CN107729910A - A kind of method, apparatus and system of figure identification - Google Patents
A kind of method, apparatus and system of figure identification Download PDFInfo
- Publication number
- CN107729910A CN107729910A CN201610659370.5A CN201610659370A CN107729910A CN 107729910 A CN107729910 A CN 107729910A CN 201610659370 A CN201610659370 A CN 201610659370A CN 107729910 A CN107729910 A CN 107729910A
- Authority
- CN
- China
- Prior art keywords
- picture
- detected
- confidence level
- shape sample
- result
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
Abstract
The application is related to computer realm, the more particularly to a kind of method, apparatus and system of figure identification.To improve the accuracy of figure identification.This method is:Detection means needs to carry out figure identification to picture to be detected, using default each shape sample, the figure included in picture to be detected is positioned respectively, obtain corresponding positioning result and confidence level, finally, confidence level highest positioning result is based on again, obtains corresponding figure recognition result;So, due to there is provided the shape sample to become more meticulous, therefore, the identification precision of the same class figure of different objects shape now can be effectively improved.
Description
Technical field
The application is related to computer realm, the more particularly to a kind of method, apparatus and system of figure identification.
Background technology
With the development of computer video, in order to cater to the use demand of user, object detection technology is in many nets
It is obtained for and makes full use of in standing, an important content being increasingly becoming in website service.For example, unmanned device passes through
Pedestrian, vehicle and road sign in object detection technology identification video;In another example Website server by object detection technology from
User's object interested is identified in the picture of family input;In another example intelligent monitoring and controlling device is regarded by the identification of object detection technology
Pedestrian, progress Face datection etc. in frequency.
In a broad aspect, object detection refers to the thing for being automatically positioned the single or multiple classification specified in picture or video
Body.Object detection mainly includes two tasks, identifies the seat of the classification of object and positioning object included in picture or video
Cursor position.
By taking the object detection technology that Website server uses as an example.At present, Website server would generally divide object
Class, then, object detection is carried out again now in each class.Specifically, it is pre- to be directed to each classification for Website server
A grader and coordinate is first trained to return device.Grader is used to carry out target object classification identification, and coordinate returns device then
For now, coordinate position of the object in picture being identified, finally again based on the coordinate position identified in respective class
Body form is determined, object detection is carried out further according to the body form.
However, under some application scenarios (e.g., user searches for the pictures of the commodity for wishing to buy in website), even returning
Belonging to the object of same class now, difference in shape is also bigger, and e.g., at " trousers ", this class has trousers, shorts etc. now,
And for example, at " skirt ", this class has one-piece dress, longuette, skirt etc. now.
For this classification, in the object detecting method used at present, device is returned to ownership usually using single coordinate
Coordinate position identification is carried out in same class purpose different objects.But the shape facility used in a coordinate returns device is logical
The sample data with shape facility general character for crossing magnanimity trains what is obtained, and therefore, a coordinate returns device often for a certain
The object of shape can carry out accurately identifying for coordinate position.However, due to belonging to same class purpose object in shape
Can have larger difference, then, using single coordinate return device carry out coordinate position identification, can cause output coordinate position and
Deviation is larger between actual coordinate position, so as to further influence the accuracy of figure identification.
The content of the invention
The embodiment of the present application provides a kind of method, apparatus and system of figure identification, to improve the accurate of figure identification
Property.
The concrete technical scheme that the embodiment of the present application provides is as follows:
A kind of pattern recognition system, including:
Client, for obtaining picture to be detected;
Server, for receiving the picture to be detected of client transmissions, using default each shape sample, respectively to institute
State the figure included in picture to be detected to be positioned, obtain corresponding positioning result and confidence level, wherein, a shape sample
The confidence level of positioning result corresponding to this, characterize the similarity of one shape sample and the figure;Based on the confidence
Degree, obtains corresponding figure recognition result.
Optionally, when obtaining picture to be detected, the client is used for:
Original image is obtained, and using the original image as picture to be detected;Or
Original image is obtained, and designated area is intercepted in the original image according to setting means, and by the finger
Region is determined as picture to be detected.
Optionally, using default each shape sample, the figure included in the picture to be detected is determined respectively
During position, the server is used for:
Using default shape sample, the figure included in the picture to be detected is positioned respectively;Or
Feature extraction is carried out to the picture to be detected, the class of the picture ownership to be detected is determined according to extraction result
Mesh, then based on the corresponding default each shape sample of classification, the figure included in the picture to be detected is carried out respectively
Positioning.
Optionally, using a default shape sample, the figure included in the picture to be detected is positioned, obtained
When obtaining corresponding positioning result, the server is used for:
Using one shape sample, Graphic Pattern Matching is carried out in the picture to be detected;
It is determined that during in the presence of the figure that the match is successful, according to one shape sample, side is drawn on the periphery of the figure
Frame;
The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
Optionally, when obtaining the confidence level of the positioning result, the server is used for:
The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, it is right
The similarity carries out Error processing, and using result as the confidence level.
Optionally, based on the confidence level, when obtaining corresponding figure recognition result, the server is used for:
Using shape sample corresponding to confidence level highest positioning result as figure recognition result.
Optionally, server is further used for:
Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and is identified knot with the figure
Object as fruit.
A kind of pattern recognition method, including:
Obtain picture to be detected;
Using default each shape sample, the figure included in the picture to be detected is positioned respectively, obtained
Corresponding positioning result and confidence level, wherein, the confidence level of positioning result corresponding to a shape sample, characterize one
The similarity of shape sample and the figure;
Based on the confidence level, corresponding figure recognition result is obtained.
Optionally, the acquisition picture to be detected, including:
Original image is obtained, and using the original image as picture to be detected;Or
Original image is obtained, and designated area is intercepted in the original image according to setting means, and by the finger
Region is determined as picture to be detected.
Optionally, it is described to use default each shape sample, the figure included in the picture to be detected is entered respectively
Row positioning, including:
Using default shape sample, the figure included in the picture to be detected is positioned respectively;Or
Feature extraction is carried out to the picture to be detected, the class of the picture ownership to be detected is determined according to extraction result
Mesh, then based on the corresponding default each shape sample of classification, the figure included in the picture to be detected is carried out respectively
Positioning.
Optionally, using a default shape sample, the figure included in the picture to be detected is positioned, obtained
Corresponding positioning result is obtained, including:
Using one shape sample, Graphic Pattern Matching is carried out in the picture to be detected;
It is determined that during in the presence of the figure that the match is successful, according to one shape sample, side is drawn on the periphery of the figure
Frame;
The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
Optionally, the confidence level of the positioning result is obtained, including:
The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, it is right
The similarity carries out Error processing, and using result as the confidence level.
Optionally, based on the confidence level, corresponding figure recognition result is obtained, including:
Using shape sample corresponding to confidence level highest positioning result as figure recognition result.
Optionally, further comprise:
Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and is identified knot with the figure
Object as fruit.
A kind of pattern recognition method, including:
Obtain picture to be detected;
The shape sample that device includes is returned using default each coordinate, respectively the figure to being included in the picture to be detected
Shape is positioned, and obtains corresponding positioning result and confidence level, wherein, it is corresponding that a coordinate returns the shape sample that device includes
Positioning result confidence level, characterize the similarity that one coordinate returns shape sample that device includes and the figure;
Based on the confidence level, corresponding figure recognition result is obtained.
Optionally, the acquisition picture to be detected, including:
Original image is obtained, and using the original image as picture to be detected;Or
Original image is obtained, and designated area is intercepted in the original image according to setting means, and by the finger
Region is determined as picture to be detected.
Optionally, the shape sample included using default each coordinate recurrence device, respectively to the mapping to be checked
The figure included in piece is positioned, including:
The shape sample that devices include is returned using default whole coordinates, respectively the figure to being included in the picture to be detected
Shape is positioned;Or
Feature extraction is carried out to the picture to be detected, the class of the picture ownership to be detected is determined according to extraction result
Mesh, then the shape sample included based on the corresponding default each coordinate recurrence device of classification, respectively to the picture to be detected
In the figure that includes positioned.
Optionally, the shape sample included using default coordinate recurrence device, to being included in the picture to be detected
Figure positioned, obtain corresponding positioning result, including:
The shape sample included using one coordinate recurrence device, Graphic Pattern Matching is carried out in the picture to be detected;
It is determined that during in the presence of the figure that the match is successful, the shape sample that device includes is returned according to one coordinate, described
Draw frame in the periphery of figure;
The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
Optionally, the confidence level of the positioning result is obtained, including:
The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, it is right
The similarity carries out Error processing, and using result as the confidence level.
Optionally, based on the confidence level, corresponding figure recognition result is obtained, including:
The shape sample that coordinate recurrence device corresponding to confidence level highest positioning result is included, identifies as figure and ties
Fruit.
Optionally, further comprise:
Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and is identified knot with the figure
Object as fruit.
A kind of pattern recognition device, including:
Acquiring unit, for obtaining picture to be detected;
Positioning unit, for using default each shape sample, the respectively figure to being included in the picture to be detected
Positioned, obtain corresponding positioning result and confidence level, wherein, the confidence of positioning result corresponding to a shape sample
Degree, characterize the similarity of one shape sample and the figure;
Recognition unit, for based on the confidence level, obtaining corresponding figure recognition result.
Optionally, when obtaining picture to be detected, the acquiring unit is used for:
Original image is obtained, and using the original image as picture to be detected;Or
Original image is obtained, and designated area is intercepted in the original image according to setting means, and by the finger
Region is determined as picture to be detected.
Optionally, using default each shape sample, the figure included in the picture to be detected is determined respectively
During position, the acquiring unit is used for:
Using default shape sample, the figure included in the picture to be detected is positioned respectively;Or
Feature extraction is carried out to the picture to be detected, the class of the picture ownership to be detected is determined according to extraction result
Mesh, then based on the corresponding default each shape sample of classification, the figure included in the picture to be detected is carried out respectively
Positioning.
Optionally, using a default shape sample, the figure included in the picture to be detected is positioned, obtained
When obtaining corresponding positioning result, the positioning unit is used for:
Using one shape sample, Graphic Pattern Matching is carried out in the picture to be detected;
It is determined that during in the presence of the figure that the match is successful, according to one shape sample, side is drawn on the periphery of the figure
Frame;
The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
Optionally, when obtaining the confidence level of the positioning result, the positioning unit is used for:
The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, it is right
The similarity carries out Error processing, and using result as the confidence level.
Optionally, based on the confidence level, when obtaining corresponding figure recognition result, the recognition unit is used for:
Using shape sample corresponding to confidence level highest positioning result as figure recognition result.
Optionally, the recognition unit is further used for:
Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and is identified knot with the figure
Object as fruit.
A kind of pattern recognition method, including:
Acquiring unit, for obtaining picture to be detected;
Positioning unit, for the shape sample included using default each coordinate recurrence device, respectively to described to be detected
The figure included in picture is positioned, and obtains corresponding positioning result and confidence level, wherein, a coordinate returns device and included
Shape sample corresponding to positioning result confidence level, characterize one coordinate and return shape sample and the figure that device includes
The similarity of shape;
Recognition unit, for based on the confidence level, obtaining corresponding figure recognition result.
Optionally, when obtaining picture to be detected, the acquiring unit is used for:
Original image is obtained, and using the original image as picture to be detected;Or
Original image is obtained, and designated area is intercepted in the original image according to setting means, and by the finger
Region is determined as picture to be detected.
Optionally, the shape sample included using default each coordinate recurrence device, respectively in the picture to be detected
Comprising figure positioned when, the positioning unit is used for:
The shape sample that devices include is returned using default whole coordinates, respectively the figure to being included in the picture to be detected
Shape is positioned;Or
Feature extraction is carried out to the picture to be detected, the class of the picture ownership to be detected is determined according to extraction result
Mesh, then the shape sample included based on the corresponding default each coordinate recurrence device of classification, respectively to the picture to be detected
In the figure that includes positioned.
Optionally, the shape sample included using default coordinate recurrence device, to being included in the picture to be detected
Figure positioned, when obtaining corresponding positioning result, the positioning unit is used for:
The shape sample included using one coordinate recurrence device, Graphic Pattern Matching is carried out in the picture to be detected;
It is determined that during in the presence of the figure that the match is successful, the shape sample that device includes is returned according to one coordinate, described
Draw frame in the periphery of figure;
The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
Optionally, when obtaining the confidence level of the positioning result, the positioning unit is used for:
The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, it is right
The similarity carries out Error processing, and using result as the confidence level.
Optionally, based on the confidence level, when obtaining corresponding figure recognition result, the recognition unit is used for:
The shape sample that coordinate recurrence device corresponding to confidence level highest positioning result is included, identifies as figure and ties
Fruit.
Optionally, the recognition unit is used for:
Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and is identified knot with the figure
Object as fruit.
The application has the beneficial effect that:
In summary, in the embodiment of the present application, detection means needs to carry out figure identification to picture to be detected, using default
Each shape sample, the figure included in picture to be detected is positioned respectively, corresponding positioning result is obtained and puts
Reliability, finally, then based on confidence level highest positioning result, obtain corresponding figure recognition result;So, because there is provided essence
The shape sample of refinement, therefore, the identification precision of the same class figure of different objects shape now can be effectively improved.
Brief description of the drawings
Fig. 1 is detection means operation logic schematic diagram in the embodiment of the present application;
Fig. 2A is pattern recognition system configuration diagram in the embodiment of the present application;
Fig. 2 B are that detection means carries out figure identification process figure in the embodiment of the present application;
Fig. 3 is positioning result schematic diagram in the embodiment of the present application;
Fig. 4 is detection means illustrative view of functional configuration in the embodiment of the present application.
Embodiment
In order to improve the accuracy of figure identification, in the embodiment of the present application, for each classification, it is respectively divided out multiple
Subclass, wherein, a subclass is characterized in a kind of body form of respective class now, then, one is set for each subclass
Coordinate returns device, and so, each coordinate returns device and carried out targetedly just for the body form that corresponding subclass characterizes
Figure positions, and therefore, can greatly promote the accuracy of figure identification.
The application preferred embodiment is described in detail below in conjunction with the accompanying drawings.
As shown in fig.1, in the present embodiment, multiple objects grader, an object classification are provided with detection means
The corresponding classification of device, such as:Jacket, skirt, footwear, bag, pet etc..Wherein, each object classification device is by magnanimity
Sample data trains what is obtained, can be by carrying out feature extraction to picture to determine the classification of picture ownership.
And some coordinates are respectively arranged with each object classification device and return device, a coordinate recurrence device corresponds to corresponding
A kind of body form (a kind of alternatively referred to as shape sample) of class now;Wherein, each coordinate returns device and is all based on magnanimity sample
What the cluster training of notebook data obtained.
For example, it is assumed that the classification that object classification device 1 characterizes is " jacket ", and three are respectively arranged with object classification device 1
Kind coordinate returns device, wherein, it is " shortage of money jacket, Aspect Ratio 1 that coordinate, which returns the body form that device 1 characterizes,:1 ", and coordinate returns
The body form for returning device 2 to characterize is " common jacket, Aspect Ratio 4:3”;And it is " long that coordinate, which returns the body form that device 3 characterizes,
Money jacket, Aspect Ratio 7:4”.
In another example, it is assumed that the classification that object classification device 2 characterizes is " bag ", and is respectively arranged with three under object classification device 2
Kind coordinate returns device, wherein, it is " circular bag " that coordinate, which returns the body form that device 1 characterizes, and coordinate returns the object shape that device 2 characterizes
Shape is " inverted trapezoidal bag ", and it is that " quadrangle bag (can be square or rectangular that coordinate, which returns the body form that device 3 characterizes,
Shape) ".
Certainly, the above-mentioned shape that jacket is described using Aspect Ratio, and using the wide approximate figure description bag of chartered steamer
Shape, it is citing, is not limited thereto, will not be repeated here in practical application.
Certainly, if the quantity of sample data is enough, and discrimination is fine enough, then can be by clustering algorithm (e.g.,
K-means more body form samples) are obtained, return device so as to train more targetedly coordinates, herein also no longer
Repeat.
Refering to shown in Fig. 2A, in the embodiment of the present invention, client and server is included in pattern recognition system,
Client 20, for obtaining picture to be detected;
Server 21, it is right respectively using default each shape sample for receiving the picture to be detected of client transmissions
The figure included in picture to be detected is positioned, and obtains corresponding positioning result and confidence level, wherein, a shape sample
The confidence level of corresponding positioning result, characterize the similarity of a shape sample and above-mentioned figure;Based on above-mentioned confidence level, obtain
Obtain corresponding figure recognition result.
Based on said system framework, refering to shown in Fig. 2 B, in the embodiment of the present application, the detailed process of object detection is as follows:
Step 200:Detection means obtains picture to be detected.
When detection means (e.g., server) obtains picture to be detected, the user of client upload can be received in client
The picture to be detected of upper input, it can also directly obtain the picture to be detected that user inputs on detection means, subsequent embodiment
In, illustrated by taking latter event as an example, wherein, during obtain picture to be detected performed by detection means one
System processing, is applied equally to client, subsequently will not be described in great detail.
In the embodiment of the present application, when user inputs the original image for needing to search on detection means, it can use a variety of
Method.
For example, user can replicate to the original image that search for of needs, then affix on detection means screen and be in
In existing detection block, then " search " button is clicked on, so, detection means can be to get original image.
In another example user, which browses some interface, sees the original image liked and it is desirable that when being scanned for it, Ke Yichang
By original image, then menu bar can be presented in detection means on screen, and user can select " search " button therein, this
Sample, detection means can also get original image.
Further, for detection means angle, after the original image of user's input is got, if original image
Size be not up to predetermined threshold value, then detection means can directly using user choose original image as picture to be detected.
And if the size of original image has reached predetermined threshold value, now illustrating that original image is very big, it is impossible to single treatment finishes, this
When, detection means can also use the original that setting means (e.g., sliding window mode or selective search mode) inputs in user
Designated area is intercepted in beginning picture, and using designated area as picture to be detected.Certainly, if in the designated area currently intercepted
Do not include special pattern, then can continue to intercept next piece of designated area in original image using above-mentioned setting means, this
Sample, can break the whole up into parts, and whether detect successively has specific figure in each designated area.
And in the present embodiment, only exemplified by including specific figure in the designated area currently intercepted, it will not be repeated here.
In practical application, detection means can be the terminal (e.g., smart mobile phone, tablet personal computer etc.) that user uses,
It can be the server of network side, will not be repeated here.
Step 201:Detection means carries out feature extraction to picture to be detected, determines that picture to be detected is returned based on extraction result
The classification of category.
Specifically, detection means can input picture to be detected respectively each object classification device, each object classification device is all
Corresponding characteristic element is provided with, by these characteristic elements, each object classification device can be in the starting stage to figure to be detected
Piece carries out classification division.
For example, the classification that object classification device 1 manages is " pet ", then feature extraction key element can include " eyes ", " ear
Piece ", " beard ", " claw " etc., can be therein a kind of or any combination.
In another example the classification that object classification device 2 manages is " bag ", then feature extraction key element can include " band ", " bag
Body ", " pure color ", " cartoon pattern " etc., can be therein a kind of or any combination.
In another example the classification that object classification device 3 manages is " footwear ", then feature extraction key element can include " two independent
Body ", " strip ", " shoestring " etc., can be therein a kind of or any combination.
Characteristic element used in each object classification device all obtains after the training of Massive Sample data, sample number
According to quantity it is more, the classification results of object classification device are also more accurate.
In the embodiment of the present application, picture to be detected is inputted each object classification device, each object by detection means respectively
Grader all can carry out feature extraction to picture to be detected, then be identified based on respective characteristic element, finally determine to be checked
The classification of mapping piece ownership.
Optionally, in the present embodiment, it is assumed that after the identification of each object classification device, determine the class of picture ownership to be detected
Mesh is " jacket ", as example in subsequent embodiment, be will not be described in great detail.
Step 202:Detection means is based on the corresponding above-mentioned default each shape sample of classification, respectively in picture to be detected
Comprising figure positioned.
In the embodiment of the present application, multiple objects grader is prefixed, and under each object classification device, further divide
Multiple subclasses, the corresponding coordinate of each subclass return device, a coordinate return device characterize respective class now one
The typical body form of kind, by being then based on the training acquisition of Massive Sample data, therefore, also known as a coordinate returns device and characterized
A character shape be a kind of shape sample.
On the other hand, in the embodiment of the present application, detection means first uses the class that object classification device belongs to picture to be detected
Mesh is determined, and is reused corresponding each coordinate under above-mentioned object classification device and is returned device to the figure that is included in picture to be detected
Positioned.
And in practical application, can be direct if object classification device does not identify the classification of picture ownership to be detected
The figure included using the coordinate recurrence device of whole to picture to be detected is positioned, or, if default coordinate returns device
Number is less, then, in order to reduce operation complexity, can also be identified without classification, and directly returned using whole coordinates
The figure for returning device to include picture to be detected positions, then, in the case of this is several, step 201 can not be performed.
And in the embodiment of the present application, only illustrated, will not be repeated here exemplified by performing step 201.
Step 203:Detection means obtains positioning result and confidence level corresponding to each shape sample.
In the embodiment of the present application, by taking any one shape sample as an example (hereinafter referred to as shape sample X), detection means uses
It is that picture to be detected is input into coordinate corresponding to shape sample X to return when shape sample X detects to picture to be detected
In device X, device X is returned by coordinate the figure in picture to be detected is matched, judge whether to match with shape sample X
Successful figure, it is determined that during in the presence of the figure that the match is successful, by shape sample X, it is (specific to draw frame on the periphery of above-mentioned figure
As shown in Figure 3), then, it is determined that the coordinate position of above-mentioned frame, then the positioning result using the coordinate position as above-mentioned figure.
Still by taking classification " jacket " as an example, it is assumed that corresponding " jacket " is provided with three kinds of coordinates and returns device, wherein, coordinate returns device
The 1 shape sample characterized is " shortage of money jacket, Aspect Ratio 1:1 ", and it is " on common that coordinate, which returns the shape sample that device 2 characterizes,
Clothing, Aspect Ratio 4:3”;And it is " long blouse, Aspect Ratio 7 that coordinate, which returns the shape sample that device 3 characterizes,:4”.
So, detection means can use " 1 successively:1”、“4:3 " and " 7:The rectangle of 4 " these three ratios (can also be it
He is more bonded the figure of jacket shape, and rectangle is only for example), search for whether to have in picture to be detected and meet above-mentioned various ratios
Figure exist, that is, judge whether that the match is successful.In the embodiment of the present application, optionally, the registration of figure and shape sample reaches
It is considered as that the match is successful to given threshold (e.g., 70%).
In the case of the first, if the figure for meeting a certain ratio, the seat of output pattern in picture to be detected be present
Cursor position is as positioning result.
For example, it is assumed that detection means uses " 7:4 " rectangle is detected the presence of in picture to be detected similar in a shape
Figure, then detection means can be in the periphery of this figure drafting one " 7:4 " rectangle covers the figure, and exports this rectangle
Top left co-ordinate and bottom right angular coordinate as positioning result.
In another example, it is assumed that detection means uses " 1:It is close that 1 " rectangle detects the presence of a shape in picture to be detected
Figure, then detection means can this figure periphery draw one " 1:1 " rectangle covers the figure, and exports this square
The center point coordinate of shape and the coordinate on one of summit are as positioning result.
Certainly, positioning result can also use the coordinate position of other forms record figure, above are only citing.
In the case of second, if the figure for meeting any one ratio is not present in picture to be detected, can directly it carry
Go out to position unsuccessful.
Further, detection means will also export the confidence level of positioning result while positioning result is exported, so-called
Confidence level, expression is the shape sample of positioning result sign and the similarity degree of the figure in picture to be detected.
Because in practical application, due to the influence of shooting angle, shooting distance etc. factor, in picture to be detected
Figure seldom can be identical with default shape sample.But fine distinction can not illustrate that above-mentioned figure is exactly not
Meet shape sample.For example, the ratio that a shortage of money jacket is presented in picture to be detected is 1:0.9, this does not simultaneously meet pin
To the default ratio of shortage of money jacket " 1:1 ", still, it can not illustrate just not including shortage of money jacket in picture to be detected, therefore, be
Avoid judging by accident, in the present embodiment, add confidence level this concept.
Still with above-mentioned " 1:Exemplified by 0.9 " shortage of money jacket, because detection means can use each default shape sample
Figure in picture to be detected is positioned, therefore, detection means can use " 1 successively:1 " rectangle, " 4:3 " rectangle,
“7:4 rectangle " positions in picture to be detected to shortage of money jacket, and respectively obtains positioning result 1, positioning result 2 and determine
Position result 3, then, detection means can calculate " 1 respectively:0.9 " and " 1:1 " similarity 1=90%, " 1:0.9 " and " 4:3 "
Similarity 2=83%, " 1:0.9 " and " 7:4 " similarity 3=63%.
Finally, detection means can be by similarity corresponding to each shape sample, directly as corresponding positioning result
Confidence level, or, can also to corresponding to each shape sample similarity carry out Error processing after (e.g., be multiplied by predetermined coefficient,
Substitute into preset formula etc.), the confidence level using each result as corresponding positioning result.
The specific calculation of confidence level can will not be repeated here depending on concrete application environment and exact requirements.
Step 204:Detection means is based on confidence level highest positioning result, obtains corresponding figure recognition result.
Optionally, detection means is known the shape sample corresponding to confidence level highest positioning result as final figure
Other result.
For example, in the example above, final figure recognition result is " shortage of money jacket, Aspect Ratio 1:1”.
Step 205:Detection means is based on figure recognition result, and object detection is carried out in setting range.
Specifically, detection means can be based on figure recognition result, carry out object detection within a preset range, filter out with
The similar object of figure recognition result.
For example, detection means can be based on " shortage of money jacket, Aspect Ratio 1:1 " carries out object inspection in electric quotient data storehouse
Survey, filter out all jackets for meeting the description information, be presented to user, for selection by the user.
In another example the figure recognition result that detection means obtains is " children's school bag, circular ", then it can be based on the description and believe
Breath carries out object detection in database, filters out all school bags for meeting the description information, is presented to user, so that user selects
Select.
Refering to shown in Fig. 2A, in the embodiment of the present application, figure identification comprises at least:
Client 20, for obtaining picture to be detected;
Server 21, it is right respectively using default each shape sample for receiving the picture to be detected of client transmissions
The figure included in the picture to be detected is positioned, and obtains corresponding positioning result and confidence level, wherein, a shape
The confidence level of positioning result corresponding to sample, characterize the similarity of one shape sample and the figure;Put based on described
Reliability, obtain corresponding figure recognition result.
Optionally, when obtaining picture to be detected, client 20 is used for:
Original image is obtained, and using the original image as picture to be detected;Or
Original image is obtained, and designated area is intercepted in the original image according to setting means, and by the finger
Region is determined as picture to be detected.
Using default each shape sample, when being positioned respectively to the figure included in the picture to be detected, clothes
Business device 21 is used for:
Using default shape sample, the figure included in the picture to be detected is positioned respectively;Or
Feature extraction is carried out to the picture to be detected, the class of the picture ownership to be detected is determined according to extraction result
Mesh, then based on the corresponding default each shape sample of classification, the figure included in the picture to be detected is carried out respectively
Positioning.
Optionally, using a default shape sample, the figure included in the picture to be detected is positioned, obtained
When obtaining corresponding positioning result, server 21 is used for:
Using one shape sample, Graphic Pattern Matching is carried out in the picture to be detected;
It is determined that during in the presence of the figure that the match is successful, according to one shape sample, side is drawn on the periphery of the figure
Frame;
The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
Optionally, when obtaining the confidence level of the positioning result, server 21 is used for:
The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, it is right
The similarity carries out Error processing, and using result as the confidence level.
Optionally, based on the confidence level, when obtaining corresponding figure recognition result, server 21 is used for:
Using shape sample corresponding to confidence level highest positioning result as figure recognition result.
Optionally, server 21 is further used for:
Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and is identified knot with the figure
Object as fruit.
As shown in fig.4, in the embodiment of the present application, it is single that detection means includes acquiring unit 40, positioning unit 41 and identification
Member 42, wherein,
Specifically, acting on for above-mentioned unit has nuance under different scenes:
Under first scene:
Acquiring unit 40, for obtaining picture to be detected;
Positioning unit 41, for using default each shape sample, the respectively figure to being included in the picture to be detected
Shape is positioned, and obtains corresponding positioning result and confidence level, wherein, the confidence of positioning result corresponding to a shape sample
Degree, characterize the similarity of one shape sample and the figure;
Recognition unit 42, for based on confidence level highest positioning result, obtaining corresponding figure recognition result.
Optionally, when obtaining picture to be detected, acquiring unit 40 is used for:
Original image is obtained, and using the original image as picture to be detected;Or
Original image is obtained, and designated area is intercepted in the original image according to setting means, and by the finger
Region is determined as picture to be detected.
Optionally, using default each shape sample, the figure included in the picture to be detected is determined respectively
During position, acquiring unit 40 is used for:
Using default shape sample, the figure included in the picture to be detected is positioned respectively;Or
Feature extraction is carried out to the picture to be detected, the class of the picture ownership to be detected is determined according to extraction result
Mesh, then based on the corresponding default each shape sample of classification, the figure included in the picture to be detected is carried out respectively
Positioning.
Optionally, using a default shape sample, the figure included in the picture to be detected is positioned, obtained
When obtaining corresponding positioning result, positioning unit 41 is used for:
Using one shape sample, Graphic Pattern Matching is carried out in the picture to be detected;
It is determined that during in the presence of the figure that the match is successful, according to one shape sample, side is drawn on the periphery of the figure
Frame;
The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
Optionally, when obtaining the confidence level of the positioning result, positioning unit 41 is used for:
The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, it is right
The similarity carries out Error processing, and using result as the confidence level.
Optionally, based on the confidence level, when obtaining corresponding figure recognition result, recognition unit 42 is used for:
Using shape sample corresponding to confidence level highest positioning result as figure recognition result.
Optionally, recognition unit 42 is further used for:
Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and is identified knot with the figure
Object as fruit.
Under second of scene:
Acquiring unit 40, for obtaining picture to be detected;
Positioning unit 41, for the shape sample included using default each coordinate recurrence device, respectively to described to be checked
The figure included in mapping piece is positioned, and obtains corresponding positioning result and confidence level, wherein, a coordinate returns device bag
The confidence level of positioning result corresponding to the shape sample contained, characterize one coordinate and return shape sample that device includes and described
The similarity of figure;
Recognition unit is used for 42, based on the confidence level, obtains corresponding figure recognition result.
Optionally, when obtaining picture to be detected, acquiring unit 40 is used for:
Original image is obtained, and using the original image as picture to be detected;Or
Original image is obtained, and designated area is intercepted in the original image according to setting means, and by the finger
Region is determined as picture to be detected.
Optionally, the shape sample included using default each coordinate recurrence device, respectively in the picture to be detected
Comprising figure positioned when, positioning unit 41 is used for:
The shape sample that devices include is returned using default whole coordinates, respectively the figure to being included in the picture to be detected
Shape is positioned;Or
Feature extraction is carried out to the picture to be detected, the class of the picture ownership to be detected is determined according to extraction result
Mesh, then the shape sample included based on the corresponding default each coordinate recurrence device of classification, respectively to the picture to be detected
In the figure that includes positioned.
Optionally, the shape sample included using default coordinate recurrence device, to being included in the picture to be detected
Figure positioned, when obtaining corresponding positioning result, positioning unit 41 is used for:
The shape sample included using one coordinate recurrence device, Graphic Pattern Matching is carried out in the picture to be detected;
It is determined that during in the presence of the figure that the match is successful, the shape sample that device includes is returned according to one coordinate, described
Draw frame in the periphery of figure;
The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
Optionally, when obtaining the confidence level of the positioning result, positioning unit 41 is used for:
The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, it is right
The similarity carries out Error processing, and using result as the confidence level.
Optionally, based on the confidence level, when obtaining corresponding figure recognition result, recognition unit 42 is used for:
The shape sample that coordinate recurrence device corresponding to confidence level highest positioning result is included, identifies as figure and ties
Fruit.
Optionally, recognition unit 42 is used for:
Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and is identified knot with the figure
Object as fruit.
In summary,, first will be such a for the classification that each body form changes greatly in classification in the embodiment of the present application
Classification is divided into some subclasses of corresponding different objects shape, then individually trains a coordinate to return for each subclass
Device, wherein, a coordinate returns device and characterizes a kind of shape sample.
So, when detection means needs to carry out figure identification to picture to be detected, picture to be detected will be inputted every
One coordinate returns device, i.e., using default each shape sample, the figure included in picture to be detected is positioned respectively,
Corresponding positioning result and confidence level are obtained, finally, then based on confidence level highest positioning result, corresponding figure is obtained and knows
Other result;So, because there is provided the shape sample to become more meticulous, therefore, the accuracy of identification that each coordinate returns device obtains
Ensure, then, can be to effectively improve the identification precision of the same class figure of different objects shape now, and then follow-up
When carrying out object detection, accuracy of detection can be further improved, effectively increases Consumer's Experience.
It should be understood by those skilled in the art that, embodiments herein can be provided as method, system or computer program
Product.Therefore, the application can use the reality in terms of complete hardware embodiment, complete software embodiment or combination software and hardware
Apply the form of example.Moreover, the application can use the computer for wherein including computer usable program code in one or more
The computer program production that usable storage medium is implemented on (including but is not limited to magnetic disk storage, CD-ROM, optical memory etc.)
The form of product.
The application is with reference to the flow according to the method for the embodiment of the present application, equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that can be by every first-class in computer program instructions implementation process figure and/or block diagram
Journey and/or the flow in square frame and flow chart and/or block diagram and/or the combination of square frame.These computer programs can be provided
The processors of all-purpose computer, special-purpose computer, Embedded Processor or other programmable data processing devices is instructed to produce
A raw machine so that produced by the instruction of computer or the computing device of other programmable data processing devices for real
The device for the function of being specified in present one flow of flow chart or one square frame of multiple flows and/or block diagram or multiple square frames.
These computer program instructions, which may be alternatively stored in, can guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works so that the instruction being stored in the computer-readable memory, which produces, to be included referring to
Make the manufacture of device, the command device realize in one flow of flow chart or multiple flows and/or one square frame of block diagram or
The function of being specified in multiple square frames.
These computer program instructions can be also loaded into computer or other programmable data processing devices so that counted
Series of operation steps is performed on calculation machine or other programmable devices to produce computer implemented processing, so as in computer or
The instruction performed on other programmable devices is provided for realizing in one flow of flow chart or multiple flows and/or block diagram one
The step of function of being specified in individual square frame or multiple square frames.
Although having been described for the preferred embodiment of the application, those skilled in the art once know basic creation
Property concept, then can make other change and modification to these embodiments.So appended claims be intended to be construed to include it is excellent
Select embodiment and fall into having altered and changing for the application scope.
Obviously, those skilled in the art can carry out various changes and modification without departing from this Shen to the embodiment of the present application
Please embodiment spirit and scope.So, if these modifications and variations of the embodiment of the present application belong to the application claim
And its within the scope of equivalent technologies, then the application is also intended to comprising including these changes and modification.
Claims (23)
- A kind of 1. pattern recognition system, it is characterised in that including:Client, for obtaining picture to be detected;Server, for receiving the picture to be detected of client transmissions, using default each shape sample, treated respectively to described The figure included in detection picture is positioned, and obtains corresponding positioning result and confidence level, wherein, a shape sample pair The confidence level for the positioning result answered, characterize the similarity of one shape sample and the figure;Based on the confidence level, obtain Obtain corresponding figure recognition result.
- 2. the system as claimed in claim 1, it is characterised in that when obtaining picture to be detected, the client is used for:Original image is obtained, and using the original image as picture to be detected;OrOriginal image is obtained, and designated area is intercepted in the original image according to setting means, and by the specified area Domain is as picture to be detected.
- 3. the system as claimed in claim 1, it is characterised in that default each shape sample is used, respectively to described to be checked When the figure included in mapping piece is positioned, the server is used for:Using default shape sample, the figure included in the picture to be detected is positioned respectively;OrFeature extraction is carried out to the picture to be detected, the classification of the picture ownership to be detected is determined according to extraction result, then Based on the corresponding default each shape sample of classification, the figure included in the picture to be detected is positioned respectively.
- 4. the system as described in claim 1,2 or 3, it is characterised in that a default shape sample is used, to described to be checked The figure included in mapping piece is positioned, and when obtaining corresponding positioning result, the server is used for:Using one shape sample, Graphic Pattern Matching is carried out in the picture to be detected;It is determined that during in the presence of the figure that the match is successful, according to one shape sample, frame is drawn on the periphery of the figure;The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
- 5. system as claimed in claim 4, it is characterised in that when obtaining the confidence level of the positioning result, the server For:The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, to described Similarity carries out Error processing, and using result as the confidence level.
- 6. the system as described in claim 1,2 or 3, it is characterised in that based on the confidence level, obtain corresponding figure identification When as a result, the server is used for:Using shape sample corresponding to confidence level highest positioning result as figure recognition result.
- 7. the system as described in claim 1,2 or 3, it is characterised in that server is further used for:Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and the figure recognition result class As object.
- A kind of 8. pattern recognition method, it is characterised in that including:Obtain picture to be detected;Using default each shape sample, the figure included in the picture to be detected is positioned respectively, obtained corresponding Positioning result and confidence level, wherein, the confidence level of positioning result, characterizes one shape corresponding to a shape sample The similarity of sample and the figure;Based on the confidence level, corresponding figure recognition result is obtained.
- 9. method as claimed in claim 8, it is characterised in that the acquisition picture to be detected, including:Original image is obtained, and using the original image as picture to be detected;OrOriginal image is obtained, and designated area is intercepted in the original image according to setting means, and by the specified area Domain is as picture to be detected.
- 10. method as claimed in claim 8, it is characterised in that it is described to use default each shape sample, respectively to described The figure included in picture to be detected is positioned, including:Using default shape sample, the figure included in the picture to be detected is positioned respectively;OrFeature extraction is carried out to the picture to be detected, the classification of the picture ownership to be detected is determined according to extraction result, then Based on the corresponding default each shape sample of classification, the figure included in the picture to be detected is positioned respectively.
- 11. the method as described in claim 8,9 or 10, it is characterised in that use a default shape sample, treated to described The figure included in detection picture is positioned, and obtains corresponding positioning result, including:Using one shape sample, Graphic Pattern Matching is carried out in the picture to be detected;It is determined that during in the presence of the figure that the match is successful, according to one shape sample, frame is drawn on the periphery of the figure;The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
- 12. method as claimed in claim 11, it is characterised in that the confidence level of the positioning result is obtained, including:The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, to described Similarity carries out Error processing, and using result as the confidence level.
- 13. the method as described in claim 8,9 or 10, it is characterised in that based on the confidence level, obtain corresponding figure and know Other result, including:Using shape sample corresponding to confidence level highest positioning result as figure recognition result.
- 14. the method as described in claim 8,9 or 10, it is characterised in that further comprise:Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and the figure recognition result class As object.
- A kind of 15. pattern recognition method, it is characterised in that including:Obtain picture to be detected;The shape sample included using default each coordinate recurrence device, is entered to the figure included in the picture to be detected respectively Row positioning, obtains corresponding positioning result and confidence level, wherein, a coordinate returns fixed corresponding to the shape sample that device includes The confidence level of position result, characterize the similarity that one coordinate returns shape sample that device includes and the figure;Based on the confidence level, corresponding figure recognition result is obtained.
- 16. method as claimed in claim 15, it is characterised in that the acquisition picture to be detected, including:Original image is obtained, and using the original image as picture to be detected;OrOriginal image is obtained, and designated area is intercepted in the original image according to setting means, and by the specified area Domain is as picture to be detected.
- 17. method as claimed in claim 15, it is characterised in that the shape included using default each coordinate recurrence device Shape sample, the figure included in the picture to be detected is positioned respectively, including:The shape sample included using default whole coordinates recurrence devices, is entered to the figure included in the picture to be detected respectively Row positioning;OrFeature extraction is carried out to the picture to be detected, the classification of the picture ownership to be detected is determined according to extraction result, then The shape sample included based on the corresponding default each coordinate recurrence device of classification, respectively to being included in the picture to be detected Figure positioned.
- 18. the method as described in claim 15,16 or 17, it is characterised in that device is returned using a default coordinate and included Shape sample, the figure included in the picture to be detected is positioned, obtains corresponding positioning result, including:The shape sample included using one coordinate recurrence device, Graphic Pattern Matching is carried out in the picture to be detected;It is determined that during in the presence of the figure that the match is successful, the shape sample that device includes is returned according to one coordinate, in the figure Periphery draw frame;The coordinate position of the frame is determined, the positioning result using the coordinate position as the figure.
- 19. method as claimed in claim 18, it is characterised in that the confidence level of the positioning result is obtained, including:The similarity of the positioning result and the figure is calculated, using the similarity as the confidence level, or, to described Similarity carries out Error processing, and using result as the confidence level.
- 20. the method as described in claim 15,16 or 17, it is characterised in that based on the confidence level, obtain corresponding figure Recognition result, including:The shape sample that coordinate recurrence device corresponding to confidence level highest positioning result is included, as figure recognition result.
- 21. the method as described in claim 15,16 or 17, it is characterised in that further comprise:Based on the figure recognition result, object detection is carried out within a preset range, is filtered out and the figure recognition result class As object.
- A kind of 22. pattern recognition device, it is characterised in that including:Acquiring unit, for obtaining picture to be detected;Positioning unit, for using default each shape sample, the figure included in the picture to be detected is carried out respectively Positioning, obtains corresponding positioning result and confidence level, wherein, the confidence level of positioning result, table corresponding to a shape sample Levy the similarity of one shape sample and the figure;Recognition unit, for based on the confidence level, obtaining corresponding figure recognition result.
- A kind of 23. pattern recognition method, it is characterised in that including:Acquiring unit, for obtaining picture to be detected;Positioning unit, for the shape sample included using default each coordinate recurrence device, respectively to the picture to be detected In the figure that includes positioned, obtain corresponding positioning result and confidence level, wherein, a coordinate returns the shape that device includes The confidence level of positioning result corresponding to shape sample, characterize one coordinate and return shape sample that device includes and the figure Similarity;Recognition unit, for based on the confidence level, obtaining corresponding figure recognition result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610659370.5A CN107729910A (en) | 2016-08-11 | 2016-08-11 | A kind of method, apparatus and system of figure identification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610659370.5A CN107729910A (en) | 2016-08-11 | 2016-08-11 | A kind of method, apparatus and system of figure identification |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107729910A true CN107729910A (en) | 2018-02-23 |
Family
ID=61200580
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610659370.5A Pending CN107729910A (en) | 2016-08-11 | 2016-08-11 | A kind of method, apparatus and system of figure identification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107729910A (en) |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101216841A (en) * | 2008-01-14 | 2008-07-09 | 南京搜拍信息技术有限公司 | Interactive type image search system and method |
CN102024049A (en) * | 2010-12-08 | 2011-04-20 | 中国科学院自动化研究所 | Image retrieval method used in e-commerce platform |
CN102063436A (en) * | 2009-11-18 | 2011-05-18 | 腾讯科技(深圳)有限公司 | System and method for realizing merchandise information searching by using terminal to acquire images |
CN102819566A (en) * | 2012-07-17 | 2012-12-12 | 杭州淘淘搜科技有限公司 | Cross-catalogue indexing method for business images |
CN104143076A (en) * | 2013-05-09 | 2014-11-12 | 腾讯科技(深圳)有限公司 | Matching method and system for face shape |
CN104584033A (en) * | 2012-06-08 | 2015-04-29 | 新加坡国立大学 | Interactive clothes searching in online stores |
CN104573715A (en) * | 2014-12-30 | 2015-04-29 | 百度在线网络技术(北京)有限公司 | Recognition method and device for image main region |
-
2016
- 2016-08-11 CN CN201610659370.5A patent/CN107729910A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101216841A (en) * | 2008-01-14 | 2008-07-09 | 南京搜拍信息技术有限公司 | Interactive type image search system and method |
CN102063436A (en) * | 2009-11-18 | 2011-05-18 | 腾讯科技(深圳)有限公司 | System and method for realizing merchandise information searching by using terminal to acquire images |
CN102024049A (en) * | 2010-12-08 | 2011-04-20 | 中国科学院自动化研究所 | Image retrieval method used in e-commerce platform |
CN104584033A (en) * | 2012-06-08 | 2015-04-29 | 新加坡国立大学 | Interactive clothes searching in online stores |
CN102819566A (en) * | 2012-07-17 | 2012-12-12 | 杭州淘淘搜科技有限公司 | Cross-catalogue indexing method for business images |
CN104143076A (en) * | 2013-05-09 | 2014-11-12 | 腾讯科技(深圳)有限公司 | Matching method and system for face shape |
CN104573715A (en) * | 2014-12-30 | 2015-04-29 | 百度在线网络技术(北京)有限公司 | Recognition method and device for image main region |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109284729B (en) | Method, device and medium for acquiring face recognition model training data based on video | |
US10007862B2 (en) | System and method for suspect search | |
TWI559242B (en) | Visual clothing retrieval | |
US9336459B2 (en) | Interactive content generation | |
TWI623842B (en) | Image search and method and device for acquiring image text information | |
CN110135246A (en) | A kind of recognition methods and equipment of human action | |
US9299009B1 (en) | Utilizing color descriptors to determine color content of images | |
JP2022510712A (en) | Neural network training method and image matching method, as well as equipment | |
CN108229324A (en) | Gesture method for tracing and device, electronic equipment, computer storage media | |
WO2017114237A1 (en) | Image query method and device | |
US11475500B2 (en) | Device and method for item recommendation based on visual elements | |
US10650234B2 (en) | Eyeball movement capturing method and device, and storage medium | |
CN105117399B (en) | Image searching method and device | |
WO2018170695A1 (en) | Method and device for recognizing descriptive attributes of appearance feature | |
CN110647906A (en) | Clothing target detection method based on fast R-CNN method | |
CN107533547A (en) | Product index editing method and its system | |
CN113557546B (en) | Method, device, equipment and storage medium for detecting associated objects in image | |
CN111126102A (en) | Personnel searching method and device and image processing equipment | |
JP5791751B2 (en) | Image recognition method and image recognition apparatus | |
CN111126457A (en) | Information acquisition method and device, storage medium and electronic device | |
CN107729910A (en) | A kind of method, apparatus and system of figure identification | |
CN111368674B (en) | Image recognition method and device | |
CN111008210B (en) | Commodity identification method, commodity identification device, codec and storage device | |
CN110363251B (en) | SKU image classification method and device, electronic equipment and storage medium | |
CN104484324B (en) | A kind of pedestrian retrieval method of multi-model and fuzzy color |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180223 |