CN109325945A - Image processing method, device, electronic equipment and storage medium - Google Patents
Image processing method, device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN109325945A CN109325945A CN201811071355.4A CN201811071355A CN109325945A CN 109325945 A CN109325945 A CN 109325945A CN 201811071355 A CN201811071355 A CN 201811071355A CN 109325945 A CN109325945 A CN 109325945A
- Authority
- CN
- China
- Prior art keywords
- sampled point
- abscissa
- ordinate
- adjusted
- convolution
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The embodiment of the present application provides a kind of image processing method, device, electronic equipment and storage medium, is related to technical field of image processing.Method includes: to obtain the relevant form parameter of shape of image sampling frame；According to the convolution nuclear parameter of form parameter and convolution kernel, each sampled point at least two sampled points of convolution kernel is adjusted from current first position to the second position corresponding with form parameter, obtain convolution kernel adjusted, wherein, the position that any two sampled point at least two sampled points is presently in is not identical, and any two sampled point position adjusted is not also identical；Process of convolution is carried out to the image in image sampling frame according to convolution kernel adjusted.Process of convolution is carried out to the image in image sampling frame with image sampling frame shape matched convolution kernel by sampling point position distribution, so that it may avoid deviation occur to the testing result of image, to also improve the precision of detection.
Description
Technical field
This application involves technical field of image processing, set in particular to a kind of image processing method, device, electronics
Standby and storage medium.
Background technique
Currently, the image sampling frame that frame selects objects in images can be generated, to adopt to image in target detection technique
Target object image in sample frame is further detected and is identified.
But the randomness of the size and location due to target object image in the picture is bigger, leads to the image generated
The shape size otherness of sample boxes is also very big.Therefore since the shape size of image sampling frame is not fixed, cause to adopt to image
During target object image in sample frame is detected, there is deviation to the testing result of object, so as to cause detection
Precision is affected.
Summary of the invention
The application is to provide a kind of image processing method, device, electronic equipment and storage medium, to effectively prevent to mesh
There is deviation in the testing result of mark object, to improve the precision of detection.
To achieve the goals above, embodiments herein is accomplished in that
In a first aspect, the embodiment of the present application provides a kind of image processing method, which comprises
Obtain the relevant form parameter of shape of image sampling frame；
It, will be at least two sampled points of the convolution kernel according to the convolution nuclear parameter of the form parameter and convolution kernel
Each sampled point is adjusted from current first position to the second position corresponding with the form parameter, obtains convolution adjusted
Core, wherein the position that any two sampled point at least two sampled point is presently in is not identical, any two sampling
Point position adjusted is not also identical；
Process of convolution is carried out according to the convolution collecting image adjusted.
With reference to first aspect, the embodiment of the present application provides the first possible embodiment of first aspect, the shape
Shape parameter includes: the minification of the length and width and described image sample boxes of described image sample boxes in down-sampling；
The convolution nuclear parameter according to the form parameter and convolution kernel, by each of at least two sampled points of the convolution kernel
Sampled point is adjusted from current first position to the second position corresponding with the form parameter, obtains convolution kernel adjusted,
Include:
According to the length, the minification and the convolution nuclear parameter, by least two samplings of the convolution kernel
The abscissa of each sampled point in point is adjusted to the second abscissa corresponding with the length from the first current abscissa；
According to the width, the minification and the convolution nuclear parameter, by the ordinate of each sampled point from current
The first ordinate be adjusted to the second ordinate corresponding with the width, wherein the coordinate of each sampled point before adjustment is
It is at current first position and adjusted to indicate that the position of each sampled point is located at for first abscissa and the first ordinate
The coordinate of each sampled point is the second abscissa and the second ordinate, indicates that the position of each sampled point is located at adjusted second
At position；
Convolution kernel adjusted is obtained, the position of each sampled point is located at the second position in the convolution kernel adjusted
Place.
With reference to first aspect, the embodiment of the present application provides second of possible embodiment of first aspect, and described
According to the length, the minification and the convolution nuclear parameter, at least two sampled points by the convolution kernel
The abscissa of each sampled point is adjusted to the second abscissa corresponding with the length from the first current abscissa, comprising:
According to the length, the minification and the convolution nuclear parameter, at least two of the convolution kernel are determined
First deviation of the abscissa of each sampled point in sampled point；
It is according to the first deviation of the abscissa of each sampled point, the abscissa of each sampled point is horizontal from current first
Coordinate Adjusting is the second abscissa corresponding with the length.
With reference to first aspect, the embodiment of the present application provides the third possible embodiment of first aspect, the volume
Product nuclear parameter include: the convolution kernel total columns and each sampled point column and the convolution kernel central point it
Between interval columns；It is described according to the length, the minification and convolution nuclear parameter, determine every in the convolution kernel
First deviation of the abscissa of a sampled point, comprising:
According to the length W, the minification S, total columns M and the interval columns m, the convolution is determined
First deviation of the abscissa of each sampled point in core is W*m/M*S.
With reference to first aspect, the embodiment of the present application provides the 4th kind of possible embodiment of first aspect, and described
According to the width, the minification and the convolution nuclear parameter, the ordinate by each sampled point is from current first
Ordinate is adjusted to the second ordinate corresponding with the width, comprising:
According to the width, the minification and the convolution nuclear parameter, the ordinate of each sampled point is determined
Second deviation；
According to the second deviation of the ordinate of each sampled point, the ordinate of each sampled point is indulged from current first
Coordinate Adjusting is the second ordinate corresponding with the length.
With reference to first aspect, the embodiment of the present application provides the 5th kind of possible embodiment of first aspect, the volume
Product nuclear parameter further include: total line number of the convolution kernel and each sampled point be expert at in the convolution kernel
Interval line number between heart point；It is described according to the width, the minification and the convolution nuclear parameter, determine each adopt
Second deviation of the ordinate of sampling point, including；
According to the width H, the minification S, total line number N and the interval line number n, each sampling is determined
Second deviation H*n/N*S of the ordinate of point.
With reference to first aspect, the first of first aspect formula, this Shen either into the 5th kind of possible embodiment
Please embodiment provide the 6th kind of possible embodiment of first aspect, it is described obtain image sampling frame shape before,
The method also includes:
Obtain the first position parameter of the central point of described image sample boxes；
According to the second position parameter of the first position parameter and the central point of the convolution kernel, by each sampled point from
Current the third place is adjusted to first position, wherein each sampled point, which is in first position, indicates the center of the convolution kernel
Point is overlapped with the central point of described image sample boxes.
