mark represent.
Fig. 1 shows the schematic diagram of images to be recognized and template.
Fig. 2 shows to make template travel through the whole flow chart of image.
Fig. 3 shows the schematic diagram of identification people's template according to an embodiment of the invention.
Fig. 4 is shown according to an embodiment of the invention using two templates while carrying out the flow chart of convolution traversal to image
400。
Specific embodiment
In the following description, with reference to each embodiment, present invention is described.However, those skilled in the art will recognize
Knowing can replace and/or addition method, material or component in the case of neither one or multiple specific details or with other
Implement each embodiment together.It is not shown or do not describe known structure, material or operation in detail in order to avoid making this in other situations
The aspects of each embodiment of invention is obscure.Similarly, for purposes of explanation, specific quantity, material and configuration are elaborated, with
Comprehensive understanding to embodiments of the invention is just provided.However, the present invention can be implemented in the case of no specific detail.This
Outward, it should be understood that each embodiment shown in accompanying drawing is illustrative expression and is not drawn necessarily to scale.
In this manual, the reference to " one embodiment " or " embodiment " means to combine embodiment description
Special characteristic, structure or characteristic are included at least one embodiment of the invention.In the short of this specification middle appearance everywhere
Language " in one embodiment " is not necessarily all referring to same embodiment.
First, the related notion used during processing image using template is introduced：
Template refers to matrix square, and its mathematical sense is a kind of convolutional calculation.
Convolutional calculation：Be considered as the process of weighted sum, each pixel in the image-region for using respectively with volume
Product core (that is, weight matrix) each element correspondence be multiplied, all sum of products as regional center pixel new value.
Convolution kernel：During convolution using to power represent that the matrix is identical with the image area size for using with a matrix,
Its row, column is all odd number, is a weight matrix.
Convolutional calculation example：
The convolutional calculation of the pixel region R and convolution kernel G of 3*3：
Vacation lets R be the pixel region of 3*3, and convolution kernel is G：
Convolution and=R1G1+R2G2+R3G3+R4G4+R5G5+R6G6+R7G7+R8G8+R9G9
The present invention proposes to use the category score of formwork calculation image, detects whether to be the thing for recognizing based on category score
Body.The detailed process of the category score of image is calculated with reference to Fig. 1 and Fig. 2 introductions.
Fig. 1 shows the schematic diagram of images to be recognized and template.As shown in figure 1, rectangular box 110 is image, by some pictures
Vegetarian refreshments is constituted, and the image has specific width and height.Shaded boxes 120 are template.By template 120 and the figure of institute overlay area
The value multiplication corresponding with institute overlay area image as carrying out the value of convolutional calculation, i.e. template every, the value for obtaining is sued for peace again, most
Category score obtaining eventually and as the image-region.Category score represents the response intensity in the region and template, and response is strong
Degree is bigger, and score is higher, otherwise score is smaller.
, it is necessary to make template travel through whole image during being identified to image.Fig. 2 shows to make template travel through whole
The flow chart of image.
First, in step 210, a pending image is received.Next, making template be opened from the original position of the image
Beginning carries out convolutional calculation traversal.For example, the coordinate for setting the original position of image is (0,0), it is starting point with original position (0,0),
Take with y-axis direction the image-region being equal to template size along the x-axis direction from the off.
In step 220, the region and template are carried out into convolution, i.e. by the pixel value in the region respectively with two templates
Respective value is multiplied and sues for peace again, obtains category score of the image-region for the template.
In step 230, starting point coordinate is added 1 along the x-axis direction, with position (1,0) for starting point along the x-axis direction with y-axis direction
Take the image-region being equal to template size.
In step 240, judge whether taken image-region exceeds the scope of image along the x-axis direction.If the figure for being taken
Scope as region without departing from image, then return to step 220, convolution is carried out by acquired image-region and template, is somebody's turn to do
Category score of the image-region for the template.
