CN109886279A - Image processing method, device, computer equipment and storage medium - Google Patents
Image processing method, device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN109886279A CN109886279A CN201910067366.3A CN201910067366A CN109886279A CN 109886279 A CN109886279 A CN 109886279A CN 201910067366 A CN201910067366 A CN 201910067366A CN 109886279 A CN109886279 A CN 109886279A
- Authority
- CN
- China
- Prior art keywords
- pixel
- sampling area
- candidate region
- characteristic pattern
- pixel value
- 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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image processing method, device, computer equipment and storage mediums, so that the calculated for pixel values result of the fixed-size characteristic pattern finally obtained is more accurate.Method part includes: to obtain candidate region characteristic pattern;Candidate region characteristic pattern is divided into the equivalent zonule of NxM size according to default candidate region pond parameter;Each zonule is averagely divided into P sampling area according to default sampling number P;Determine the intersection pixel intersected in the characteristic pattern of candidate region with sampling area;According to intersection pixel, the pixel value of the center position of sampling area is determined;According to the pixel value of the center position of the corresponding each sampling area in zonule, the corresponding pixel value for determining each zonule;Fixed-size candidate region characteristic pattern is obtained according to the pixel value of each zonule.
Description
Technical field
The present invention relates to field of image processing more particularly to a kind of image processing method, device, computer equipment and storages
Medium.
Background technique
In field of image processing, it usually needs detect and analyze certain block region, referred to as target detection.In object detection field
Numerous frames in, such as Fast-RCNN, Faster-RCNN, RFCN, ROI Pooling effect be according to pre-selection frame
Corresponding region pond is turned to fixed-size spy in characteristic pattern by Region Proposal namely the position coordinates of candidate region
Sign figure, to carry out subsequent sort operation.But the location information of Region Proposal is returned by model,
Obtained location information is usually the form of floating number, and the characteristic pattern desired size of pondization operation is fixed.Therefore by corresponding region
Pond turns to this process of fixed-size characteristic pattern and there is the operation for being rounded quantization twice: (1) by the boundary Region Proposal
Rounding is quantified as integral point coordinate value.(2) borderline region being rounded after quantifying averagely is divided into N xN unit, to each
The boundary of unit carries out rounding quantization again.
However pass through and be rounded quantization twice, obtained candidate frame and most start to return the position come out have it is certain inclined
Difference, this deviation will affect the accuracy of detection or segmentation, referred to as region mismatch problem (misalignment).Traditional
It is that above-mentioned integral quantization has been taken to operate in scheme, obtains the pixel that coordinate is floating number using the method for bilinear interpolation
Image values on point, to convert a continuous operation for entire feature accumulation process, specifically: first passing through bilinearity
Interpolation algorithm calculates one " red crunode ", and according to the fixed position in the side of red crunode, such as unifies the picture where red crunode
The lower-left of vegetarian refreshments, bottom right pixel point calculate the position of the pixel where red crunode, however, the fixed position institute in the side of red crunode
The pixel of pixel be not necessarily the pixel close to the pixel where red crunode, it is thus possible to will lead to and calculate
As a result there is a certain error.
Summary of the invention
The present invention provides a kind of image processing method, device, computer equipment and storage mediums, so that finally obtain
The calculated for pixel values result of fixed-size characteristic pattern is more accurate.
A kind of image processing method, comprising:
Candidate region characteristic pattern is obtained, the candidate region characteristic pattern is acquired after candidate region to be mapped to characteristic pattern
, the characteristic pattern is obtained by carrying out feature extraction to input picture by convolutional neural networks, and the candidate region is root
The input picture is carried out obtained by the detection of target area according to goal-selling detection algorithm;
The candidate region characteristic pattern is divided into the equivalent cell of NxM size according to default candidate region pond parameter
Domain, for positive integer and more than or equal to 1, the default candidate region pond parameter includes the width for pondization processing by described N, M
Spend parameter and length parameter;
Each zonule is averagely divided into P sampling area according to default sampling number P, the P is positive integer
And it is greater than or equal to 2;
Determine the intersection pixel intersected in the candidate region characteristic pattern with the sampling area;
According to the intersection pixel intersected in the candidate region characteristic pattern with the sampling area, the sampling is determined
The pixel value of the center position in region;
It is corresponding to determine often according to the pixel value of the center position of the corresponding each sampling area in the zonule
The pixel value of a zonule;
The fixed-size candidate region characteristic pattern is obtained according to the pixel value of each zonule.
A kind of image processing apparatus, comprising:
First obtains module, and for obtaining candidate region characteristic pattern, the candidate region characteristic pattern is to reflect candidate region
It is incident upon obtained after characteristic pattern, the characteristic pattern is to be carried out obtained by feature extraction as convolutional neural networks to input picture
It arrives, the candidate region is obtained by carrying out target area detection to the input picture according to goal-selling detection algorithm;
First division module, for being obtained described in module acquisition according to default candidate region pond parameter by described first
Candidate region characteristic pattern is divided into the equivalent zonule of NxM size, and described N, M are described pre- for positive integer and more than or equal to 1
If candidate region pond parameter includes the width parameter and length parameter for pondization processing;
Second division module, the institute for dividing each first division module according to sampling number P is preset
It states zonule and is averagely divided into P sampling area, the P is positive integer and is greater than or equal to 2;
First determining module, for determine it is described first obtain module obtain the candidate region characteristic pattern in institute
State the intersection pixel for the sampling area intersection that the second division module divides;
Second determining module, for according to first determining module determine the candidate region characteristic pattern in it is described
The intersection pixel of sampling area intersection, determines the pixel value of the center position of the sampling area;
Third determining module, the zonule for being determined according to second determining module is corresponding each described to adopt
The pixel value of the center position in sample region, the corresponding pixel value for determining each zonule;
Second obtains module, for being obtained according to the pixel value of each of the third determining module determination zonule
The fixed-size candidate region characteristic pattern.
A kind of computer equipment, including memory, processor and storage are in the memory and can be in the processing
The computer program run on device, the processor realize above-mentioned image processing method when executing the computer program.It is a kind of
Computer readable storage medium, the computer-readable recording medium storage have computer program, and the computer program is located
Reason device realizes above-mentioned image processing method when executing.
In the scheme that above-mentioned image processing method, device, computer equipment and storage medium are realized, according to sampling area
It with the overlapping relation of pixel, is effectively guaranteed and corresponding candidate pool area is turned into fixed-size characteristic pattern, and due to being
It being calculated according to intersection pixel, the calculated result of the pixel value of final sampling area is more accurate, so that finally obtain
The calculated for pixel values result of fixed-size characteristic pattern is also relatively more accurate.
Detailed description of the invention
It, below will be attached needed in the description of this invention in order to illustrate more clearly of technical solution of the present invention
Figure is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this field
For those of ordinary skill, without any creative labor, it can also be obtained according to these attached drawings other attached
Figure.
Fig. 1 is one schematic diagram of system framework figure applied by image processing method in the present invention;
Fig. 2 is image processing method one embodiment flow diagram in the present invention;
Fig. 3 is a schematic diagram of the sampling area of candidate region characteristic pattern in the present invention;
Fig. 4 is another embodiment flow diagram of image processing method in the present invention;
Fig. 5 is another schematic diagram of the sampling area of candidate region characteristic pattern in the present invention;
Fig. 6 is another schematic diagram of the sampling area of candidate region characteristic pattern in the present invention;
Fig. 7 is another embodiment flow diagram of image processing method in the present invention;
Fig. 8 is another embodiment flow diagram of image processing method in the present invention;
Fig. 9 is another embodiment flow diagram of image processing method in the present invention;
Figure 10 is another schematic diagram of the sampling area of candidate region characteristic pattern in the present invention;
Figure 11 is another embodiment flow diagram of image processing method in the present invention;
Figure 12 is image processing apparatus one embodiment structural schematic diagram in the present invention;
Figure 13 is one embodiment structural schematic diagram of computer equipment in the present invention.
Specific embodiment
Below in conjunction with the attached drawing in the present invention, the technical solution in the present invention is clearly and completely described, is shown
So, described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on the implementation in the present invention
Example, every other embodiment obtained by those of ordinary skill in the art without making creative efforts belong to
The scope of protection of the invention.
The embodiment of the present invention provides image processing method, can be applicable in the system framework schematic diagram such as Fig. 1, server can
Processing mode provided by this image processing method is executed to input picture, to obtain fixed-size feature after being handled
Figure, wherein server can realize with the server cluster of independent server either multiple servers composition, below it is right
The embodiment of the present invention describes in detail:
In one embodiment, as shown in Fig. 2, providing a kind of image processing method, include the following steps:
S10: obtaining candidate region characteristic pattern, and candidate region is is mapped to institute after characteristic pattern by the candidate region characteristic pattern
It obtains, the characteristic pattern is the candidate region obtained by carrying out feature extraction to input picture by convolutional neural networks
Obtained by carrying out target area detection to the input picture according to goal-selling detection algorithm.
The embodiment of the present invention is applied to convolutional neural networks (the Regions with Convolutional based on region
Neural Networks, RCNN) in, including but not limited to Fast RCNN, Faster RCNN and based on the complete of region
Convolutional network (Region-based Fully Convolutional Networks, RFCN).It is appreciated that being based on above-mentioned
It, can input picture progress convolutional layer, pond layer, area-of-interest pond layer (region of in the convolutional neural networks in region
Interest pooling layer, ROI pooling) and full articulamentum treatment process, and the embodiment of the present invention be
Obtain the candidate region (Region Proposal) of input picture and characteristic pattern (namely the feature of input picture
Maps after), pooling layers of ROI are handled.
Wherein, candidate region is exactly to find out the position namely region of interest that the middle target of input picture is likely to occur in advance
Domain, it is described by detecting the candidate region of input picture using characteristic informations such as texture, edge, colors in input picture
Candidate region is obtained by carrying out target area detection to the input picture according to goal-selling detection algorithm, wherein specific
Ground, in Fast RCNN, goal-selling detection algorithm is specifically to use selective search (selective search) directly right
Input picture carries out the extraction of candidate region, and in Faster RCNN, goal-selling detection algorithm refer to input picture into
After the feature extraction of row convolutional layer obtains characteristic pattern, using region candidate network (region proposal network,
PRN the extraction of candidate region) is carried out to characteristic pattern, no further details are given here for detailed process, but it is understood that, process is above-mentioned
The processing of goal-selling detection algorithm can get the corresponding candidate region of input picture.
Candidate region characteristic pattern refers to candidate region mapping to obtained figure after characteristic pattern corresponding to input picture
Picture, it will be understood that input picture mentions after the convolutional layer of convolutional neural networks of the input based on region by the feature of convolutional layer
Take processing, the corresponding characteristic pattern of available input object, the characteristic extraction procedure of specific convolutional layer is not also unfolded to go to live in the household of one's in-laws on getting married here
It states.In embodiments of the present invention, the corresponding candidate region characteristic pattern of input picture can be got, it should be noted that the present invention
The candidate region characteristic pattern of embodiment meaning refers to the corresponding each candidate region characteristic pattern of input picture, for the ease of retouching
It states, hereafter the image processing method that candidate region special zone figure proposes the embodiment of the present invention is described.
S20: it is equivalent that NxM size is divided into the candidate region characteristic pattern according to default candidate region pond parameter
Zonule, described N, M be positive integer and be greater than or equal to 1, the default candidate region pond parameter include width parameter and
Length parameter.
Wherein, presetting candidate region pond parameter is preset parameter in ROI pooling layers, above-mentioned default candidate region
Pond parameter is intended to turn in candidate region characteristic pattern pond the parameter of fixed-size characteristic pattern, and specific address includes width parameter
(pooled-h) and length parameter (pooled-w).
Illustratively, it is assumed that the size of input picture is 800*800, and the convolutional neural networks convolutional layer based on region uses
Be VGG16 network, feat-stride=32 (indicates that input picture picture after convolutional layer is handled is reduced into the 1/ of original image
32), namely after VGG16 network processes obtaining the corresponding characteristic pattern of input picture is 25*25, it is assumed that input picture has a time
Favored area, size 665*665, then after the candidate region maps to characteristic pattern, the size of obtained candidate region characteristic pattern are as follows:
665/32 ≈ 20.78, i.e. candidate region characteristic pattern are as follows: 20.78*20.78.Needing explanation is, in embodiments of the present invention,
In pooling layers for the treatment of process of ROI, for ease of description, usually retain two floating numbers of decimal point, when there is floating number
Retain 2 significant digits floating number to be illustrated.Assuming that the width parameter and length parameter of default candidate region pond parameter
It is respectively as follows: pooled-h=7, pooled-w=7, then the candidate region characteristic pattern is fixed into the spy of 7*7 size after treatment
Sign figure, in other words, is divided into 7*7=49 size etc. for the candidate region of the 20.78*20.78 mapped on characteristic pattern
Same zonule, the size of each zonule are 20.78/7 ≈ 2.97, that is to say, that each zonule is 2.97*2.97.
It should be noted that in embodiments of the present invention, according to default candidate region pond parameter to the candidate region
Characteristic pattern is divided into the equivalent zonule of NxM size, and the specific size of N, M are configured by practical application request, and default
Candidate region pond parameter is related, and the embodiment of the present invention is not specifically limited, and illustratively, above-mentioned N, M can also be respectively 8,
At this point, the candidate region characteristic pattern is fixed into the characteristic pattern of 8*8 size after treatment.
S30: each zonule is averagely divided by P sampling area according to default sampling number P, the P is positive
Integer and be greater than or equal to 2.
In the embodiment of the present invention, the candidate region characteristic pattern is divided into according to default candidate region pond parameter
After the equivalent zonule of NxM size, each zonule is handled as follows: according to default sampling number P by each institute
It states zonule and is averagely divided into P sampling area, the P is for positive integer and more than or equal to 2, the sampling area and the time
The shape type of the pixel of favored area characteristic pattern is identical.Wherein, which is for calculating each zonule
Pixel value set sampling number, illustratively, which can be 4,8 etc., and the specific present invention is implemented
Example is without limitation.
As described above, illustratively, the candidate region of the 20.78*20.78 mapped on characteristic pattern is divided into 7*
Behind the equivalent zonule of 7 sizes of ≈ 49, if default sampling number P is 4, by equivalent each of the zonule of 49 sizes
Zonule is averagely divided into 4 sampling areas.
S40: the intersection pixel intersected in the candidate region characteristic pattern with the sampling area is determined.
Passing through step 30, each zonule is averagely divided by P sampling area according to default sampling number P
Afterwards, the intersection pixel intersected in the candidate region characteristic pattern with the sampling area is determined.Namely passing through step S30
After processing obtains the corresponding each sampling area in each zonule, determines in the characteristic pattern of candidate region, adopted respectively with each
The intersection pixel of sample region intersection.It should be noted that specifically, the coordinate position and time of each sampling area can be passed through
The coordinate position of the pixel of favored area special zone figure determines the intersection pixel intersected with each sampling area.Illustratively,
It is appreciated that passing through step S40, the center position of corresponding 4 sampling areas in each zonule can be obtained, and small with this
The intersection pixel of the corresponding 4 sampling areas intersection in region.
Illustratively, as shown in Figure 3:
Wherein, region shown in Fig. 3 is a part of region in the characteristic pattern of candidate region, including 1-16 pixel,
Region where thick line box A, B, C and D is sampling area, and sampling area A, B, C and D constitute a zonule.With sampling
For the A of region, the center position of sampling area A can be determined by bilinearity difference interpolation algorithm, and determine to adopt with described
The intersection pixel of sample region A intersection, that is, pixel 1,2,5 and 6.Sampling area corresponding for each zonule,
The intersection pixel that can determine that center position and intersect with sampling area, for example, for sampling area B, with sample region
The target pixel points that domain B has intersection are 2,3,6 and 7.
S50: it according to the intersection pixel intersected in the candidate region characteristic pattern with the sampling area, determines described
The pixel value of the center position of sampling area.
It is determining the center position of the sampling area, and is determining to adopt in the candidate region characteristic pattern with described
After the intersection pixel of sample region intersection, according to the intersection picture intersected in the candidate region characteristic pattern with the sampling area
Vegetarian refreshments determines the pixel value of the center position of the sampling area.
As shown in figure 3, can have the intersection pixel 1,2,3 intersected according to sampling area A by taking sampling area A as an example
And 4 determine sampling area A center position pixel value.It is appreciated that being based on same calculation, can distinguish
Obtain the pixel value of the center position of the corresponding each sampling area in each zonule.
S60: corresponding according to the pixel value of the center position of the corresponding each sampling area in each zonule
Determine the pixel value of each zonule.
After step S50, it can get in candidate region, the pixel of the center position of the sampling area of each zonule
It is worth, it is right according to the pixel value of the center position of the corresponding each sampling area in the zonule in the embodiment of the present invention
It should determine the pixel value of each zonule.
As shown in figure 3, the pixel of the center position of zonule corresponding sampling area A, B, C and D can be obtained respectively
Value, correspondence is determined in Fig. 3, by the pixel value of the zonule constituted sampling area A, B, C and D.It is calculated based on same
Mode can get the corresponding pixel value in each zonule, different one illustrate here.
S70: the fixed-size candidate region characteristic pattern is obtained according to the pixel value of each zonule.
It is corresponding true in the pixel value according to the center position of the corresponding each sampling area in each zonule
It is fixed-size after being handled according to the pixel value of each zonule after the pixel value for making each zonule
The candidate region characteristic pattern.
As described above, illustratively, the candidate region of the 20.78*20.78 mapped on characteristic pattern is being divided into 7*
Behind the equivalent zonule of 7 sizes of ≈ 49, each zonule is 2.97*2.97, by preceding step S10-S70, acquisition it is every
The corresponding pixel value in a zonule, obtains the pixel value of 49 zonules, so that output obtains the candidate region characteristic pattern of 7*7,
Therefore, the exportable candidate region characteristic pattern for obtaining fixed-size 7*7, the fixed-size candidate region characteristic pattern can be used for
The subsequent classification of convolutional neural networks based on region and recurrence processing.As it can be seen that the embodiment of the invention provides at a kind of image
Reason method is effectively guaranteed corresponding candidate pool area turning to fixed ruler according to the overlapping relation of sampling area and pixel
Very little characteristic pattern, and by being calculated thus according to intersection pixel, the calculated result of the pixel value of final sampling area is more
Accurately, so that the calculated for pixel values result of the fixed-size characteristic pattern finally obtained is also relatively more accurate.
It should be noted that in conjunction with above-described embodiment, according to the pixel in sampling area and candidate region characteristic pattern
The size relation of size, the embodiment of the invention also provides with specific reference in the candidate region characteristic pattern with the sampling area
The target of intersection intersects pixel, determines the implementation of the pixel value of the center position of the sampling area, divides below
It is not introduced:
In one embodiment, as shown in figure 4, before step 50 namely it is described according in the candidate region characteristic pattern with
The target of sampling area intersection intersects pixel, before the pixel value for determining the center position of the sampling area,
The method also includes following steps:
S80: judge whether the size of the sampling area is greater than or equal to the pixel of the candidate region characteristic pattern
Size.
In step s 30, after each zonule being averagely divided into P sampling area according to default sampling number P,
Whether the sampling area for judging each zonule is greater than or equal to the size of the pixel in the characteristic pattern of candidate region, Ke Yili
Solution, since the size of each pixel of input picture is consistent, the candidate region characteristic pattern mapped it is every
The size of a pixel is also the same, and because the size for dividing each sampling area of obtained candidate region is equivalent, because
This may determine that whether any one sampling area is greater than or equal to the size of any one pixel of candidate region characteristic pattern.
S90: if judging, the size of the sampling area is greater than or equal to the size of the pixel of the candidate region image,
The corresponding auxiliary frame of the sampling area, the shape of the auxiliary frame and the time are then generated centered on the center position
The pixel of favored area characteristic pattern is identical, and the size of the auxiliary frame is then less than or equal to the pixel of the candidate region characteristic pattern
The size of point.
When judge sampling area size be greater than or equal to the candidate region image pixel size, then with described
The corresponding auxiliary frame of the sampling area is generated centered on center position, that is to say, that each of each zonule is adopted
Sample region can all generate corresponding auxiliary frame, and the shape of the auxiliary frame is identical as the pixel of the candidate region characteristic pattern,
The size of the auxiliary frame is then less than or equal to the size of the pixel of the candidate region characteristic pattern.It should be noted that this
The center position that bilinear interpolation determines each sampling area can be used in inventive embodiments, specifically calculate and each adopt
No further details are given here for the process of the center position in sample region.
Illustratively, referring to Fig. 5, as shown in Figure 5, the size of sampling area is greater than or equal to the candidate region image
Pixel size (being illustrated so that auxiliary frame and pixel are in the same size as an example in Fig. 5), by taking sampling area D as an example, this
In inventive embodiments, the corresponding auxiliary frame d of sampling area D can be generated with the center position of sampling area D, in Fig. 5
Shown in slash box, wherein the shape of the corresponding auxiliary frame d of sampling area D and the pixel of the candidate region characteristic pattern
Identical, the size of the corresponding auxiliary frame d of sampling area D is then less than or equal to the pixel of the candidate region characteristic pattern.
It should be noted that being only illustrated by taking sampling area D as an example here, for generating each cell in the characteristic pattern of candidate region
Each sampling area in domain produces corresponding auxiliary frame through the embodiment of the present invention.
It is described according to the intersection pixel intersected in the candidate region characteristic pattern with the sampling area in step S50,
The pixel value for determining the center position of the sampling area, specifically comprises the following steps:
S50`: according to the intersection pixel that intersects in the candidate region characteristic pattern with the sampling area and described
The corresponding auxiliary frame of sampling area, determines the pixel value of the center position of the sampling area.
In generating candidate region characteristic pattern after the auxiliary frame of each sampling area, according in the candidate region characteristic pattern
The intersection pixel and corresponding auxiliary frame intersected with the sampling area, determines the center position of the sampling area
Pixel value.
In order to make it easy to understand, referring again to Fig. 5, by taking sampling area D as an example, it is seen then that there is the phase intersected with sampling area D
Friendship pixel is pixel 6,7,10 and 11, and the corresponding auxiliary frame of sampling area D is auxiliary frame d, then according to pixel 6,7,
10,11 and auxiliary frame d determines the pixel value of the center position of sampling area D.
It should be noted that for other sampling areas of candidate region characteristic pattern, the center based on above-mentioned sampling area D
The calculated for pixel values mode of point position is determined, illustratively, referring to Fig. 6, for sampling area A, it is seen then that with sample region
The intersection pixel that domain A has intersection is pixel 1,2,5 and 6, and the corresponding auxiliary frame of sampling area A is auxiliary frame a, then root
The pixel value of the center position of sampling area a, other sampling areas are determined according to pixel 1,2,5,6 and auxiliary frame a
Central point pixel value calculation, do not illustrate one by one here.
In one embodiment, as shown in fig. 7, in step S50` namely it is described according in the candidate region characteristic pattern with
The intersection pixel and the corresponding auxiliary frame of the sampling area of the sampling area intersection, determine the sampling area
The pixel value of center position, specifically comprises the following steps:
S51`: phase of the corresponding auxiliary frame of the sampling area respectively with each pixel in the intersection pixel is obtained
Cross surface product.
S52`: according to the intersecting area of the corresponding auxiliary frame of the sampling area and each pixel and described
The pixel value of each pixel determines the corresponding first object pixel value of the sampling area.
S53`: the center position by the corresponding first object pixel value of the sampling area, as the sampling area
Pixel value.
For step S51`, please continue to refer to Fig. 5, by taking sampling area D as an example, it is seen that the corresponding auxiliary region sampling area D
Domain d respectively with intersect pixel namely pixel 6,7,10 and 11 and intersect, in this step, it may be determined that auxiliary frame d respectively with
The intersecting area for intersecting pixel 6,7,10 and 11, is denoted as: C respectively in the present embodiment6d、C7d、C10dAnd C11d。
For step S52`-S53`, equally by taking sampling area D as an example, C is being obtained6d、C7d、C10dAnd C11dLater, root
According to C6d、C7d、C10dAnd C11d, and intersect pixel 6,7,10 and 11 and determine the corresponding first object picture of sampling area D
Element value, and using the first object pixel value as the pixel value of the center position of sampling area D.It should be noted that for
The corresponding center point of sampling area can be obtained by step S51`-S53` in other sampling areas of candidate region characteristic pattern
The pixel value set does not repeat to repeat here one by one.
In one embodiment, as shown in figure 8, it is in step S53` namely described according to the corresponding auxiliary of the sampling area
Frame and the intersecting area of each pixel and the pixel value of each pixel, determine the sampling area pair
The first object pixel value answered, specifically comprises the following steps:
S531`: respectively correspond calculate each pixel pixel value and each pixel intersecting area it
Between product.
S532`: by the product phase between the pixel value of each pixel and the intersecting area of each pixel
Add, to obtain the first sum of products.
S533`: calculating sum of the intersecting area of each pixel, with obtain the first intersection pixel point areas with.
S534`: the first sum of products is calculated with described first and intersects quotient between pixel point areas, to obtain the sample region
The corresponding first object pixel value in domain.
Here by taking the sampling area D in the characteristic pattern of candidate region as an example, the embodiment of the present invention is illustrated:
For step S531`, C is being obtained6d、C7d、C10dAnd C11dLater, it respectively corresponds and calculates sampling area correspondence
Intersection pixel in each pixel pixel value and C6d、C7d、C10dAnd C11dBetween product, namely calculate separately out
C6dWith the pixel value A of pixel 66Product, be denoted as: A6C6d;Calculate C7dWith the pixel value A of pixel 77Product, be denoted as:
A7C7d;Calculate C10dWith the pixel value A of pixel 1010Product, be denoted as: A10C10d;Calculate C11dWith the picture of pixel 11
Plain value A11Product, be denoted as: A11C11d。
For step S532`, A is calculated6C6d、A7C7d、A10C10dAnd A11C11dLater, above-mentioned product is subjected to phase
Add to obtain the first sum of products namely A6C6d+A7C7d+A10C10d+A11C11d。
For step S533`, the intersecting area of the corresponding auxiliary frame d of sampling area D and each pixel and, the
One intersection pixel point areas and, that is, C6d+C7d+C10d+C11d。
For step S234`, corresponding first sum of products of sampling area D is calculated, intersects pixel point areas with described first
Quotient U between andD, it is specific as follows to state shown in formula:
It should be noted that being illustrated by taking sampling area D as an example in the embodiment of the present invention, for candidate region feature
Other sampling areas in figure, the corresponding first object pixel value of D for seeing above-mentioned sampling area obtain calculation progress
It calculates, does not repeat to repeat here.
It should be noted that according to the size relation of the size of the pixel in sampling area and candidate region characteristic pattern,
The embodiment of the invention also provides with specific reference to the target intersection intersected in the candidate region characteristic pattern with the sampling area
Pixel determines the implementation of the pixel value of the center position of the sampling area:
As shown in figure 9, after step S80 namely whether the size for judging the sampling area is greater than or equal to institute
After the size for stating the pixel of candidate region characteristic pattern, which further includes following steps:
S100: if judging, the size of the sampling area is less than the size of the pixel of the candidate region characteristic pattern,
According to the pixel value of each intersection pixel intersected in the candidate region characteristic pattern with the sampling area and described adopt
Sample region and each intersecting area for intersecting pixel, determine corresponding second target pixel value of the sampling area.
S110: using corresponding second target pixel value of the sampling area as the center position of the sampling area
Pixel value.
It has been observed that default sampled point is preset configuration, according to default sampling number P to each cell in candidate region
Domain marks off the size for the sampling area come, it is possible to less than the size of pixel in the characteristic pattern of candidate region, for example, candidate regions
When domain is Small object, mapping to the corresponding obtained candidate region characteristic pattern of input picture will be smaller, each of marks off and to adopt
The size in sample region is likely less than the size of the pixel of candidate region spy's figure.
Illustratively, as shown in Figure 10: as it can be seen that sampling area A, B, C and D that zonule marks off are less than pixel
Size, then according to the pixel value of each intersection pixel intersected in the candidate region characteristic pattern with the sampling area, with
And the sampling area and each intersecting area for intersecting pixel, determine corresponding second target of the sampling area
Pixel value.By taking sampling area B as an example, each intersection pixel intersected with sampling area B is respectively pixel 5 and 6, is remembered respectively
The pixel value of pixel 5 is A5, the pixel value of pixel 6 is A6, sampling area B and the intersecting area of pixel 5 and 6 are remembered respectively
For C5B、C6B, then according to A5、A6、C5BAnd C6BDetermine corresponding second target pixel value of sampling area B, and by sampling area
Pixel value of corresponding second target pixel value of B as the center position of sampling area B.Candidate region is less than for size
Other sampling areas of the pixel spot size of characteristic pattern can be based on aforesaid way, determine the heavy central point of sampling area respectively
The pixel value of position does not repeat to repeat here one by one.
In one embodiment, as shown in figure 11, in step S100, if judging, the size of the sampling area is less than the time
The size of the pixel of favored area characteristic pattern, it is described each according to intersecting in the candidate region characteristic pattern with the sampling area
The pixel value and the sampling area and each intersecting area for intersecting pixel of a intersection pixel, determine institute
Corresponding second target pixel value of sampling area is stated, is specifically comprised the following steps:
S101: the pixel value for calculating each intersection pixel is respectively corresponded, with each phase for intersecting pixel
Product between cross surface product.
S102: by the pixel value of each intersection pixel, between each intersecting area for intersecting pixel
Product addition, to obtain the second sum of products.
S103: calculating the sum between the intersecting area of each intersection pixel, to obtain the second intersection pixel face
Product and.
S104: the second sum of products is calculated with described second and intersects quotient between pixel point areas, to obtain the sample region
Corresponding second target pixel value in domain.
For step S101-S104, please continue to refer to Figure 10, by taking the sampling area B in the characteristic pattern of candidate region as an example,
Obtaining is C5B、C6BLater, the pixel value and C for calculating each pixel in intersection pixel are respectively corresponded5B、C6BBetween multiply
Product, namely calculate separately out C5BWith the pixel value A of pixel 55Product, be denoted as: A5C5B;Calculate C6BWith the picture of pixel 6
Plain value A6Product, be denoted as: A6C6B。
By taking the sampling area B in the characteristic pattern of candidate region as an example, C is calculated in acquisition5BWith the pixel value A of pixel 55's
Product A5C5B, and calculate C6BWith the pixel value A of pixel 66Product A6C6BLater, product is added to obtain second
Sum of products, namely: A5C5B+A6C6B。
Sum of the intersecting area of the corresponding intersection pixel of sampling area B, i.e., the first intersection pixel point areas and are as follows: C5B+
C6B。
Second sum of products intersects quotient between pixel point areas with described second, and the specific following formula that samples calculates:
It should be noted that being illustrated by taking sampling area B as an example in the embodiment of the present invention, for candidate region feature
Other sampling areas in figure, corresponding second target pixel value of B for seeing above-mentioned sampling area obtain calculation progress
It calculates, does not repeat to repeat here.
It should be understood that the size of the serial number of each step is not meant that the order of the execution order in above-described embodiment, each process
Execution sequence should be determined by its function and internal logic, the implementation process without coping with the embodiment of the present invention constitutes any limit
It is fixed.
In one embodiment, a kind of image processing apparatus is provided, in the image processing apparatus and above-described embodiment at image
Reason method corresponds.As shown in figure 12, which includes the first acquisition module 101, the first division module
102, the second division module 103, the first determining module 104, the second determining module 105, third determining module 106 and second obtain
Module 107.Detailed description are as follows for each functional module:
First obtains module, and for obtaining candidate region characteristic pattern, the candidate region characteristic pattern is to reflect candidate region
It is incident upon obtained after characteristic pattern, the characteristic pattern is to be carried out obtained by feature extraction as convolutional neural networks to input picture
It arrives, the candidate region is obtained by carrying out target area detection to the input picture according to goal-selling detection algorithm;
First division module, for being obtained described in module acquisition according to default candidate region pond parameter by described first
Candidate region characteristic pattern is divided into the equivalent zonule of NxM size, and described N, M are described pre- for positive integer and more than or equal to 1
If candidate region pond parameter includes the width parameter and length parameter for pondization processing;
Second division module, the institute for dividing each first division module according to sampling number P is preset
It states zonule and is averagely divided into P sampling area, the P is positive integer and is greater than or equal to 2;
First determining module, for determine it is described first obtain module obtain the candidate region characteristic pattern in institute
State the intersection pixel for the sampling area intersection that the second division module divides;
Second determining module, for according to first determining module determine the candidate region characteristic pattern in it is described
The intersection pixel of sampling area intersection, determines the pixel value of the center position of the sampling area;
Third determining module, the zonule for being determined according to second determining module is corresponding each described to adopt
The pixel value of the center position in sample region, the corresponding pixel value for determining each zonule;
Second obtains module, for being obtained according to the pixel value of each of the third determining module determination zonule
The fixed-size candidate region characteristic pattern.
In one embodiment, described image processing unit further includes the 4th determining module, generation module;
4th determining module is adopted according in the candidate region characteristic pattern with described for second determining module
The intersection pixel of sample region intersection, before the pixel value for determining the center position of the sampling area, is adopted described in judgement
Whether the size in sample region is greater than or equal to the size of the pixel of the candidate region characteristic pattern;
The generation module, if judging that the size of the sampling area is greater than or equal to institute for the 4th determining module
The size of the pixel of candidate region image is stated, then generates the sample region centered on the center position of the sampling area
The corresponding auxiliary frame in domain, the shape of the auxiliary frame is identical as the pixel of the candidate region characteristic pattern, the auxiliary frame
Size is then less than or equal to the size of the pixel of the candidate region characteristic pattern;
Second determining module, is specifically used for: intersecting according in the candidate region characteristic pattern with the sampling area
Intersection pixel and the corresponding auxiliary frame of the sampling area, determine the pixel of the center position of the sampling area
Value.
In one embodiment, second determining module be used for according in the candidate region characteristic pattern with the sample region
The intersection pixel and the corresponding auxiliary frame of the sampling area of domain intersection, determine the center position of the sampling area
Pixel value, comprising:
Second determining module is used for:
Obtain intersection of the corresponding auxiliary frame of the sampling area respectively with each pixel in the intersection pixel
Product;
According to the intersecting area and each picture of the corresponding auxiliary frame of the sampling area and each pixel
The pixel value of vegetarian refreshments determines the corresponding first object pixel value of the sampling area;
By the corresponding first object pixel value of the sampling area, the pixel of the center position as the sampling area
Value.
In one embodiment, second determining module is for described according to the corresponding auxiliary frame of the sampling area and institute
The intersecting area of each pixel and the pixel value of each pixel are stated, determines the sampling area corresponding
One target pixel value, comprising:
Second determining module is specifically used for:
Respectively correspond multiplying between the intersecting area of the pixel value and each pixel that calculate each pixel
Product;
By the product addition between the pixel value of each pixel and the intersecting area of each pixel, to obtain
To the first sum of products;
Calculate the sum of the intersecting area of each pixel, with obtain the first intersection pixel point areas and;
It calculates the first sum of products and intersects quotient between pixel point areas with described first, it is corresponding to obtain the sampling area
First object pixel value.
In one embodiment, described image processing unit further includes the 5th determining module:
5th determining module: whether the size for judging the sampling area is greater than or equal to the candidate region
After the size of the pixel of characteristic pattern, if judging, the size of the sampling area is less than the pixel of the candidate region characteristic pattern
The size of point, then according to the pixel of each intersection pixel intersected in the candidate region characteristic pattern with the sampling area
Value and the sampling area and each intersecting area for intersecting pixel determine the sampling area corresponding the
Two target pixel values;
Using corresponding second target pixel value of the sampling area as the pixel of the center position of the sampling area
Value.
In one embodiment, the 5th determining module for it is described according in the candidate region characteristic pattern with the sample region
The pixel value and the sampling area and each intersection for intersecting pixel of each intersection pixel of domain intersection
Product, determines corresponding second target pixel value of the sampling area, comprising:
5th determining module, is specifically used for:
The pixel value for calculating each target intersection pixel is respectively corresponded, intersects pixel with each target
Product between intersecting area;
By the pixel value of each intersection pixel, with the product between each intersecting area for intersecting pixel
It is added, to obtain the second sum of products;
Calculate it is described it is each intersection pixel intersecting area between sum, with obtain second intersection pixel point areas with;
It calculates the second sum of products and intersects quotient between pixel point areas with described second, it is corresponding to obtain the sampling area
The second target pixel value.
Specific about image processing apparatus limits the restriction that may refer to above for image processing method, herein not
It repeats again.Modules in above-mentioned image processing apparatus can be realized fully or partially through software, hardware and combinations thereof.On
Stating each module can be embedded in the form of hardware or independently of in the processor in computer equipment, can also store in a software form
In memory in computer equipment, the corresponding operation of the above modules is executed in order to which processor calls.
In one embodiment, a kind of computer equipment is provided, which can be server, internal junction
Composition can be as shown in figure 13.The computer equipment include by system bus connect processor, memory, network interface and
Database.Wherein, the processor of the computer equipment is for providing calculating and control ability.The memory packet of the computer equipment
Include non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system, computer program and data
Library.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.The calculating
The database of machine equipment is for storing characteristic pattern etc..The network interface of the computer equipment is used to pass through network with external terminal
Connection communication.To realize a kind of image processing method when the computer program is executed by processor.
In one embodiment, a kind of computer equipment is provided, including memory, processor and storage are on a memory
And the computer program that can be run on a processor, processor perform the steps of when executing computer program
Candidate region characteristic pattern is obtained, the candidate region characteristic pattern is acquired after candidate region to be mapped to characteristic pattern
, the characteristic pattern is obtained by carrying out feature extraction to input picture by convolutional neural networks, and the candidate region is root
The input picture is carried out obtained by the detection of target area according to goal-selling detection algorithm;
The candidate region characteristic pattern is divided into the equivalent cell of NxM size according to default candidate region pond parameter
Domain, for positive integer and more than or equal to 1, the default candidate region pond parameter includes the width for pondization processing by described N, M
Spend parameter and length parameter;
Each zonule is averagely divided into P sampling area according to default sampling number P, the P is positive integer
And it is greater than or equal to 2;
Determine the intersection pixel intersected in the candidate region characteristic pattern with the sampling area;
According to the intersection pixel intersected in the candidate region characteristic pattern with the sampling area, the sampling is determined
The pixel value of the center position in region;
It is corresponding to determine often according to the pixel value of the center position of the corresponding each sampling area in the zonule
The pixel value of a zonule;
The fixed-size candidate region characteristic pattern is obtained according to the pixel value of each zonule.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
Machine program performs the steps of when being executed by processor
Candidate region characteristic pattern is obtained, the candidate region characteristic pattern is acquired after candidate region to be mapped to characteristic pattern
, the characteristic pattern is obtained by carrying out feature extraction to input picture by convolutional neural networks, and the candidate region is root
The input picture is carried out obtained by the detection of target area according to goal-selling detection algorithm;
The candidate region characteristic pattern is divided into the equivalent cell of NxM size according to default candidate region pond parameter
Domain, for positive integer and more than or equal to 1, the default candidate region pond parameter includes the width for pondization processing by described N, M
Spend parameter and length parameter;
Each zonule is averagely divided into P sampling area according to default sampling number P, the P is positive integer
And it is greater than or equal to 2;
Determine the intersection pixel intersected in the candidate region characteristic pattern with the sampling area;
According to the intersection pixel intersected in the candidate region characteristic pattern with the sampling area, the sampling is determined
The pixel value of the center position in region;
It is corresponding to determine often according to the pixel value of the center position of the corresponding each sampling area in the zonule
The pixel value of a zonule;
The fixed-size candidate region characteristic pattern is obtained according to the pixel value of each zonule.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein,
To any reference of memory, storage, database or other media used in each embodiment provided by the present invention,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of described device is divided into different functional unit or module, more than completing
The all or part of function of description.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations;Although referring to aforementioned reality
Applying example, invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each
Technical solution documented by embodiment is modified or equivalent replacement of some of the technical features;And these are modified
Or replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution should all
It is included within protection scope of the present invention.
Claims (10)
1. a kind of image processing method characterized by comprising
Obtain candidate region characteristic pattern, the candidate region characteristic pattern be candidate region is mapped to it is obtained after characteristic pattern,
The characteristic pattern is obtained by carrying out feature extraction to input picture by convolutional neural networks, and the candidate region is according to pre-
If algorithm of target detection carries out obtained by the detection of target area the input picture;
The candidate region characteristic pattern is divided into the equivalent zonule of NxM size according to default candidate region pond parameter,
For positive integer and more than or equal to 1, the default candidate region pond parameter includes the width ginseng for pondization processing by described N, M
Several and length parameter;
Each zonule is averagely divided into P sampling area according to default sampling number P, the P is positive integer and big
In or equal to 2;
Determine the intersection pixel intersected in the candidate region characteristic pattern with the sampling area;
According to the intersection pixel intersected in the candidate region characteristic pattern with the sampling area, the sampling area is determined
Center position pixel value;
According to the pixel value of the center position of the corresponding each sampling area in the zonule, correspondence determines each institute
State the pixel value of zonule;
The fixed-size candidate region characteristic pattern is obtained according to the pixel value of each zonule.
2. image processing method as described in claim 1, which is characterized in that it is described according in the candidate region characteristic pattern with
The intersection pixel of sampling area intersection, it is described before the pixel value for determining the center position of the sampling area
Method further include:
Judge whether the size of the sampling area is greater than or equal to the size of the pixel of the candidate region characteristic pattern;
If judging, the size of the sampling area is greater than or equal to the size of the pixel of the candidate region image, with described
Generate the corresponding auxiliary frame of the sampling area centered on the center position of sampling area, the shape of the auxiliary frame with it is described
The pixel of candidate region characteristic pattern is identical, and the size of the auxiliary frame is then less than or equal to the picture of the candidate region characteristic pattern
The size of vegetarian refreshments;
It is described according to the intersection pixel intersected in the candidate region characteristic pattern with the sampling area, determine the sampling
The pixel value of the center position in region, comprising:
According to the intersection pixel and the sampling area pair intersected in the candidate region characteristic pattern with the sampling area
The auxiliary frame answered determines the pixel value of the center position of the sampling area.
3. image processing method as claimed in claim 2, which is characterized in that it is described according in the candidate region characteristic pattern with
The intersection pixel and the corresponding auxiliary frame of the sampling area of the sampling area intersection, determine the sampling area
The pixel value of center position, comprising:
Obtain intersecting area of the corresponding auxiliary frame of the sampling area respectively with each pixel in the intersection pixel;
According to the intersecting area and each pixel of the corresponding auxiliary frame of the sampling area and each pixel
Pixel value, determine the corresponding first object pixel value of the sampling area;
By the corresponding first object pixel value of the sampling area, the pixel value of the center position as the sampling area.
4. image processing method as claimed in claim 3, which is characterized in that described according to the corresponding auxiliary of the sampling area
Frame and the intersecting area of each pixel and the pixel value of each pixel, determine the sampling area pair
The first object pixel value answered, comprising:
Respectively correspond the product between the intersecting area of the pixel value and each pixel that calculate each pixel;
By the product addition between the pixel value of each pixel and the intersecting area of each pixel, to obtain
One sum of products;
Calculate the sum of the intersecting area of each pixel, with obtain the first intersection pixel point areas and;
It calculates the first sum of products and intersects quotient between pixel point areas with described first, to obtain the sampling area corresponding
One target pixel value.
5. image processing method as claimed in claim 2, which is characterized in that whether the size for judging the sampling area
After size more than or equal to the pixel of the candidate region characteristic pattern, the method also includes:
If judging, the size of the sampling area is less than the size of the pixel of the candidate region characteristic pattern, according to the time
The pixel value of each intersection pixel intersected in favored area characteristic pattern with the sampling area and the sampling area and institute
The intersecting area for stating each intersection pixel, determines corresponding second target pixel value of the sampling area;
Using corresponding second target pixel value of the sampling area as the pixel value of the center position of the sampling area.
6. image processing method as claimed in claim 5, which is characterized in that it is described according in the candidate region characteristic pattern with
The pixel value of each intersection pixel of the sampling area intersection and the sampling area and each intersection pixel
Intersecting area, determine corresponding second target pixel value of the sampling area, comprising:
The pixel value for calculating each target intersection pixel is respectively corresponded, the intersection of pixel is intersected with each target
Product between area;
By the pixel value of each intersection pixel, with the product phase between each intersecting area for intersecting pixel
Add, to obtain the second sum of products;
Calculate it is described it is each intersection pixel intersecting area between sum, with obtain second intersection pixel point areas with;
It calculates the second sum of products and intersects quotient between pixel point areas with described second, to obtain the sampling area corresponding
Two target pixel values.
7. a kind of image processing apparatus characterized by comprising
First obtains module, and for obtaining candidate region characteristic pattern, the candidate region characteristic pattern is to map to candidate region
Obtained after characteristic pattern, the characteristic pattern is institute obtained by carrying out feature extraction to input picture by convolutional neural networks
Stating candidate region is obtained by carrying out target area detection to the input picture according to goal-selling detection algorithm;
First division module, the candidate for obtaining module acquisition for described first according to candidate region pond parameter is preset
Provincial characteristics figure is divided into the equivalent zonule of NxM size, and described N, M are positive integer and are greater than or equal to 1, the default time
Favored area pond parameter includes the width parameter and length parameter for pondization processing;
Second division module is described small for being divided each first division module according to default sampling number P
Region is averagely divided into P sampling area, and the P is positive integer and is greater than or equal to 2;
First determining module, for determining that described first obtains in the candidate region characteristic pattern that module obtains with described the
The intersection pixel for the sampling area intersection that two division modules divide;
Second determining module, for according to first determining module determine the candidate region characteristic pattern in the sampling
The intersection pixel of region intersection, determines the pixel value of the center position of the sampling area;
Third determining module, the corresponding each sample region in the zonule for being determined according to second determining module
The pixel value of the center position in domain, the corresponding pixel value for determining each zonule;
Second obtains module, for being fixed according to the pixel value of each of the third determining module determination zonule
The candidate region characteristic pattern of size.
8. image processing apparatus as claimed in claim 7, which is characterized in that described image processing unit further includes the 4th determining
Module, generation module;
4th determining module, for second determining module according in the candidate region characteristic pattern with the sample region
The intersection pixel of domain intersection, before the pixel value for determining the center position of the sampling area, judges the sample region
Whether the size in domain is greater than or equal to the size of the pixel of the candidate region characteristic pattern;
The generation module, if judging that the size of the sampling area is greater than or equal to the time for the 4th determining module
The size of the pixel of constituency area image then generates the sampling area pair centered on the center position of the sampling area
The auxiliary frame answered, the shape of the auxiliary frame is identical as the pixel of the candidate region characteristic pattern, the size of the auxiliary frame
Then less than or equal to the size of the pixel of the candidate region characteristic pattern;
Second determining module, is specifically used for: according to the phase intersected in the candidate region characteristic pattern with the sampling area
Pixel and the corresponding auxiliary frame of the sampling area are handed over, determines the pixel value of the center position of the sampling area.
9. a kind of computer equipment, including memory, processor and storage are in the memory and can be in the processor
The computer program of upper operation, which is characterized in that the processor realized when executing the computer program as claim 1 to
6 described in any item image processing methods.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In realization such as image processing method as claimed in any one of claims 1 to 6 when the computer program is executed by processor.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910067366.3A CN109886279B (en) | 2019-01-24 | 2019-01-24 | Image processing method, device, computer equipment and storage medium |
PCT/CN2019/089196 WO2020151153A1 (en) | 2019-01-24 | 2019-05-30 | Image processing method and apparatus, and computer device and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910067366.3A CN109886279B (en) | 2019-01-24 | 2019-01-24 | Image processing method, device, computer equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109886279A true CN109886279A (en) | 2019-06-14 |
CN109886279B CN109886279B (en) | 2023-09-29 |
Family
ID=66926823
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910067366.3A Active CN109886279B (en) | 2019-01-24 | 2019-01-24 | Image processing method, device, computer equipment and storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109886279B (en) |
WO (1) | WO2020151153A1 (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110415239A (en) * | 2019-08-01 | 2019-11-05 | 腾讯科技(深圳)有限公司 | Image processing method, device, equipment, medical treatment electronic equipment and medium |
CN110646006A (en) * | 2019-09-02 | 2020-01-03 | 平安科技(深圳)有限公司 | Assembly path planning method and related device |
CN111462094A (en) * | 2020-04-03 | 2020-07-28 | 联觉(深圳)科技有限公司 | PCBA component detection method and device and computer readable storage medium |
CN112256906A (en) * | 2020-10-23 | 2021-01-22 | 安徽启新明智科技有限公司 | Method, device and storage medium for marking annotation on display screen |
CN115393586A (en) * | 2022-08-18 | 2022-11-25 | 北京爱科农科技有限公司 | Farmland breeding region dividing method and device, computer equipment and medium |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113553685B (en) * | 2021-07-27 | 2024-03-22 | 久瓴(江苏)数字智能科技有限公司 | Lighting device arrangement method and device and electronic equipment |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2141928A1 (en) * | 2008-06-30 | 2010-01-06 | Thomson Licensing S.A. | Device and method for analysing an encoded image |
CN106599866A (en) * | 2016-12-22 | 2017-04-26 | 上海百芝龙网络科技有限公司 | Multidimensional user identity identification method |
US20170220904A1 (en) * | 2015-04-02 | 2017-08-03 | Tencent Technology (Shenzhen) Company Limited | Training method and apparatus for convolutional neural network model |
CN107145889A (en) * | 2017-04-14 | 2017-09-08 | 中国人民解放军国防科学技术大学 | Target identification method based on double CNN networks with RoI ponds |
CN107808141A (en) * | 2017-11-08 | 2018-03-16 | 国家电网公司 | A kind of electric transmission line isolator explosion recognition methods based on deep learning |
CN108764143A (en) * | 2018-05-29 | 2018-11-06 | 北京字节跳动网络技术有限公司 | Image processing method, device, computer equipment and storage medium |
CN108876791A (en) * | 2017-10-23 | 2018-11-23 | 北京旷视科技有限公司 | Image processing method, device and system and storage medium |
CN109117876A (en) * | 2018-07-26 | 2019-01-01 | 成都快眼科技有限公司 | A kind of dense small target deteection model building method, model and detection method |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107301383B (en) * | 2017-06-07 | 2020-11-24 | 华南理工大学 | Road traffic sign identification method based on Fast R-CNN |
CN108133217B (en) * | 2017-11-22 | 2018-10-30 | 北京达佳互联信息技术有限公司 | Characteristics of image determines method, apparatus and terminal |
-
2019
- 2019-01-24 CN CN201910067366.3A patent/CN109886279B/en active Active
- 2019-05-30 WO PCT/CN2019/089196 patent/WO2020151153A1/en active Application Filing
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2141928A1 (en) * | 2008-06-30 | 2010-01-06 | Thomson Licensing S.A. | Device and method for analysing an encoded image |
US20170220904A1 (en) * | 2015-04-02 | 2017-08-03 | Tencent Technology (Shenzhen) Company Limited | Training method and apparatus for convolutional neural network model |
CN106599866A (en) * | 2016-12-22 | 2017-04-26 | 上海百芝龙网络科技有限公司 | Multidimensional user identity identification method |
CN107145889A (en) * | 2017-04-14 | 2017-09-08 | 中国人民解放军国防科学技术大学 | Target identification method based on double CNN networks with RoI ponds |
CN108876791A (en) * | 2017-10-23 | 2018-11-23 | 北京旷视科技有限公司 | Image processing method, device and system and storage medium |
CN107808141A (en) * | 2017-11-08 | 2018-03-16 | 国家电网公司 | A kind of electric transmission line isolator explosion recognition methods based on deep learning |
CN108764143A (en) * | 2018-05-29 | 2018-11-06 | 北京字节跳动网络技术有限公司 | Image processing method, device, computer equipment and storage medium |
CN109117876A (en) * | 2018-07-26 | 2019-01-01 | 成都快眼科技有限公司 | A kind of dense small target deteection model building method, model and detection method |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110415239A (en) * | 2019-08-01 | 2019-11-05 | 腾讯科技(深圳)有限公司 | Image processing method, device, equipment, medical treatment electronic equipment and medium |
CN110415239B (en) * | 2019-08-01 | 2022-12-16 | 腾讯科技(深圳)有限公司 | Image processing method, image processing apparatus, medical electronic device, and medium |
CN110646006A (en) * | 2019-09-02 | 2020-01-03 | 平安科技(深圳)有限公司 | Assembly path planning method and related device |
CN111462094A (en) * | 2020-04-03 | 2020-07-28 | 联觉(深圳)科技有限公司 | PCBA component detection method and device and computer readable storage medium |
CN112256906A (en) * | 2020-10-23 | 2021-01-22 | 安徽启新明智科技有限公司 | Method, device and storage medium for marking annotation on display screen |
CN115393586A (en) * | 2022-08-18 | 2022-11-25 | 北京爱科农科技有限公司 | Farmland breeding region dividing method and device, computer equipment and medium |
Also Published As
Publication number | Publication date |
---|---|
CN109886279B (en) | 2023-09-29 |
WO2020151153A1 (en) | 2020-07-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109886279A (en) | Image processing method, device, computer equipment and storage medium | |
CN110517278B (en) | Image segmentation and training method and device of image segmentation network and computer equipment | |
CN111079632A (en) | Training method and device of text detection model, computer equipment and storage medium | |
US11282271B2 (en) | Method in constructing a model of a scenery and device therefor | |
CN110675440B (en) | Confidence evaluation method and device for three-dimensional depth data and computer equipment | |
CN109711419A (en) | Image processing method, device, computer equipment and storage medium | |
CN113850807B (en) | Image sub-pixel matching positioning method, system, device and medium | |
CN115601774B (en) | Table recognition method, apparatus, device, storage medium and program product | |
CN109102524B (en) | Tracking method and tracking device for image feature points | |
CN108122280A (en) | The method for reconstructing and device of a kind of three-dimensional point cloud | |
CN114254584A (en) | Comparison method, modeling method and device of chip products and storage medium | |
CN112348116A (en) | Target detection method and device using spatial context and computer equipment | |
CN113160330A (en) | End-to-end-based camera and laser radar calibration method, system and medium | |
CN115713487A (en) | Defect identification method, device and storage medium for X-ray welding seam image | |
CN114048845B (en) | Point cloud repairing method and device, computer equipment and storage medium | |
CN111143146A (en) | Health state prediction method and system of storage device | |
US20160267352A1 (en) | System and method for constructing a statistical shape model | |
CN109360215B (en) | Method, device and equipment for searching outer contour of three-dimensional model and storage medium | |
CN115082592A (en) | Curve generation method, system, computer equipment and storage medium | |
CN111311731B (en) | Random gray level map generation method and device based on digital projection and computer equipment | |
CN117193278A (en) | Method, apparatus, computer device and storage medium for dynamic edge path generation | |
CN114677468A (en) | Model correction method, device, equipment and storage medium based on reverse modeling | |
CN110489510B (en) | Road data processing method and device, readable storage medium and computer equipment | |
CN110211230B (en) | Space planning model integration method and device, computer equipment and storage medium | |
CN114612609A (en) | Curved surface reflection line calculation method, curved surface reflection line calculation device, computer equipment and 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 |