With reference to first aspect, the embodiment of the present application provides the 7th kind of possible embodiment of first aspect, and described
One location parameter includes: the first center abscissa and the first center ordinate, and the second position parameter includes: the second center cross
Coordinate and the second center ordinate, it is described to be joined according to the second position of the first position parameter and the central point of the convolution kernel
Number, each sampled point is adjusted from current the third place to first position, comprising:
According to first center abscissa and second center abscissa, by the abscissa of each sampled point from current
Third abscissa be adjusted to the first abscissa；
According to first center ordinate and second center ordinate, by the ordinate of each sampled point from current
Third ordinate be adjusted to the first ordinate, wherein the coordinate of each sampled point be third abscissa and third ordinate table
Show that the position of each sampled point is located at current the third place.
With reference to first aspect, the embodiment of the present application provides the 8th kind of possible embodiment of first aspect, and described
According to first center abscissa and second center abscissa, by the abscissa of each sampled point from the current horizontal seat of third
Mark is adjusted to the first abscissa, comprising:
According to first center abscissa and second center abscissa, determine first center abscissa with
Third deviation between the abscissa of second center；
According to the third deviation, the abscissa of each sampled point is adjusted to the first cross from current third abscissa
Coordinate.
With reference to first aspect, the embodiment of the present application provides the 9th kind of possible embodiment of first aspect, and described
According to first center ordinate and second center ordinate, the ordinate of each sampled point is sat from current third is vertical
Mark is adjusted to the first ordinate, comprising:
According to first center ordinate and second center ordinate, determine first center ordinate with
The 4th deviation between the ordinate of second center；
According to the 4th deviation, the ordinate of each sampled point is adjusted to first from current third ordinate and is indulged
Coordinate.
In conjunction with second aspect, the embodiment of the present application provides a kind of image processing apparatus, and described device includes:
Shape obtains module, the relevant form parameter of shape for obtaining image sampling frame；
Adjusting Shape module, for the convolution nuclear parameter according to the form parameter and convolution kernel, by the convolution kernel
Each sampled point at least two sampled points is adjusted from current first position to second corresponding with the form parameter
It sets, obtains convolution kernel adjusted, wherein the position that any two sampled point at least two sampled point is presently in
Not identical, any two sampled point position adjusted is not also identical；
Convolution module, for carrying out process of convolution according to the convolution collecting image adjusted.
In conjunction with second aspect, the embodiment of the present application provides the first possible embodiment of second aspect, the shape
Shape parameter includes: the minification of the length and width and described image sample boxes of described image sample boxes in down-sampling；
The Adjusting Shape module is also used to according to the length, the minification and the convolution nuclear parameter, by institute
State the abscissa of each sampled point at least two sampled points of convolution kernel from the first current abscissa be adjusted to it is described
Corresponding second abscissa of length；According to the width, the minification and the convolution nuclear parameter, by each sampled point
Ordinate is adjusted to the second ordinate corresponding with the width from the first current ordinate, wherein adopts each of before adjustment
The coordinate of sampling point is the first abscissa and the first ordinate, indicates that the position of each sampled point is located at current first position,
And the coordinate of each sampled point adjusted is the second abscissa and the second ordinate, indicates that the position of each sampled point is located at
The second place adjusted；Convolution kernel adjusted is obtained, the position position of each sampled point in the convolution kernel adjusted
In the second place.
In conjunction with second aspect, the embodiment of the present application provides second of possible embodiment of second aspect, and described
According to the length, the minification and the convolution nuclear parameter,
The Adjusting Shape module is also used to be determined according to the length, the minification and the convolution nuclear parameter
First deviation of the abscissa of each sampled point at least two sampled points of the convolution kernel out；According to each sampled point
Abscissa the first deviation, the abscissa of each sampled point is adjusted to and the length pair from the first current abscissa
The second abscissa answered.
In conjunction with second aspect, the embodiment of the present application provides the third possible embodiment of second aspect, the volume
Product nuclear parameter include: the convolution kernel total columns and each sampled point column and the convolution kernel central point it
Between interval columns；
The Adjusting Shape module is also used to according to the length W, the minification S, total columns M and described
It is spaced columns m, determines that the first deviation of the abscissa of each sampled point in the convolution kernel is W*m/M*S.
In conjunction with second aspect, the embodiment of the present application provides the 4th kind of possible embodiment of second aspect, and described
According to the width, the minification and the convolution nuclear parameter,
The Adjusting Shape module is also used to be determined according to the width, the minification and the convolution nuclear parameter
Second deviation of the ordinate of each sampled point out；According to the second deviation of the ordinate of each sampled point, adopted each
The ordinate of sampling point is adjusted to the second ordinate corresponding with the length from the first current ordinate.
In conjunction with second aspect, the embodiment of the present application provides the 5th kind of possible embodiment of second aspect, the volume
Product nuclear parameter further include: total line number of the convolution kernel and each sampled point be expert at in the convolution kernel
Interval line number between heart point；
The Adjusting Shape module is also used to according to the width H, the minification S, total line number N and described
It is spaced line number n, determines the second deviation H*n/N*S of the ordinate of each sampled point.
In conjunction with second aspect, second aspect the first either into the 5th kind of possible embodiment formula, this Shen
Please embodiment provide the 6th kind of possible embodiment of second aspect, described device further include:
Position obtains module, the first position parameter of the central point for obtaining described image sample boxes；
Position adjusting type modules, for being joined according to the second position of the first position parameter and the central point of the convolution kernel
Number, each sampled point is adjusted from current the third place to first position, wherein each sampled point is in first position expression
The central point of the convolution kernel is overlapped with the central point of described image sample boxes.
In conjunction with second aspect, the embodiment of the present application provides the 7th kind of possible embodiment of second aspect, and described
One location parameter includes: the first center abscissa and the first center ordinate, and the second position parameter includes: the second center cross
Coordinate and the second center ordinate,
The position adjusting type modules are also used to according to first center abscissa and second center abscissa, will
The abscissa of each sampled point is adjusted to the first abscissa from current third abscissa；According to first center ordinate and
The ordinate of each sampled point is adjusted to the first ordinate from current third ordinate by second center ordinate,
In, the coordinate of each sampled point is that third abscissa and third ordinate indicate that the position of each sampled point is located at current third
At position.
In conjunction with second aspect, the embodiment of the present application provides the 8th kind of possible embodiment of second aspect,
The position adjusting type modules are also used to according to first center abscissa and second center abscissa, really
Make the third deviation between first center abscissa and second center abscissa；According to the third deviation
Value, is adjusted to the first abscissa from current third abscissa for the abscissa of each sampled point.
In conjunction with second aspect, the embodiment of the present application provides the 9th kind of possible embodiment of second aspect,
The position adjusting type modules are also used to according to first center ordinate and second center ordinate, really
Make the 4th deviation between first center ordinate and second center ordinate；According to the 4th deviation
Value, is adjusted to the first ordinate from current third ordinate for the ordinate of each sampled point.
The third aspect, the embodiment of the present application provide a kind of electronic equipment, and the electronic equipment includes: processor, storage
Device, bus and communication interface；The processor, the communication interface and memory are connected by the bus；
The memory, for storing program；
The processor, for the program by calling storage in the memory, to execute first aspect and first
Image processing method described in any embodiment of aspect.
Fourth aspect, the embodiment of the present application provide a kind of calculating of non-volatile program code that can be performed with computer
The readable storage medium of machine, said program code make any embodiment party of the computer execution first aspect and first aspect
Image processing method described in formula.
The beneficial effect of the embodiment of the present application is:
According to the relevant form parameter of shape of image sampling frame, by every at least two sampled points by convolution kernel
A sampled point is adjusted from current first position to the second position corresponding with form parameter, so that convolution kernel adjusted
The position distribution of interior at least two sampled point is matched with the shape of image sampling frame.Therefore, it is distributed and schemes by sampling point position
As the matched convolution kernel of sample boxes shape to carry out process of convolution to the image in image sampling frame, so that it may avoid to image
There is deviation in testing result, to also improve the precision of detection.
To enable the above objects, features, and advantages of the application to be clearer and more comprehensible, preferred embodiment is cited below particularly, and cooperate
Appended attached drawing, is described in detail below.
Detailed description of the invention
Technical solution in ord to more clearly illustrate embodiments of the present application, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only some embodiments of the application, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 shows the structural block diagram of a kind of electronic equipment of the application first embodiment offer；
Fig. 2 shows a kind of first pass figures for image processing method that the application second embodiment provides；
Fig. 3 shows the sub-process figure of step S120 in a kind of image processing method of the application second embodiment offer；
Fig. 4 shows the sub-process figure of step S140 in a kind of image processing method of the application second embodiment offer；
Fig. 5 shows a kind of the first schematic diagram of a scenario of image processing method of the application second embodiment offer；
Fig. 6 shows a kind of the second schematic diagram of a scenario of image processing method of the application second embodiment offer；
Fig. 7 shows a kind of structural block diagram of image processing apparatus of the application 3rd embodiment offer.
Specific embodiment
Below in conjunction with attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Ground description, it is clear that described embodiments are only a part of embodiments of the present application, instead of all the embodiments.Usually exist
The component of the embodiment of the present application described and illustrated in attached drawing can be arranged and be designed with a variety of different configurations herein.Cause
This, is not intended to limit claimed the application's to the detailed description of the embodiments herein provided in the accompanying drawings below
Range, but it is merely representative of the selected embodiment of the application.Based on embodiments herein, those skilled in the art not into
Row goes out every other embodiment obtained under the premise of creative work, shall fall in the protection scope of this application.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.Term " first ", " the
Two " etc. are only used for distinguishing description, are not understood to indicate or imply relative importance.
First embodiment
Referring to Fig. 1, the embodiment of the present application provides electronic equipment 10, electronic equipment 10 can be terminal device or be
Server.Wherein, terminal device can be PC (personal computer, PC), tablet computer, smart phone, a
Personal digital assistant (personal digital assistant, PDA) etc.；Server can take for network server, database
Business device, Cloud Server or the server set that is made of multiple child servers at etc..
In the present embodiment, which may include: memory 11, communication module 12, bus 13 and processor 14.
Wherein, processor 14, communication module 12 and memory 11 are connected by bus 13.Processor 14 is deposited in memory 11 for executing
The executable module of storage, such as computer program.The component and structure of electronic equipment 10 shown in FIG. 1 are only exemplary, and
Unrestricted, as needed, electronic equipment 10 also can have other assemblies and structure.
Wherein, memory 11 may include high-speed random access memory (Random Access Memory RAM),
It may further include non-labile memory (non-volatile memory), for example, at least a magnetic disk storage.This implementation
In example, memory 11 stores program required for executing image processing method.
Bus 13 can be isa bus, pci bus or eisa bus etc..It is total that bus can be divided into address bus, data
Line, control bus etc..Only to be indicated with a four-headed arrow in Fig. 1, it is not intended that an only bus or one convenient for indicating
The bus of seed type.
Processor 14 may be a kind of IC chip, the processing capacity with signal.During realization, above-mentioned side
Each step of method can be completed by the integrated logic circuit of the hardware in processor 14 or the instruction of software form.Above-mentioned
Processor 14 can be general processor, including central processing unit (Central Processing Unit, abbreviation CPU), network
Processor (Network Processor, abbreviation NP) etc.；It can also be digital signal processor (DSP), specific integrated circuit
(ASIC), ready-made programmable gate array (FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components.It may be implemented or execute disclosed each method, step and the logic diagram in the embodiment of the present invention.It is general
Processor can be microprocessor or the processor is also possible to any conventional processor etc..In conjunction with institute of the embodiment of the present invention
The step of disclosed method, can be embodied directly in hardware decoding processor and execute completion, or with the hardware in decoding processor
And software module combination executes completion.Software module can be located at random access memory, and flash memory, read-only memory may be programmed read-only
In the storage medium of this fields such as memory or electrically erasable programmable memory, register maturation.
Method performed by the device of stream process or definition that any embodiment of the embodiment of the present invention discloses can be applied to
In processor 14, or realized by processor 14.Processor 14 is stored in after receiving and executing instruction by the calling of bus 13
After program in memory 11, processor 14, which controls communication module 12 by bus 13, can then execute the stream of image processing method
Journey.
Second embodiment
Present embodiments provide a kind of image processing method, it should be noted that step shown in the flowchart of the accompanying drawings
It can execute in a computer system such as a set of computer executable instructions, although also, showing patrol in flow charts
Sequence is collected, but in some cases, it can be with the steps shown or described are performed in an order that is different from the one herein.Below to this
Embodiment describes in detail.
Referring to Fig. 2, in image processing method provided in this embodiment, the image processing method include: step S110,
Step S120, step S130, step S140 and step S150.
Step S110: the first position parameter of the central point of described image sample boxes is obtained.
Step S120:, will be every according to the second position parameter of the first position parameter and the central point of the convolution kernel
A sampled point is adjusted from current the third place to first position, wherein each sampled point, which is in first position, indicates the volume
The central point of product core is overlapped with the central point of described image sample boxes.
Step S130: the relevant form parameter of shape of image sampling frame is obtained.
Step S140: according to the convolution nuclear parameter of the form parameter and convolution kernel, by at least two of the convolution kernel
Each sampled point in sampled point is adjusted from current first position to the second position corresponding with the form parameter, is adjusted
Convolution kernel after whole, wherein the position that any two sampled point at least two sampled point is presently in is not identical, appoints
Two sampled point positions adjusted of anticipating are not also identical.
Step S150: process of convolution is carried out according to the convolution collecting image adjusted.
Below in conjunction with Fig. 2 to Fig. 4, each step of the application is described in detail.
Step S110: the first position parameter of the central point of described image sample boxes is obtained.
Electronic equipment can carry out detection identification to the object in original image, for example, electronic equipment can use
The modes such as FasterRCNN, RFCN, SSD, RetinaNet, RefineDet and YOLOv2 carry out the object in original image
Detection identification.So electronic equipment can according to preset control program, and according to the position of object in original image and
Size generates the image sampling frame that frame selects object in original image；Or electronic equipment can also respond the image of user
Sample boxes generate operation, generate the desired image sampling frame of user.
It is understood that image sampling frame can exist in a virtual manner, i.e., the actual original image can not have
One actual frame that can be seen by the user.
In the present embodiment, electronic equipment can be pre-established according to the relative position of pixel each in original image with reference to seat
Mark system.And since image sampling frame is located in the original image, then image sampling frame is also located at the reference frame accordingly
In.Therefore during electronic equipment generates the image sampling frame, electronic equipment can also obtain the image sampling frame accordingly and exist
Position in reference frame, i.e. electronic equipment can also obtain the central point of the image sampling frame in reference frame accordingly
In first position parameter.First position parameter can for the image sampling frame central point in reference frame first in
Heart abscissa and the first center ordinate.
Step S120:, will be every according to the second position parameter of the first position parameter and the central point of the convolution kernel
A sampled point is adjusted from current the third place to first position, wherein each sampled point, which is in first position, indicates the volume
The central point of product core is overlapped with the central point of described image sample boxes.
In the present embodiment, step S120 may include: step S121 and step S122.
Step S121: according to first center abscissa and second center abscissa, by the cross of each sampled point
Coordinate is adjusted to the first abscissa from current third abscissa.
Step S122: according to first center ordinate and second center ordinate, by the vertical of each sampled point
Coordinate is adjusted to the first ordinate from current third ordinate, wherein the coordinate of each sampled point is third abscissa and the
Three ordinates indicate that the position of each sampled point is located at current the third place.
While electronic equipment generates image sampling frame, electronic equipment can also be according to preset control Program Generating convolution
Core to carry out process of convolution to image, to realize the sampling to image, wherein the image can be based on original graph for electronic equipment
Comprising any image including original image in the feature pyramid of the image generated as down-sampling.
In the present embodiment, therefore to need during convolution in stepping convenient for accurately convolution can be carried out to image
The central point of convolution kernel can be overlapped with the central point of image sampling frame when to a certain position.So, in order to realize this effect,
First the central point of convolution kernel can be adjusted to be overlapped with the central point of image sampling frame after generating convolution kernel, thus can
Enough guarantee that the convolution kernel can step to central point during subsequent convolution to be overlapped with the central point of image sampling frame.
It should be noted that if the case where image sampling frame is the image sampling frame generation operation in response to user and generates
Under, since image sampling frame is generated based on user's operation, then user can guarantee in image sampling frame when generating
Heart point is overlapped with the central point of the convolution kernel of generation.If but when image sampling frame the case where being automatically generated for electronic equipment, due to
The image sampling frame automatically generated at this time can generate after image sampling frame carries out time sampling user's operation for electronic equipment and return
Return the sample boxes more accurate relative to image out, therefore the central point of image sampling frame can not be with the central point of convolution kernel
It is overlapped.
This programme for ease of understanding is adjusted to by the central point of the central point of convolution kernel and image sampling frame in the present embodiment
To the explanation of coincidence, is generated not for image sampling frame based on user's operation or automatically generated based on electronic equipment, but
It is not intended as the restriction to scheme.Under actual conditions, if image sampling frame is to be generated based on user's operation, electronic equipment can not
Adjust the central point of convolution kernel.
In the present embodiment, electronic equipment, which executes step S121 and step S122, can adjust the central point and figure of convolution kernel
As the central point coincidence of sample boxes, will be illustrated below to step S121 and step S122 is executed:
In detail, the position that is located in the reference frame of the electronic equipment based on convolution kernel, electronic equipment can obtain
Second position parameter of the central point of the convolution kernel in reference frame.Second position parameter can be the center of the convolution kernel
Second center abscissa and second center ordinate of the point in reference frame.
Again since convolution kernel can have at least two sampled points, for example, convolution kernel A is the convolution kernel of 5*5, then the volume
It then can have 25 sampled points in product core A.Therefore for for convolution kernel, each of at least two samplings of convolution kernel are adopted
Sampling point can then be located at current the third place relative to the central point of convolution kernel, and any two at least two sampled points
The position that a sampled point is presently in is not identical, i.e., any two the third place where any two sampled point is not identical.
When the central point of the convolution kernel to be adjusted to be overlapped with the central point of figure sample boxes, due to the center of convolution kernel
Point position change, then the position of each sampled point is also required to change accordingly in convolution kernel.Therefore it can be according in convolution kernel
Difference between heart point and the central point of figure sample boxes is as the foundation being adjusted to each sampled point, thus by each
Sampled point is adjusted from current the third place to first position, wherein the position of each sampled point is located at current the third place
The coordinate that can indicate each sampled point is third abscissa and third ordinate.
Electronic equipment determines the first center abscissa and second according to the first center abscissa and the second center abscissa
Third deviation between the abscissa of center, electronic equipment is according to the third deviation, so that it may by the horizontal seat of each sampled point
Mark is adjusted to the first abscissa from current third abscissa.And electronic equipment is further according to the first center ordinate and described
Second center ordinate determines the 4th deviation between the first center ordinate and the second center ordinate, electronic equipment
According to the 4th deviation, so that it may which the ordinate of each sampled point is adjusted to the first vertical seat from current third ordinate
Mark.In this way, it is subsequent continue adjustment before, the coordinate of each sampled point is that the first abscissa and the first ordinate can indicate
Each sampled point has currently had been adjusted at first position, wherein any two sampled point is worked as at least two sampled points
Any two first position at preceding place is not also identical.
It is assumed that the convolution kernel A generated is the convolution kernel of 5*5, then seat of the central point of convolution kernel A in reference frame
Target second position parameter includes: (Xa, Ya).And at least two sampled points are in convolution kernel A with the seat of the central point of convolution kernel A
It is denoted as reference, then the coordinate of at least two sampled points can be in convolution kernel A are as follows:
Wherein, formula 1 is the distribution of at least two second center abscissas of at least two sampled points of convolution kernel A, and formula 2 is then
For the distribution of at least two second center ordinates of at least two sampled points of convolution kernel A, formula 1 and formula 2 then represent each
Sample is in current the third place.
The first position parameter of coordinate of the central point of image sampling frame B in reference frame includes: (Xb, Yb), root
According to first position parameter (Xb, Yb) and second position parameter (Xa, Ya) by the midpoint of convolution kernel A adjust to image sampling frame B
Central point be overlapped when, then the coordinate of at least two sampled points corresponding can adjust in convolution kernel A are as follows:
Wherein, formula 3 is at least two second center abscissas of at least two sampled points of convolution kernel A convolution kernel A's
Midpoint is adjusted to distribution when being overlapped with the central point of image sampling frame B, and formula 4 is then at least two sampled points of convolution kernel A
At least two second center ordinates adjust at the midpoint of convolution kernel A to distribution when being overlapped with the central point of image sampling frame B,
Formula 3 and formula 4 then represent each sample in current first position.
In the present embodiment, since the shape of the convolution kernel of generation is generally the shape of standard, i.e. each of convolution kernel is adopted
Sampling point and adjacent sampled point are equidistant, allow for the shape and the very strong image sampling frame of shape randomness of convolution kernel in this way
Shape mismatches, and deviation occurs so as to cause sampling.Therefore the central point of convolution kernel not only can be adjusted to and be schemed by electronic equipment
As the central point coincidence of sample boxes, step S130 to step S140 is can also be performed by the Adjusting Shape of convolution kernel in electronic equipment
Extremely matched with the shape of image sampling frame.
Step S130: the relevant form parameter of shape of image sampling frame is obtained.
Since the number of pixel can determine the size of the size of image, then the size of image can also correspond to expression
The size of image sampling frame can also indicate in reference frame in the reference frame and in the image.Therefore in life
While at image sampling frame, electronic equipment can also obtain the relevant form parameter of shape in the image sampling frame, the shape
Relevant form parameter may include: the length and width of the image sampling frame in reference frame.Due in detection process
In, image is comprising any image including original image in the feature pyramid of image, then the image is compared to original graph
As may reduce, then the image sampling frame reduced in the image compared to being automatically, or based on user's operation generation at the beginning
Image sampling frame can also be with Scaling, therefore the relevant form parameter of shape that electronic equipment obtains can also include: the image
Minification of the sample boxes in down-sampling, i.e. the image sampling frame minification can then obtain the image reduced in image and adopt
Sample frame.
Step S140: according to the convolution nuclear parameter of the form parameter and convolution kernel, by at least two of the convolution kernel
Each sampled point in sampled point is adjusted from current first position to the second position corresponding with the form parameter, is adjusted
Convolution kernel after whole, wherein the position that any two sampled point at least two sampled point is presently in is not identical, appoints
Two sampled point positions adjusted of anticipating are not also identical.
In the present embodiment, step S140 may include: step S141, step S142 and step S143.
Step S141: according to the length, the minification and the convolution nuclear parameter, at least by the convolution kernel
The abscissa of each sampled point in two sampled points is adjusted to corresponding with the length second from the first current abscissa
Abscissa.
Step S142: according to the width, the minification and the convolution nuclear parameter, by the vertical seat of each sampled point
Mark is adjusted to the second ordinate corresponding with the width from the first current ordinate.
Step S143: convolution kernel adjusted is obtained, the position of each sampled point is located in the convolution kernel adjusted
The second place.
In the present embodiment, while electronic equipment generates convolution kernel, electronic equipment can also obtain the convolution of the convolution kernel
Nuclear parameter.Wherein, convolution nuclear parameter may include: convolution kernel total line number and total columns, and, at the center for determining convolution kernel
Point in the case where, convolution nuclear parameter can also be include: between the column of each sampled point and the central point of convolution kernel between
Every columns, the interval line number of each sampled point being expert between the central point of convolution kernel.
For example, convolution kernel A is the convolution kernel of 5*5, then total line number of convolution kernel A and total columns can be with are as follows: 5 rows and 5 column.
The interval columns of the central point of sampled point and convolution kernel A on a most marginal column can be 3.And it is located at and most side
The interval line number of the central point of sampled point and convolution kernel A in the adjacent a line of a line of edge can be 2.
In the present embodiment, for electronic equipment by executing step S141 and step S142, electronic equipment can be by each sampling
The first current abscissa of point is adjusted to the second abscissa corresponding with length, and can be again by the current of each sampled point
The first ordinate be adjusted to the mode of the second ordinate corresponding with length.To realize the coordinate of each sampled point with length
Degree is corresponding with width, then realizes that the shape of convolution kernel is matched with the shape of image sampling frame, so that it may adjust in convolution kernel
Heart point is overlapped with the central point of image sampling frame.It will be illustrated below to step S141, step S142 and step S143 is executed:
Electronic equipment is according to length, minification and according to the total columns and interval columns in convolution nuclear parameter, electronics
Equipment is assured that out that the abscissa of each sampled point at least two sampled points of convolution kernel is reduced relative to this
The first deviation that the length of small multiple generates.Then, the of the abscissa according to each sampled point that electronic equipment is determined
It is horizontal then can be adjusted to corresponding with length second from the first current abscissa by one deviation for the abscissa of each sampled point
Coordinate.
Simultaneously, electronic equipment according to width, minification and according in convolution nuclear parameter total line number and interval
Line number, electronic equipment are assured that out the ordinate of each sampled point at least two sampled points of convolution kernel relative to this
Reduce the second deviation that the width of minification generates.Then, the indulging according to each sampled point that electronic equipment is determined
The ordinate of each sampled point then can be adjusted to corresponding with width from the first current ordinate by the second deviation of coordinate
The second ordinate.In this way, the coordinate of each sampled point adjusted is that the second abscissa and the second ordinate can indicate every
A sampled point has currently had been adjusted to the second place, wherein any two sampled point is current at least two sampled points
Any two first position at place is not also identical.
Optionally it is determined that the first deviation and length that go out, minification, total columns and be spaced the relationship of columns can be with
Are as follows: length W, minification S, total columns M and interval columns m, the abscissa of each sampled point in the convolution kernel determined
First deviation can be W*m/M*S.And the second deviation determined and width, minification, total line number and spaced rows
Several relationships can be with are as follows: width H, minification S, total line number N and interval line number n, each sampling in the convolution kernel determined
Second deviation of the ordinate of point can be H*n/N*S.
It is understood that the sampled point in the row or column at convolution kernel edge is more proximate in convolution kernel, these sampled points
Adjustment amplitude is bigger.
Continue above-mentioned it is assumed that formula 3 and formula 4 then represent each sample in convolution kernel A in current first
It sets.If the length of image sampling frame B is W, minification S, total columns are M and interval columns is m and image sampling frame B
Width be H, total line number be N and interval behavior number n.Each sample of convolution kernel A so adjusted is in current
The coordinate distribution of two positions can be as follows:
Wherein, formula 5 is abscissa of each sample when the second place at least two sampled points of convolution kernel A
Distribution and formula 6 are ordinate of each sample when the second place point at least two sampled points of convolution kernel A
Cloth.So formula 5 and formula 6 then indicates modulated convolution kernel.
Step S150: process of convolution is carried out according to the convolution collecting image adjusted.
After electronic equipment obtains convolution kernel adjusted, the image of image can be adopted according to the convolution kernel adjusted
Feature in sample frame carries out process of convolution.
It is understood that convolution kernel adjusted is carried out at convolution with the characteristic target object in the image sampling frame to image
Reason can be with are as follows: the convolution kernel adjusted carries out process of convolution to the feature at the convolution kernel present position adjusted.
It, can be according to preceding method again if step to next sampling location in image sampling frame with convolution kernel preset step-length
The position distribution for adjusting at least two sampled points of convolution kernel is matched with the shape of image sampling frame, to obtain again adjusted
Convolution kernel samples next sampling location.
It should be noted that convolution kernel adjusted is during convolution, each sampled point in convolution kernel adjusted
The sampling feature of sampling can obtain for the value away from each nearest four dispersed feature points of sampled point by bilinear interpolation.
As shown in Figure 5 and Figure 6, the convolution kernel A1 in Fig. 5 before adjustment is the convolution kernel of 3*3, each sampling in convolution kernel A1
First position of the point (circle shown in dotted line) before adjustment, and image sampling frame is B.So, by adjusting acquisition and image
Sample boxes are that each sampled point (circle shown in solid) is in the second position before adjustment in the convolution kernel A2 adjusted of B.
3rd embodiment
Referring to Fig. 7, the embodiment of the present application provides a kind of image processing apparatus 100, which can be with
Applied to electronic equipment, which includes:
Shape obtains module 110, the relevant form parameter of shape for obtaining image sampling frame.
Adjusting Shape module 120, for the convolution nuclear parameter according to the form parameter and convolution kernel, by the convolution kernel
At least two sampled points in each sampled point adjust from current first position to corresponding with the form parameter second
Position obtains convolution kernel adjusted, wherein the position that any two sampled point at least two sampled point is presently in
Set not identical, any two sampled point position adjusted is not also identical.
Convolution module 130, for carrying out process of convolution according to the convolution collecting image adjusted.
It should be noted that due to it is apparent to those skilled in the art that, for the convenience and letter of description
Clean, system, the specific work process of device and unit of foregoing description can be with reference to corresponding in preceding method embodiment
Journey, details are not described herein.
It should be understood by those skilled in the art that, the embodiment of the present application can provide as the production of method, system or computer program
Product.Therefore, in terms of the embodiment of the present application can be used complete hardware embodiment, complete software embodiment or combine software and hardware
Embodiment form.Moreover, it wherein includes computer available programs generation that the embodiment of the present application, which can be used in one or more,
The meter implemented in the computer-usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) of code
The form of calculation machine program product.
In conclusion the embodiment of the present application provides a kind of image processing method, device, electronic equipment and storage medium.
Method includes: to obtain the relevant form parameter of shape of image sampling frame；According to the convolution nuclear parameter of form parameter and convolution kernel,
Each sampled point at least two sampled points of convolution kernel is adjusted from current first position to corresponding with form parameter
The second position obtains convolution kernel adjusted, wherein the position that any two sampled point at least two sampled points is presently in
Set not identical, any two sampled point position adjusted is not also identical；According to convolution kernel adjusted in image sampling frame
Image carry out process of convolution.
According to the relevant form parameter of shape of image sampling frame, by every at least two sampled points by convolution kernel
A sampled point is adjusted from current first position to the second position corresponding with form parameter, so that convolution kernel adjusted
The position distribution of interior at least two sampled point is matched with the shape of image sampling frame.Therefore, it is distributed and schemes by sampling point position
As the matched convolution kernel of sample boxes shape to carry out process of convolution to the image in image sampling frame, so that it may avoid to image
There is deviation in testing result, to also improve the precision of detection.
The above is only preferred embodiment of the present application, are not intended to limit this application, for those skilled in the art
For member, various changes and changes are possible in this application.Within the spirit and principles of this application, it is made it is any modification,
Equivalent replacement, improvement etc., should be included within the scope of protection of this application.It should also be noted that similar label and letter are under
Similar terms are indicated in the attached drawing in face, therefore, once being defined in a certain Xiang Yi attached drawing, are not then needed in subsequent attached drawing
It is further defined and explained.
More than, the only specific embodiment of the application, but the protection scope of the application is not limited thereto, and it is any to be familiar with
Those skilled in the art within the technical scope of the present application, can easily think of the change or the replacement, and should all cover
Within the protection scope of the application.Therefore, the protection scope of the application should be subject to the protection scope in claims.
Claims (13)
1. a kind of image processing method, which is characterized in that the described method includes:
Obtain the relevant form parameter of shape of image sampling frame；
According to the convolution nuclear parameter of the form parameter and convolution kernel, by each of at least two sampled points of the convolution kernel
Sampled point is adjusted from current first position to the second position corresponding with the form parameter, obtains convolution kernel adjusted,
Wherein, the position that any two sampled point at least two sampled point is presently in is not identical, any two sampled point
Position adjusted is not also identical；
Process of convolution is carried out according to the convolution collecting image adjusted.
2. image processing method according to claim 1, which is characterized in that the form parameter includes: that described image is adopted
Minification of the length and width and described image sample boxes of sample frame in down-sampling；It is described according to the form parameter
With the convolution nuclear parameter of convolution kernel, by each sampled point at least two sampled points of the convolution kernel from current first
It sets and adjusts to the second position corresponding with the form parameter, obtain convolution kernel adjusted, comprising:
It, will be at least two sampled points of the convolution kernel according to the length, the minification and the convolution nuclear parameter
The abscissa of each sampled point be adjusted to the second abscissa corresponding with the length from the first current abscissa；
According to the width, the minification and the convolution nuclear parameter, by the ordinate of each sampled point from current
One ordinate is adjusted to the second ordinate corresponding with the width, wherein the coordinate of each sampled point before adjustment is first
Abscissa and the first ordinate indicate that the position of each sampled point is located at current first position and adjusted each
The coordinate of sampled point is the second abscissa and the second ordinate, indicates that the position of each sampled point is located at the second position adjusted
Place；
Convolution kernel adjusted is obtained, the position of each sampled point is located at the second place in the convolution kernel adjusted.
3. image processing method according to claim 2, which is characterized in that it is described according to the length, it is described reduce times
The several and convolution nuclear parameter, the abscissa of each sampled point at least two sampled points by the convolution kernel is from working as
The first preceding abscissa is adjusted to the second abscissa corresponding with the length, comprising:
According to the length, the minification and the convolution nuclear parameter, at least two samplings of the convolution kernel are determined
First deviation of the abscissa of each sampled point in point；
According to the first deviation of the abscissa of each sampled point, by the abscissa of each sampled point from the first current abscissa
It is adjusted to the second abscissa corresponding with the length.
4. image processing method according to claim 3, which is characterized in that the convolution nuclear parameter includes: the convolution
Interval columns between total columns of core and the column of each sampled point and the central point of the convolution kernel；The basis
The length, the minification and convolution nuclear parameter determine the of the abscissa of each sampled point in the convolution kernel
One deviation, comprising:
According to the length W, the minification S, total columns M and the interval columns m, determine in the convolution kernel
Each sampled point abscissa the first deviation be W*m/M*S.
5. image processing method according to claim 2, which is characterized in that it is described according to the width, it is described reduce times
The several and convolution nuclear parameter, the ordinate by each sampled point are adjusted to and the width from the first current ordinate
Corresponding second ordinate, comprising:
According to the width, the minification and the convolution nuclear parameter, the second of the ordinate of each sampled point is determined
Deviation；
According to the second deviation of the ordinate of each sampled point, by the ordinate of each sampled point from the first current ordinate
It is adjusted to the second ordinate corresponding with the length.
6. image processing method according to claim 5, which is characterized in that the convolution nuclear parameter further include: the volume
Total line number of product core and the interval line number of each sampled point being expert between the central point of the convolution kernel；Institute
It states according to the width, the minification and the convolution nuclear parameter, determines that the second of the ordinate of each sampled point is inclined
Difference, including；
According to the width H, the minification S, total line number N and the interval line number n, each sampled point is determined
Second deviation H*n/N*S of ordinate.
7. image processing method described in -6 any claims according to claim 1, which is characterized in that in the acquisition image sampling
Before the shape of frame, the method also includes:
Obtain the first position parameter of the central point of described image sample boxes；
According to the second position parameter of the first position parameter and the central point of the convolution kernel, by each sampled point from current
The third place adjust to first position, wherein each sampled point be in first position indicate the central point of the convolution kernel with
The central point of described image sample boxes is overlapped.
8. image processing method according to claim 7, which is characterized in that the first position parameter includes: in first
Heart abscissa and the first center ordinate, the second position parameter include: the second center abscissa and the second center ordinate,
The second position parameter according to the first position parameter and the central point of the convolution kernel, by each sampled point from current
The third place adjust to first position, comprising:
According to first center abscissa and second center abscissa, by the abscissa of each sampled point from current
Three abscissas are adjusted to the first abscissa；
According to first center ordinate and second center ordinate, by the ordinate of each sampled point from current
Three ordinates are adjusted to the first ordinate, wherein the coordinate of each sampled point is that third abscissa and third ordinate indicate every
The position of a sampled point is located at current the third place.
9. image processing method according to claim 8, which is characterized in that it is described according to first center abscissa and
The abscissa of each sampled point is adjusted to the first abscissa from current third abscissa by second center abscissa, packet
It includes:
According to first center abscissa and second center abscissa, determine first center abscissa with it is described
Third deviation between second center abscissa；
According to the third deviation, the abscissa of each sampled point is adjusted to the first horizontal seat from current third abscissa
Mark.
10. image processing method according to claim 8, which is characterized in that described according to first center ordinate
With second center ordinate, the ordinate of each sampled point is adjusted to the first ordinate from current third ordinate,
Include:
According to first center ordinate and second center ordinate, determine first center ordinate with it is described
The 4th deviation between second center ordinate；
According to the 4th deviation, the ordinate of each sampled point is adjusted to the first vertical seat from current third ordinate
Mark.
11. a kind of image processing apparatus, which is characterized in that described device includes:
Shape obtains module, the relevant form parameter of shape for obtaining image sampling frame；
Adjusting Shape module, for the convolution nuclear parameter according to the form parameter and convolution kernel, at least by the convolution kernel
Each sampled point in two sampled points is adjusted from current first position to the second position corresponding with the form parameter, is obtained
Convolution kernel adjusted, wherein the position that any two sampled point at least two sampled point is presently in not phase
Together, any two sampled point position adjusted is not also identical；
Convolution module, for carrying out process of convolution according to the convolution collecting image adjusted.
12. a kind of electronic equipment, which is characterized in that the electronic equipment includes: processor, memory, bus and communication interface；
The processor, the communication interface and memory are connected by the bus；
The memory, for storing program；
The processor, for the program by calling storage in the memory, to execute as claim 1-10 is any
Image processing method described in claim.
13. a kind of computer-readable storage media for the non-volatile program code that can be performed with computer, which is characterized in that institute
Stating program code makes the computer execute the image processing method as described in any claim of claim 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811071355.4A CN109325945B (en) | 2018-09-13 | 2018-09-13 | Image processing method, image processing device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811071355.4A CN109325945B (en) | 2018-09-13 | 2018-09-13 | Image processing method, image processing device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109325945A true CN109325945A (en) | 2019-02-12 |
| CN109325945B CN109325945B (en) | 2021-01-08 |
Family
ID=65265372
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811071355.4A Active CN109325945B (en) | 2018-09-13 | 2018-09-13 | Image processing method, image processing device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109325945B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112308105A (en) * | 2019-08-02 | 2021-02-02 | 北京图森智途科技有限公司 | Target detection method, target detector and related equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550750A (en) * | 2015-12-21 | 2016-05-04 | 长沙网动网络科技有限公司 | Method for improving identification precision of convolutional neural network |
| US9665802B2 (en) * | 2014-11-13 | 2017-05-30 | Nec Corporation | Object-centric fine-grained image classification |
| WO2017155691A1 (en) * | 2016-03-11 | 2017-09-14 | Nec Laboratories America, Inc. | Deep deformation network for object landmark localization |
| CN107292319A (en) * | 2017-08-04 | 2017-10-24 | 广东工业大学 | The method and device that a kind of characteristic image based on deformable convolutional layer is extracted |
| CN107609541A (en) * | 2017-10-17 | 2018-01-19 | 哈尔滨理工大学 | A kind of estimation method of human posture based on deformable convolutional neural networks |
| CN108510467A (en) * | 2018-03-28 | 2018-09-07 | 西安电子科技大学 | SAR image target recognition method based on variable depth shape convolutional neural networks |
-
2018
- 2018-09-13 CN CN201811071355.4A patent/CN109325945B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9665802B2 (en) * | 2014-11-13 | 2017-05-30 | Nec Corporation | Object-centric fine-grained image classification |
| CN105550750A (en) * | 2015-12-21 | 2016-05-04 | 长沙网动网络科技有限公司 | Method for improving identification precision of convolutional neural network |
| WO2017155691A1 (en) * | 2016-03-11 | 2017-09-14 | Nec Laboratories America, Inc. | Deep deformation network for object landmark localization |
| CN107292319A (en) * | 2017-08-04 | 2017-10-24 | 广东工业大学 | The method and device that a kind of characteristic image based on deformable convolutional layer is extracted |
| CN107609541A (en) * | 2017-10-17 | 2018-01-19 | 哈尔滨理工大学 | A kind of estimation method of human posture based on deformable convolutional neural networks |
| CN108510467A (en) * | 2018-03-28 | 2018-09-07 | 西安电子科技大学 | SAR image target recognition method based on variable depth shape convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| JIFENG DAI，ET AL.: "Deformable Convolutional Networks", 《THE IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV)》 * |
| 欧阳针 等: "基于可变形卷积神经网络的图像分类研究", 《软件导刊》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112308105A (en) * | 2019-08-02 | 2021-02-02 | 北京图森智途科技有限公司 | Target detection method, target detector and related equipment |
| CN112308105B (en) * | 2019-08-02 | 2024-04-12 | 北京图森智途科技有限公司 | Target detection method, target detector and related equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109325945B (en) | 2021-01-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104217202B (en) | Information identifying method, equipment and system | |
| CN109977935A (en) | A kind of text recognition method and device | |
| CN109598272A (en) | Recognition methods, device, equipment and the medium of character row image | |
| CN109285136A (en) | A kind of Multiscale Fusion method, apparatus, storage medium and the terminal of image | |
| CN110211119A (en) | Image quality measure method, apparatus, electronic equipment and readable storage medium storing program for executing | |
| CN108989804A (en) | The coding method of image and device | |
| CN110784644B (en) | Image processing method and device | |
| CN109857984A (en) | A kind of homing method and device of boiler load factor-efficacy curve | |
| CN107941202A (en) | A kind of instant inspection method of aeroplane photography flight reappearance and device | |
| CN105975132B (en) | A kind of detection method and device of touch screen accuracy | |
| CN109615171A (en) | Characteristic threshold value determines that method and device, problem objects determine method and device | |
| CN109325945A (en) | Image processing method, device, electronic equipment and storage medium | |
| CN113516697B (en) | Image registration method, device, electronic equipment and computer readable storage medium | |
| CN109242868A (en) | Image processing method, device, electronic equipment and storage medium | |
| CN113034582A (en) | Pose optimization device and method, electronic device and computer readable storage medium | |
| CN107085843A (en) | System and method for estimating the modulation transfer function in optical system | |
| CN116012354A (en) | Method, device, equipment and storage medium for detecting silk-screen defect of chip capacitor | |
| CN112465917B (en) | Distortion calibration method, system, equipment and storage medium of lens module | |
| CN113177988B (en) | Spherical screen camera and laser calibration method, device, equipment and storage medium | |
| CN116012242A (en) | Camera distortion correction effect evaluation method, device, medium and equipment | |
| CN110971895B (en) | Video jitter detection method and device | |
| CN113051778A (en) | Clothing design method and device, electronic equipment and storage medium | |
| CN114241044A (en) | Loop detection method, device, electronic equipment and computer readable medium | |
| CN113438468A (en) | Dynamic control method and device for video quality, storage medium and electronic equipment | |
| CN111199514B (en) | Image background blurring method, device and equipment and readable storage medium |
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 |