Next, circulated between step 220 to 240, until judging scope of the taken image-region beyond image, then
Step 250 is turned to, the x values of starting point coordinate are set to starting position coordinates value, y values increase by 1.The edge since new start position
X-axis direction and y-axis direction take the image-region being equal to template size.In step 260, judge taken image-region along y-axis
Whether direction exceeds the scope of image.If the image-region for being taken is without departing from the scope of image, return to step 220, by institute
The image-region of acquirement carries out convolution with template, obtains category score of the image-region for the template.Next, in step
Circulated between 220 to 250, until judging taken image-region along the y-axis direction beyond the scope of image, then whole image convolution
Complete.
In an embodiment of the present invention, the template for being used is assembled for training by specific program and mass data and got.Below
The training process of template is briefly introduced with reference to specific example.
In an embodiment of the present invention, using variable to train comprising the picture largely with rectangle frame label target example
The problem reduction of shape component model is two classification problems, and classification is then trained to target using SVM.Fig. 3 is shown according to this
The schematic diagram of identification people's template of one embodiment of invention.One detection template contains 1 root wave filter and 8 part filtering
Device, template threshold value 95% position before after all positive sample target score descending sorts in counting training pictures is divided
Count as the threshold value of detection template.
Template training
Root filter initialization：For each target classification, the statistics of target rectangle frame size is concentrated according to training data
Value, automatically selects the size of root wave filter.
Root filter update：The initial root wave filter that given previous step training is obtained, for each rectangle in training set
Frame, under conditions of root wave filter and rectangle frame are significantly overlapped (overlap more than 50%), find wave filter highest scoring one
Position.
Part filter initialization：Using a kind of simple heuristic according to the root filter initialization trained above
Eight part wave filters.Selection area a first so that 8a is equal to the 80% of root filter area.Filtered from root using exhaustive method
It is the rectangular area with the positive weights quadratic sum of all cell units in the maximum region of a that area is selected in ripple device, by this
All weights in region reset and then proceed to selection, until selecting eight rectangular areas.
Model modification：By build new training data triple (<x1,z1,y1>,...,<xn,zn,yn>) wherein xi is
Sample, yi is sample class, zi be last iterative learning to model in be best suitable for the label of xi and carry out more new model.For instruction
Practice each positive sample rectangle frame in data set, in the case where at least 50% overlap is ensured with existing detector in be possible to position
Put and detected with yardstick, the position with top score is selected wherein as the corresponding positive sample of this rectangle frame, be put into sample
In this buffering area.
Detection score position high is selected in the picture not comprising target object as negative sample, constantly to Sample Buffer
Negative sample is added in area to be limited until file is maximum.New model is trained on positive negative sample in the buffer, all samples are all
There are component locations to mark.
In order to from complex scene separate and identify each object, it is necessary to be traveled through to image using substantial amounts of template
Convolutional calculation.In order to improve the speed of classification identification, and consider the similitude of template, during template training, train sometimes
The template for going out occurs in pairs, i.e. symmetrical template, therefore when convolutional calculation is carried out every time, while being carried out using two templates
Convolution, reduces the reading times of data, for the whole calculating of image, can save substantial amounts of data read time, shows
Write and improve recognition speed.
Fig. 4 is shown according to an embodiment of the invention using two templates while carrying out the flow chart of convolution traversal to image
400。
In step 401, image to be identified is received.
In step 402, judge to use two template sizes whether identical.If two sizes of template are different, flow
Journey terminates.
If the size of two templates is identical, in step 403, image is entered simultaneously using two templates from original position
Row convolutional calculation.Two templates travel through whole image simultaneously.For example, the coordinate for setting the original position of image is (0,0), with starting
Position (0,0) is starting point, takes with y-axis direction the image-region being equal to template size along the x-axis direction from the off.By the area
Domain carries out convolutional calculation with two templates respectively, i.e. the respective value by the pixel value in the region respectively with two templates is multiplied again
Summation, obtains image-region category score corresponding with two templates respectively.
In actual applications, the picture in the region of convolutional calculation can be carried out from image reading by CPU (CPU)
Plain value matrix, R (0,0), being utilized respectively two templates carries out convolutional calculation S1 (0,0)=R (0,0) ⊙ G1 and S2 (0,0)=R
(0,0)⊙G2.Due to the superperformance that GPU (GPU) shows in terms of matrix operation, therefore can be in the control of CPU
Convolutional calculation is completed by GPU under system.Then, starting point coordinate is added 1 along the x-axis direction, with position (1,0) for starting point along the x-axis direction
The image-region being equal to template size is taken with y-axis direction, and judges whether taken image-region exceeds image along the x-axis direction
Scope, if the region is carried out convolution by the image-region for being taken with two templates respectively without departing from the scope of image, obtain
To S1 (1,0)=R (1,0) ⊙ G1 and S2 (1,0)=R (1,0) ⊙ G2, then individual element is incremented by and judges taken image district
Whether domain exceeds the scope of image along the x-axis direction, if the image-region for being taken is without departing from the scope of image, carries out convolution meter
Calculate.When the image-region for being taken exceeds the scope of image along the x-axis direction, x values are set into 0, y values increases by 1, from new starting point position
Put (0, y) start to be taken with y-axis direction along the x-axis direction the image-region being equal to template size, and judge taken image-region
Whether exceed the scope of image along the y-axis direction, if the image-region for being taken is without departing from the scope of image, by region difference
Convolution is carried out with two templates, the category score in the region is obtained, next, individual element is incremental along the x-axis direction carries out convolution
Calculate.When the image-region for being taken exceeds the scope of image along the x-axis direction, x values are set into 0, y values increases by 1.Circulate successively,
Until the image-region for being taken is along the y-axis direction beyond the scope of image, whole image convolution is completed.
In step 404, judge whether include what is recognized in the image-region using the category score of each image-region
Object.When the category score of a certain image-region is more than specific threshold, then judge to include recognized thing in the image-region
Body.When the category score of a certain image-region is less than or equal to specific threshold, then judge in the image-region not comprising being known
Other object.
Described using two templates of same size above in association with Fig. 4 while travel through whole image, two same sizes
Template can be symmetrical template.And it is it should be appreciated by one skilled in art that same using the template of two same sizes
When travel through the mode of whole image and be not limited to the mode introduced in step 403.For example, template can be traveled through along the y-axis direction first
Image, then increasing a pixel along the x-axis direction, traversing graph picture along the y-axis direction again, until whole image convolution is completed.
In another embodiment of the present invention, the starting point for convolution traversal being carried out to image is not the original position of image
Or the origin of coordinates, but from a certain pixel (x in the central area of imagei, yi) start to take the image being equal to template size
Region, then carries out convolutional calculation by the region and template, obtains category score of the image-region for the template.Connect down
Come, translated to the outer peripheral areas of image according to specific rule individual element, so as to travel through whole image.For example, can first in picture
Vegetarian refreshments (xi, yi) upper individual element of being expert at is incremented by and/or successively decreases and obtain the image-region that is equal to template size and carry out convolution
Calculate, line number is incremented by and/or a pixel of successively decreasing, individual element is incremented by and/or successively decreases acquisition and template on the row
The equivalent image-region of size simultaneously carries out convolutional calculation, until whole image convolution is completed.
Although described above is various embodiments of the present invention, however, it is to be understood that they are intended only as example to present
, and without limitation.For those skilled in the relevant art it is readily apparent that various combinations, modification can be made to it
With change without departing from the spirit and scope of the present invention.Therefore, the width and scope of the invention disclosed herein should not be gone up
State disclosed exemplary embodiment to be limited, and should be defined according only to appended claims and its equivalent.
Claims (10)
1. a kind of image-recognizing method, including：
Receive image to be identified；
Convolutional calculation is carried out to described image simultaneously using two templates, so as to obtain each region of described image with it is described two
The corresponding category score of template；
Judge whether include recognized object in image-region based on the category score.
2. method as claimed in claim 1, it is characterised in that methods described is additionally included in using two templates simultaneously to the figure
Before as carrying out convolutional calculation, judge that described two template sizes are identical, if described two template sizes are different, the side
Method terminates.
3. method as claimed in claim 1, it is characterised in that described two templates are symmetrical template.
4. method as claimed in claim 1, it is characterised in that described two templates pass through specific program and mass data collection training
Obtain.
5. method as claimed in claim 1, it is characterised in that convolutional calculation bag is carried out to described image simultaneously using two templates
Include：According to specific rule, described two templates are made to travel through whole image pixel by pixel.
6. method as claimed in claim 5, it is characterised in that according to specific rule, makes described two templates pixel by pixel time
Going through whole image includes：
A) original position with described image takes the image district being equal to template size with y-axis direction along the x-axis direction as starting point
Domain；
B described image region) is carried out into convolutional calculation with described two templates respectively, obtain described image region respectively with two
The corresponding category score of template；
C) starting point coordinate is added 1 along the x-axis direction, take what is be equal to template size with y-axis direction along the x-axis direction based on starting point coordinate
Image-region；
D) judge whether taken image-region exceeds the scope of image along the x-axis direction, if the image-region for being taken is along x-axis side
To the scope without departing from image, then return to step B), and repeat step B) to D), until the image-region that is taken along the x-axis direction
Beyond the scope of image, then step E is proceeded to)；
E the x values of starting point coordinate) are set to starting position coordinates value and y values is increased by 1, based on starting point coordinate along the x-axis direction and y
Direction of principal axis takes the image-region being equal to template size；
F) judge whether taken image-region exceeds the scope of image along the y-axis direction, if the image-region for being taken is along y-axis side
To the scope without departing from image, then return to step B), and repeat step B) to E), until the image-region that is taken along the y-axis direction
Beyond the scope of image.
7. method as claimed in claim 6, it is characterised in that CPU will carry out the figure of convolutional calculation from image reading
As region pixel matrix, and under the control of the CPU by graphics processing unit complete image-region with
Two convolutional calculations of template.
8. method as claimed in claims 6 or 7, it is characterised in that image-region includes the every point of template with the convolutional calculation of template
The value value corresponding with image-region be multiplied, the value for obtaining is sued for peace again, classification finally giving and as the image-region
Score.
9. method as claimed in claim 1, it is characterised in that when the category score of a certain image-region is more than specific threshold,
Then judge to include recognized object in the image-region；When the category score of a certain image-region is less than or equal to specific threshold
When, then judge not including recognized object in the image-region.
10. method as claimed in claim 9, it is characterised in that by counting all positive sample target scores in training pictures
After descending sort preceding 95% position fraction as template the specific threshold.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710056968.XA CN106845550B (en) | 2017-01-22 | 2017-01-22 | Image identification method based on multiple templates |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710056968.XA CN106845550B (en) | 2017-01-22 | 2017-01-22 | Image identification method based on multiple templates |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106845550A true CN106845550A (en) | 2017-06-13 |
| CN106845550B CN106845550B (en) | 2020-03-17 |
Family
ID=59122959
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710056968.XA Active CN106845550B (en) | 2017-01-22 | 2017-01-22 | Image identification method based on multiple templates |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106845550B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109308476A (en) * | 2018-09-06 | 2019-02-05 | 邬国锐 | Billing information processing method, system and computer readable storage medium |
| CN113361553A (en) * | 2020-03-06 | 2021-09-07 | 株式会社理光 | Image processing method, image processing apparatus, storage medium, and system |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102521826A (en) * | 2011-11-22 | 2012-06-27 | 苏州科雷芯电子科技有限公司 | Image registration device and method |
| CN104504381A (en) * | 2015-01-09 | 2015-04-08 | 博康智能网络科技股份有限公司 | Non-rigid target detection method and system thereof |
| KR20150120805A (en) * | 2014-04-18 | 2015-10-28 | 한양대학교 산학협력단 | Method and system for detecting human in range image |
| CN105069428A (en) * | 2015-07-29 | 2015-11-18 | 天津市协力自动化工程有限公司 | Multi-template iris identification method based on similarity principle and multi-template iris identification device based on similarity principle |
| CN105160330A (en) * | 2015-09-17 | 2015-12-16 | 中国地质大学（武汉） | Vehicle logo recognition method and vehicle logo recognition system |
| CN105260740A (en) * | 2015-09-23 | 2016-01-20 | 广州视源电子科技股份有限公司 | Element recognition method and apparatus |
| CN105320935A (en) * | 2015-07-29 | 2016-02-10 | 江苏邦融微电子有限公司 | Multiple-template fingerprint identification method |
-
2017
- 2017-01-22 CN CN201710056968.XA patent/CN106845550B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102521826A (en) * | 2011-11-22 | 2012-06-27 | 苏州科雷芯电子科技有限公司 | Image registration device and method |
| KR20150120805A (en) * | 2014-04-18 | 2015-10-28 | 한양대학교 산학협력단 | Method and system for detecting human in range image |
| CN104504381A (en) * | 2015-01-09 | 2015-04-08 | 博康智能网络科技股份有限公司 | Non-rigid target detection method and system thereof |
| CN105069428A (en) * | 2015-07-29 | 2015-11-18 | 天津市协力自动化工程有限公司 | Multi-template iris identification method based on similarity principle and multi-template iris identification device based on similarity principle |
| CN105320935A (en) * | 2015-07-29 | 2016-02-10 | 江苏邦融微电子有限公司 | Multiple-template fingerprint identification method |
| CN105160330A (en) * | 2015-09-17 | 2015-12-16 | 中国地质大学（武汉） | Vehicle logo recognition method and vehicle logo recognition system |
| CN105260740A (en) * | 2015-09-23 | 2016-01-20 | 广州视源电子科技股份有限公司 | Element recognition method and apparatus |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109308476A (en) * | 2018-09-06 | 2019-02-05 | 邬国锐 | Billing information processing method, system and computer readable storage medium |
| CN113361553A (en) * | 2020-03-06 | 2021-09-07 | 株式会社理光 | Image processing method, image processing apparatus, storage medium, and system |
| CN113361553B (en) * | 2020-03-06 | 2024-02-02 | 株式会社理光 | Image processing method, image processing apparatus, storage medium, and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106845550B (en) | 2020-03-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6159489B2 (en) | Face authentication method and system | |
| CN105447529B (en) | Method and system for detecting clothes and identifying attribute value thereof | |
| CN104809187B (en) | A kind of indoor scene semanteme marking method based on RGB D data | |
| CN106682697A (en) | End-to-end object detection method based on convolutional neural network | |
| CN107408211A (en) | Method for distinguishing is known again for object | |
| CN102682287B (en) | Pedestrian detection method based on saliency information | |
| CN107169421A (en) | A kind of car steering scene objects detection method based on depth convolutional neural networks | |
| CN107133616A (en) | A kind of non-division character locating and recognition methods based on deep learning | |
| CN108921058A (en) | Fish identification method, medium, terminal device and device based on deep learning | |
| CN106709568A (en) | RGB-D image object detection and semantic segmentation method based on deep convolution network | |
| CN106874894A (en) | A kind of human body target detection method based on the full convolutional neural networks in region | |
| US20220215227A1 (en) | Neural Architecture Search Method, Image Processing Method And Apparatus, And Storage Medium | |
| CN109255375A (en) | Panoramic picture method for checking object based on deep learning | |
| CN106778835A (en) | The airport target by using remote sensing image recognition methods of fusion scene information and depth characteristic | |
| CN107871102A (en) | A kind of method for detecting human face and device | |
| WO2021218786A1 (en) | Data processing system, object detection method and apparatus thereof | |
| CN105608456A (en) | Multi-directional text detection method based on full convolution network | |
| CN109446922B (en) | Real-time robust face detection method | |
| CN107808129A (en) | A kind of facial multi-characteristic points localization method based on single convolutional neural networks | |
| CN109558902A (en) | A kind of fast target detection method | |
| CN105469041A (en) | Facial point detection system based on multi-task regularization and layer-by-layer supervision neural networ | |
| CN107944437A (en) | A kind of Face detection method based on neutral net and integral image | |
| CN105005798B (en) | One kind is based on the similar matched target identification method of structures statistics in part | |
| CN112364931A (en) | Low-sample target detection method based on meta-feature and weight adjustment and network model | |
| CN107767416A (en) | The recognition methods of pedestrian's direction in a kind of low-resolution image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |