CN105321141B - A kind of picture load equalization processing method and device - Google Patents
A kind of picture load equalization processing method and device Download PDFInfo
- Publication number
- CN105321141B CN105321141B CN201510593034.0A CN201510593034A CN105321141B CN 105321141 B CN105321141 B CN 105321141B CN 201510593034 A CN201510593034 A CN 201510593034A CN 105321141 B CN105321141 B CN 105321141B
- Authority
- CN
- China
- Prior art keywords
- boundary
- image
- block
- image block
- textural characteristics
- 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.)
- Active
Links
Abstract
The invention discloses a kind of picture load equalization processing method and devices, include the following steps:Original image f (x, y) is divided into multiple images block with lateral along longitudinal direction;The textural characteristics of two image blocks in left and right of longitudinal boundary are calculated, the mobile longitudinal boundary keeps the textural characteristics similarity between two image blocks of described longitudinal boundary or so maximum;The textural characteristics of two image blocks up and down of horizontal boundary are calculated, the mobile horizontal boundary keeps the textural characteristics similarity between two image blocks of the longitudinal boundary or more maximum;The determining corresponding final image block of final horizontal boundary and longitudinal boundary is respectively allocated to different computing units and carries out parallel processing.The load balancing between different parallel processing elements may be implemented in this method, reduces power consumption, while meeting degree of parallelism demand, realizes that efficient, real-time image Parallel Processing, parallel efficiency are optimal.
Description
【Technical field】
The present invention relates to a kind of picture load equalization processing method and devices.
【Background technology】
The development of computing technique vision technique, the proposition of video coding technique of new generation, let us enjoy high-resolution
The vision grand banquet come with ultrahigh resolution image and video tape, but this is but also calculating in image procossing and Video coding
Complexity increases substantially, and huge challenge, while parallel processing architecture are brought for parallel super-resolution image/video processing in real time
It is the problem of load imbalance, power consumption penalty may also be brought to increase, how that the parallel processing strategy of real-time high-efficiency and load is equal
Weighing apparatus technological perfectionism combines the development trend for having become computer vision and digital image processing field.
Either image procossing or Video coding, the important composition part of concurrent technique are the methods of piecemeal.Mesh
Preceding method of partition be easy to cause the load imbalance between different parallel processing elements.
【Invention content】
In order to solve the deficiencies in the prior art, the present invention provides a kind of picture load equalization processing method and device, from
And make the load between different parallel processing elements as balanced as possible.
A kind of picture load equalization processing method, includes the following steps:
Original image f (x, y) is divided into multiple images block by S1 with lateral along longitudinal direction;
S2, calculates the textural characteristics of two image blocks in left and right of longitudinal boundary, and the mobile longitudinal boundary makes the longitudinal direction
Textural characteristics similarity between the image block of two, boundary or so is maximum;
S3, calculates the textural characteristics of two image blocks up and down of horizontal boundary, and the mobile horizontal boundary makes the longitudinal direction
Textural characteristics similarity above and below boundary between two image blocks is maximum;
The determining corresponding final image block of final horizontal boundary and longitudinal boundary is respectively allocated to different meters by S4
It calculates unit and carries out parallel processing.
In one embodiment, in step S2 and S3, the textural characteristics are texture gradient feature, as follows
Calculate the texture gradient feature of image block:
S21, with image f of Gaussian function G (x, y) and original image f (x, y) progress convolution one width of formation after smooths(x,
y):
fs(x, y)=G (x, y) * f (x, y);
S22, the image f after calculating smoothlysThe gradient amplitude M of the pixel (x, y) of (x, y)(x,y):
S23 calculates the texture gradient feature X (i) of i-th piece of vertical or horizontal image block in original image f (x, y):
X (i)=Σ M(x,y)(i)。
In one embodiment, in step S2 and S3, determine that the texture between two image blocks is special as follows
It is maximum to levy similarity:
S31 calculates the texture gradient of the texture gradient feature X (i) and i+1 block image block of i-th piece of adjacent image block
Feature X (i+1);
S32 compares the texture gradient feature X (i) of i-th piece of image block and the texture gradient feature X (i of i+1 block image block
+ 1) size between, using i+1 block image block to the direction of i-th piece of image block as the shifting on corresponding boundary if X (i) is larger
Dynamic direction, using i-th piece of image block to the direction of i+1 block image block as the mobile side on corresponding boundary if X (i+1) is larger
To;
S33, along moving direction using basic unit as step-length moving boundary, two image blocks of cycle calculations boundaries on either side
Texture gradient feature difference Δ, until | Δ -2Ct| it is more than Δ or reaches both sides boundary;Wherein, Δ=| X (i+1)-X (i) |, Ct
Indicate the texture gradient feature of the basic unit moved through;
S34, by the Gradient Features C of the last one basic unittPrevious basic unit boundary as the i-th block diagram picture
Boundary between block and i+1 block image block.
The present invention also provides a kind of picture load equilibrium treatment devices, including such as lower unit:
First processing units, for original image f (x, y) to be divided into multiple images block with lateral along longitudinal direction;
Second processing unit, the textural characteristics of two image blocks in left and right for calculating longitudinal boundary, the mobile longitudinal direction
Boundary keeps the textural characteristics similarity between two image blocks of described longitudinal boundary or so maximum;
Third processing unit, the textural characteristics of two image blocks up and down for calculating horizontal boundary, the mobile transverse direction
Boundary keeps the textural characteristics similarity between two image blocks of the longitudinal boundary or more maximum;
Fourth processing unit, for distinguishing the corresponding final image block of the final horizontal boundary and longitudinal boundary determined
It distributes to different computing units and carries out parallel processing.
In one embodiment, the textural characteristics are texture gradient feature, in third processing unit and fourth process list
In member, the texture gradient feature of image block is calculated as follows:
S21, with image f of Gaussian function G (x, y) and original image f (x, y) progress convolution one width of formation after smooths(x,
y):
fs(x, y)=G (x, y) * f (x, y);
S22, the image f after calculating smoothlysThe gradient amplitude M of the pixel (x, y) of (x, y)(x,y):
S23 calculates the texture gradient feature X (i) of i-th piece of vertical or horizontal image block in original image f (x, y):
X (i)=Σ M(x,y)(i)。
In one embodiment, in third processing unit and fourth processing unit, two figures are determined as follows
As the textural characteristics similarity between block is maximum:
S31 calculates the texture gradient of the texture gradient feature X (i) and i+1 block image block of i-th piece of adjacent image block
Feature X (i+1);
S32 compares the texture gradient feature X (i) of i-th piece of image block and the texture gradient feature X (i of i+1 block image block
+ 1) size between, using i+1 block image block to the direction of i-th piece of image block as the shifting on corresponding boundary if X (i) is larger
Dynamic direction, using i-th piece of image block to the direction of i+1 block image block as the mobile side on corresponding boundary if X (i+1) is larger
To;
S33, along moving direction using basic unit as step-length moving boundary, two image blocks of cycle calculations boundaries on either side
Texture gradient feature difference Δ, until | Δ -2Ct| it is more than Δ or reaches both sides boundary;Wherein, Δ=| X (i+1)-X (i) |, Ct
Indicate the texture gradient feature of the basic unit moved through;
S34, by the Gradient Features C of the last one basic unittPrevious basic unit boundary as the i-th block diagram picture
Boundary between block and i+1 block image block.
The present invention determines Boundary Moving direction according to initial adjacent image block texture gradient feature difference size, and then presses
According to the principle moving boundary for so that adjacent two image blocks texture gradient feature difference reduces, once there is the line between adjacent image block
Reason characteristic similarity reduction then stops immediately, which is effectively guaranteed adjacent parallel block and has similar grain feature, simultaneously
Reduce computation complexity.
The load balancing between different parallel processing elements may be implemented in this method, reduces power consumption, while meeting parallel
Degree demand realizes efficient, real-time image Parallel Processing, and parallel efficiency is optimal, will make computer vision and image procossing
Each research field is benefited extensively.
【Description of the drawings】
Fig. 1 is the flow chart of the picture load equalization processing method of an embodiment of the present invention;
Fig. 2 is the Boundary Moving schematic diagram of the image block of an embodiment of the present invention.
【Specific implementation mode】
The following further describes in detail the preferred embodiments of the invention.
As shown in Figure 1, a kind of picture load equalization processing method of embodiment, includes the following steps:
S1:Input original image f (x, y).Obtained by original image f (x, y) can be acquired as video camera, camera, or
Video image of the person as obtained by PC Tools, data format are unlimited.
S2, by original image f (x, y) along longitudinal direction be laterally divided into M × N number of image block, vertical and horizontal be respectively M and
N.Rectangular uniform division can be carried out as unit of minimum unit to the original image f (x, y) of input, form initial M × N number of
Primary image block.The boundary position that current longitudinally divided image block is characterized with i, is characterized with j and currently laterally divides image block
Boundary position, it is 1 that when initialization assigns i and j initial values respectively.
S3 calculates the gradient amplitude of each location of pixels in original image f (x, y) using Canny operators.
S31, with image f of Gaussian function G (x, y) and original image f (x, y) progress convolution one width of formation after smooths(x,
y):
fs(x, y)=G (x, y) * f (x, y);
S32, the image f after calculating smoothlysThe gradient amplitude M of the pixel (x, y) of (x, y)(x,y):
S4 calculates the texture gradient feature X (i) of longitudinal i-th piece of image block in original image f (x, y):
X (i)=Σ M(x,y)(i)
Wherein X (i) values are bigger, show that texture is more complicated.
S5 determines i-th initial moving direction in boundary.Calculate the texture ladder of i-th boundary or so two adjacent images block
Feature X (i) and X (i+1) is spent, compares the size between X (i) and X (i+1), by the module direction with larger Gradient Features
As Boundary Moving direction, even X (i+1) is larger, then i-th boundary is moved to i+1 image block, if X (i) is larger,
It is moved to i-th of image block on i-th boundary.
S6, after the moving direction for determining i-th boundary, in the direction using basic unit as step-length L moving boundaries, wherein respectively
The texture gradient feature of basic unit is with CtIt indicates, the texture gradient of the left and right sides image block on i-th boundary of cycle calculations is special
Discrepancy delta is levied, until | Δ -2Ct| it is more than Δ or reaches both sides boundary.
Wherein, it is moved to the left boundary and the boundary that moves right is poor for the texture gradient feature of adjacent two two image blocks
The computational methods of different Δ simultaneously differ, wherein calculation criterion to the left is as follows:
Δ=X (i)-X (i+1)
Move right boundary calculation criterion it is as follows:
Δ=X (i+1)-X (i)
S7, by the Gradient Features C of the last one basic unittPrevious basic unit boundary as final i-th
Boundary.For example, after i-th boundary moves right 5 basic units, | Δ -2Ct| it is more than Δ, then i-th boundary is being moved
The boundary obtained after the 4th basic unit is moved as i-th article of final boundary.
S8 after the determination for completing i-th boundary, judges whether i is less than and divides total number of borders M-1, i adds 1 if being less than
After carry out being recycled to S5, otherwise enter S9.
S9 determines that the final position of horizontal boundary j, method are identical as the final position of longitudinal boundary i is determined.To most
The image block for completing the load balancing of whole image eventually divides.
The determining corresponding final image block of final horizontal boundary and longitudinal boundary is respectively allocated to different by S10
Computing unit carries out parallel processing.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that
The specific implementation of the present invention is confined to these explanations.For those of ordinary skill in the art to which the present invention belongs, exist
Under the premise of not departing from present inventive concept, a number of simple deductions or replacements can also be made, all shall be regarded as belonging to the present invention by
The scope of patent protection that the claims submitted determine.
Claims (2)
1. a kind of picture load equalization processing method, characterized in that include the following steps:
Original image f (x, y) is divided into multiple images block by S1 with lateral along longitudinal direction;
S2, calculates the textural characteristics of two image blocks in left and right of longitudinal boundary, and the mobile longitudinal boundary makes the longitudinal boundary
Textural characteristics similarity between two image blocks in left and right is maximum;
S3, calculates the textural characteristics of two image blocks up and down of horizontal boundary, and the mobile horizontal boundary makes the horizontal boundary
Textural characteristics similarity between upper and lower two image blocks is maximum;
The determining corresponding final image block of final horizontal boundary and longitudinal boundary is respectively allocated to different calculating lists by S4
Member carries out parallel processing;
In step S2 and S3, the textural characteristics are texture gradient feature, calculate the texture ladder of image block as follows
Spend feature:
S21, with image f of Gaussian function G (x, y) and original image f (x, y) progress convolution one width of formation after smooths(x,y):
fs(x, y)=G (x, y) * f (x, y);
Wherein, Gaussian function
S22, the image f after calculating smoothlysThe gradient amplitude M of the pixel (x, y) of (x, y)(x,y):
Wherein,
S23 calculates the texture gradient feature X (i) of i-th piece of vertical or horizontal image block in original image f (x, y):
X (i)=∑ M(x,y)(i);
In step S2 and S3, determine that the textural characteristics similarity between two image blocks is maximum as follows:
S31 calculates the texture gradient feature X (i) of i-th piece of adjacent image block and the texture gradient feature X of i+1 block image block
(i+1);
S32 compares the texture gradient feature X (i) of i-th piece of image block and the texture gradient feature X (i+1) of i+1 block image block
Between size, using i+1 block image block to the direction of i-th piece of image block as the mobile side on corresponding boundary if X (i) is larger
To using i-th piece of image block to the direction of i+1 block image block as the moving direction on corresponding boundary if X (i+1) is larger;
S33, along moving direction using basic unit as step-length moving boundary, the texture of two image blocks of cycle calculations boundaries on either side
Gradient Features discrepancy delta, until | Δ -2Ct| it is more than Δ or reaches both sides boundary;Wherein, Δ=| X (i+1)-X (i) |, CtIt indicates
The texture gradient feature of the basic unit moved through;
S34, by the Gradient Features C of the last one basic unittPrevious basic unit boundary as i-th piece of image block with
Boundary between i+1 block image block.
2. a kind of picture load equilibrium treatment device, characterized in that including such as lower unit:
First processing units, for original image f (x, y) to be divided into multiple images block with lateral along longitudinal direction;
Second processing unit, the textural characteristics of two image blocks in left and right for calculating longitudinal boundary, the mobile longitudinal boundary
Keep the textural characteristics similarity between two image blocks of described longitudinal boundary or so maximum;
Third processing unit, the textural characteristics of two image blocks up and down for calculating horizontal boundary, the mobile horizontal boundary
Keep the textural characteristics similarity between two image blocks of the horizontal boundary or more maximum;
Fourth processing unit, the corresponding final image block of final horizontal boundary and longitudinal boundary for will determine distribute respectively
Parallel processing is carried out to different computing units;
The textural characteristics are that texture gradient feature is counted as follows in second processing unit and third processing unit
Calculate the texture gradient feature of image block:
S21, with image f of Gaussian function G (x, y) and original image f (x, y) progress convolution one width of formation after smooths(x,y):
fs(x, y)=G (x, y) * f (x, y);
Wherein, Gaussian function
S22, the image f after calculating smoothlysThe gradient amplitude M of the pixel (x, y) of (x, y)(x,y):
Wherein,
S23 calculates the texture gradient feature X (i) of i-th piece of vertical or horizontal image block in original image f (x, y):
X (i)=∑ M(x,y)(i);
In second processing unit and third processing unit, the textural characteristics phase between two image blocks is determined as follows
It is maximum like degree:
S31 calculates the texture gradient feature X (i) of i-th piece of adjacent image block and the texture gradient feature X of i+1 block image block
(i+1);
S32 compares the texture gradient feature X (i) of i-th piece of image block and the texture gradient feature X (i+1) of i+1 block image block
Between size, using i+1 block image block to the direction of i-th piece of image block as the mobile side on corresponding boundary if X (i) is larger
To using i-th piece of image block to the direction of i+1 block image block as the moving direction on corresponding boundary if X (i+1) is larger;
S33, along moving direction using basic unit as step-length moving boundary, the texture of two image blocks of cycle calculations boundaries on either side
Gradient Features discrepancy delta, until | Δ -2Ct| it is more than Δ or reaches both sides boundary;Wherein, Δ=| X (i+1)-X (i) |, CtIt indicates
The texture gradient feature of the basic unit moved through;
S34, by the Gradient Features C of the last one basic unittPrevious basic unit boundary as i-th piece of image block with
Boundary between i+1 block image block.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510593034.0A CN105321141B (en) | 2015-09-17 | 2015-09-17 | A kind of picture load equalization processing method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510593034.0A CN105321141B (en) | 2015-09-17 | 2015-09-17 | A kind of picture load equalization processing method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105321141A CN105321141A (en) | 2016-02-10 |
CN105321141B true CN105321141B (en) | 2018-07-20 |
Family
ID=55248467
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510593034.0A Active CN105321141B (en) | 2015-09-17 | 2015-09-17 | A kind of picture load equalization processing method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105321141B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111182301A (en) * | 2018-11-12 | 2020-05-19 | 北京眼神科技有限公司 | Method, device, equipment and system for selecting optimal quantization parameter during image compression |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8358846B2 (en) * | 2007-11-13 | 2013-01-22 | Symantec Corporation | Scanning images for pornography |
CN103745447A (en) * | 2014-02-17 | 2014-04-23 | 东南大学 | Fast parallel achieving method for non-local average filtering |
CN103942554A (en) * | 2014-05-07 | 2014-07-23 | 苏州搜客信息技术有限公司 | Image identifying method and device |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8023732B2 (en) * | 2006-07-26 | 2011-09-20 | Siemens Aktiengesellschaft | Accelerated image registration by means of parallel processors |
-
2015
- 2015-09-17 CN CN201510593034.0A patent/CN105321141B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8358846B2 (en) * | 2007-11-13 | 2013-01-22 | Symantec Corporation | Scanning images for pornography |
CN103745447A (en) * | 2014-02-17 | 2014-04-23 | 东南大学 | Fast parallel achieving method for non-local average filtering |
CN103942554A (en) * | 2014-05-07 | 2014-07-23 | 苏州搜客信息技术有限公司 | Image identifying method and device |
Also Published As
Publication number | Publication date |
---|---|
CN105321141A (en) | 2016-02-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Volz et al. | Modeling temporal coherence for optical flow | |
US20190347847A1 (en) | View generation from a single image using fully convolutional neural networks | |
CN109087273A (en) | Image recovery method, storage medium and the system of neural network based on enhancing | |
CN103402045A (en) | Image de-spin and stabilization method based on subarea matching and affine model | |
CN100446573C (en) | Implementation device in VLSI of filter for removing blocking effect based on AVS | |
CN102184533A (en) | Non-local-restriction-based total variation image deblurring method | |
CN102243711A (en) | Neighbor embedding-based image super-resolution reconstruction method | |
CN110533579B (en) | Video style conversion method based on self-coding structure and gradient order preservation | |
CN106558040A (en) | Character image treating method and apparatus | |
CN109949221A (en) | A kind of image processing method and electronic equipment | |
CN107515736A (en) | A kind of method for accelerating depth convolutional network calculating speed on embedded device | |
CN111598918A (en) | Video image stabilization motion estimation method based on reference frame optimization and foreground and background separation | |
CN106897983A (en) | The processing method and image processing apparatus of a kind of multiple image set | |
CN107067452A (en) | A kind of film 2D based on full convolutional neural networks turns 3D methods | |
CN102355589B (en) | Super-resolution reconstruction method based on parameter self-optimizing multi-swarm particle swarm optimization | |
CN105321141B (en) | A kind of picture load equalization processing method and device | |
CN102932643A (en) | Expanded variable block movement estimation circuit suitable for HEVC (high efficiency video coding) standard | |
CN108171651B (en) | Image alignment method based on multi-model geometric fitting and layered homography transformation | |
CN116152121B (en) | Curved surface screen generating method and correcting method based on distortion parameters | |
CN105279732B (en) | A kind of method and system applying to the quick deformation of flat image | |
CN103258318B (en) | A kind of image noise reduction disposal route and system | |
CN106454386A (en) | JND (Just-noticeable difference) based video encoding method and device | |
CN111161134A (en) | Image artistic style conversion method based on gamma conversion | |
CN106600629B (en) | A kind of light stream estimation method towards Large Scale Motion | |
CN105224290B (en) | A kind of picture load equalization processing method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |