CN106780418B - A kind of Image Intensified System - Google Patents
A kind of Image Intensified System Download PDFInfo
- Publication number
- CN106780418B CN106780418B CN201611055338.2A CN201611055338A CN106780418B CN 106780418 B CN106780418 B CN 106780418B CN 201611055338 A CN201611055338 A CN 201611055338A CN 106780418 B CN106780418 B CN 106780418B
- Authority
- CN
- China
- Prior art keywords
- block
- pixels
- image
- processing unit
- high frequency
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Abstract
The invention discloses a kind of Image Intensified Systems.The system includes: central processing unit, and the central processing unit receives image；Graphics processing unit；And memory, wherein under the control of the central processing unit, the graphics processing unit: according to scheduled magnification ratio, being amplified image, the low frequency part as enlarged drawing；Obtain the high frequency section of described image；The central processing unit: block is drawn to the low frequency part of the enlarged drawing, to match the block of pixels of described image；Using the block of pixels of described image, the corresponding relationship of the block of pixels of the low frequency part of the enlarged drawing and the block of pixels of high frequency section is established；The block of pixels of high frequency section is added to the low frequency part of the enlarged drawing, the increased image of resolution ratio is obtained.This method is simple, efficient, greatly reduces the calculation amount of image procossing, can be realized the real-time conversion and viewing of high-definition image.
Description
Technical field
The present invention relates to field of image processing more particularly to a kind of Image Intensified Systems based on Heterogeneous Computing.
Background technique
In field of image processing, image enhancement technique is critically important research direction.Image enhancement technique mainly study by
Originally unsharp image is apparent from or enhances the feature of certain concerns, inhibits the feature of non-interesting, to reach improvement figure
Image quality amount, abundant information amount reinforce the purpose of image interpretation and recognition effect.
The existing technology for increasing image resolution ratio can be divided into two major classes, and one kind is led to based on multiple low-resolution images
It crosses and image is corrected, a secondary high-definition picture, such methods are then obtained in the way of interpolation, reconstruction or study etc.
The superiority and inferiority of effect is heavily dependent on the kinematic parameter between two sub-pictures, and computation complexity is higher.Second class side
Method is to obtain high-definition picture by various interpolation algorithms based on single low-resolution image.Since this method pertains only to
Low-resolution image itself, therefore the scope of application is wider.
The interpolation algorithm being widely used includes arest neighbors interpolation, bilinear interpolation and cube sum etc..Wherein, most adjacent
Nearly interpolation algorithm is most simple, but to be also most also easy to produce pixel value discontinuous for closest interpolation algorithm, so as to cause blocking artifact, Jin Erzao
Fuzzy at image, picture quality effect is generally not ideal enough after amplification.Bilinear interpolation algorithm is complex, and bilinear interpolation is calculated
Method is not in the discontinuous situation of pixel value, and amplified picture quality is higher, but since bilinear interpolation has low pass filtered
The property of wave device keeps high fdrequency component impaired, it is possible that the edge contour of each theme and detail section can be made in image one
Determine to thicken in degree.Cube sum algorithm is complicated, can retain comparatively clearly edge contour and details, can
Reduce or avoid the pectination phenomenon of the crenellated phenomena of each theme edge contour and detail section in amplified image, interpolation effect
Fruit is relatively true, makes picture quality more improvement after amplification, however computation complexity also highest.
Summary of the invention
The object of the present invention is to provide a kind of systems for generating high-definition image by increasing image resolution ratio, existing to solve
There is the shortcomings that scheme computation complexity for increasing image resolution ratio is high, cannot achieve conversion in real time and watches.
According to an aspect of the present invention, a kind of Image Intensified System is provided, comprising: central processing unit, the center
Processing unit receives image；Graphics processing unit；And memory, wherein described under the control of the central processing unit
Graphics processing unit: according to scheduled magnification ratio, amplifying image, the low frequency part as enlarged drawing；Obtain institute
State the high frequency section of image；The central processing unit: block is drawn to the low frequency part of the enlarged drawing, to match described image
Block of pixels；Using the block of pixels of described image, the block of pixels and high frequency section of the low frequency part of the enlarged drawing are established
The corresponding relationship of block of pixels；The block of pixels of high frequency section is added to the low frequency part of the enlarged drawing, resolution ratio is obtained and increases
The image added.
Further, the central processing unit judges the coded format of described image, such as after receiving described image
Fruit described image is RGB coded image, then instructs the graphics processing unit that RGB coded image is converted into YCbCr code pattern
Picture.
Further, the graphics processing unit carries out bi-cubic interpolation to the CbCr component of image.
Further, under the control of the central processing unit, the graphics processing unit: according to scheduled amplification ratio
Rate amplifies the Y-component of image, the low frequency part of the Y-component as enlarged drawing；Obtain the Y-component of described image
High frequency section；The central processing unit: block is drawn to the low frequency part of the Y-component of the enlarged drawing, to match described image
Y-component block of pixels；Using the block of pixels of the Y-component of described image, the low frequency part of the Y-component of the enlarged drawing is established
Block of pixels and high frequency section block of pixels corresponding relationship；The block of pixels of high frequency section is added to the Y of the enlarged drawing
The low frequency part of component obtains the Y-component of the increased image of resolution ratio；Using after interpolation CbCr component and resolution ratio it is increased
Y-component synthesizes new images.
Further, the central processing unit judges whether the increased image of the resolution ratio meets resolution requirement,
If being unsatisfactory for the resolution requirement, the centre is supplied to using the increased image of the resolution ratio as initial pictures
Manage unit.
Further, image is amplified including repeatedly putting step by step to described image progress according to scheduled magnification ratio
Greatly.
Further, the magnification ratio amplified step by step every time in the multiple amplification step by step is N+1/N, and N is integer.
Further, the central processing unit establishes the low frequency portion of the enlarged drawing using the block of pixels of described image
The corresponding relationship of the block of pixels of the block of pixels and high frequency section divided includes: to obscure to described image, and then down-sampling, obtains
Downscaled images, the minification of the downscaled images and the magnification ratio of enlarged drawing are corresponding, search and institute in described image
Corresponding second block of pixels of the first block of pixels in the low frequency part of enlarged drawing is stated, is searched in the downscaled images and the
The most like third block of pixels of two block of pixels, wherein by the high frequency of the 4th block of pixels corresponding with third block of pixels in described image
Partial stack is to the first block of pixels.
Further, the central processing unit is increased and the matched high frequency section of specific position by nearest neighbor algorithm
Block of pixels.
Further, it the central processing unit: is obtained by nearest neighbor algorithm with the matched k candidate's of specific position
The block of pixels of high frequency section；The block of pixels of the high frequency section candidate from k obtains a tentative optimal block of pixels；Using tentative
Optimal block of pixels spliced, to obtain initial high frequency imaging；Final high frequency imaging, the conclusion are obtained by induction
Method includes: high frequency imaging N+1 to be calculated based on high frequency imaging N, in high frequency imaging N 1) for the high frequency imaging N obtained
Pixel block analysis centered on (px, py) finds optimal block of pixels P from k candidate block of pixels so that block of pixels P satisfaction with
Lower condition: the boundary of block of pixels extension is as consistent with the value of the corresponding region of high frequency imaging N as possible, and 2) utilize optimal picture
Plain block splices high frequency imaging N+1,3) it repeats step 1) and 2) there is no variation up to resulting image, or reach iteration most
Big number.
Compared with prior art, the invention has the advantages that
Compared with existing interpolation algorithm, the system disclosed by the invention for increasing image resolution ratio is simple, efficient, drops significantly
The low calculation amount of image procossing, and the computational efficiency of lifting system can be realized the real-time conversion and viewing of high-definition image.
Detailed description of the invention
For the above and other advantages and features for each embodiment that the present invention is furture elucidated, will be presented with reference to attached drawing
The more specific description of various embodiments of the present invention.It is appreciated that these attached drawings only describe exemplary embodiments of the invention, therefore
It is not to be regarded as being restriction on its scope.In the accompanying drawings, identical or corresponding component will be indicated with same or similar label.
Fig. 1 shows heterogeneous computing system 100 according to an embodiment of the invention.
Fig. 2 shows the flow charts of the method 200 according to an embodiment of the invention for improving image resolution ratio.
Fig. 3 shows the example schematic diagram for improving image resolution ratio according to one embodiment of present invention.
Fig. 4 is shown to be searched in the image of different frequency to improve image matching accuracy according to one embodiment of present invention
The schematic diagram of rope respective pixel block.
Fig. 5 is shown according to one embodiment of present invention for ensuring to splice the schematic diagram of continuity
Specific embodiment
In the following description, with reference to each embodiment, present invention is described.However, those skilled in the art will recognize
Know can in the case where none or multiple specific details or with other replacements and/or addition method, material or component
Implement each embodiment together.In other situations, well known method or operation are not shown or are not described in detail in order to avoid causing this hair
The aspects of bright each embodiment is obscure.Similarly, for purposes of explanation, specific quantity and configuration are elaborated, in order to provide
To the comprehensive understanding of the embodiment of the present invention.However, the present invention can be implemented in the case where no specific detail.
Fig. 1 shows heterogeneous computing system 100 according to an embodiment of the invention.During heterogeneous computing system 100 includes
Central Processing Unit (CPU) 110 and graphics processing unit (GPU) 120.Central processing unit (CPU) 110 and graphics processing unit
(GPU) 120 receiving host side command queue.
Image includes high frequency section and low frequency part, and wherein high frequency section is distributed mainly on the side of each main body in image
Edge outline portion and detail section, low frequency part are distributed mainly on the non-edge outline portion of each main body in image.Pass through
The high frequency section and enlarged drawing for obtaining original image respectively, the low frequency part of Block- matching original image is drawn to enlarged drawing, can be obtained
To the matched piece of coordinate in original image, according to the height of the available original image corresponding position of coordinate of the block in original image
High frequency section is added to and is amplified the low frequency part of image by frequency part, to obtain new full rate image, i.e. high-resolution
Image.
Below with reference to the method for raising image resolution ratio of Fig. 1 and Fig. 2 description based on Heterogeneous Computing.
In the method disclosed in the present application for improving image resolution ratio, CPU 110 is for carrying out image reception, image texture
Lookup synthesized with matching, image and whole control flow because wherein being handled comprising a large amount of logic judgment.GPU
120 for zooming in and out processing by upper and lower sampling matrix operation, and a large amount of parallel work-flows of this thread independence are very suitable
It closes and is executed on the large-scale parallel hardware as GPU 120.By the reasonable utilization to CPU 110 and GPU 120, by isomery
The huge advantage of calculating is discharged, so that spending 2-3 hours by the computer that is preferably configured in the prior art just can be with
The HD video of conversion can carry out converting and watching in real time via Heterogeneous Computing method of the invention.
Fig. 2 shows the flow charts of the method 200 according to an embodiment of the invention for improving image resolution ratio.The present invention
The disclosed method for improving image resolution ratio is completed under the control of CPU 110.
Firstly, in step 201 image can be received by CPU 110.Under the control of CPU 110, proceed in step 202,
By GPU 120 according to scheduled magnification ratio, to original image I0It amplifies, obtains original image I0Enlarged drawing, amplification again
(upsampling) or image interpolation (interpolating) are referred to as up-sampled, main purpose is amplification original image, to image
Zoom operations can not bring the more information about the image, the quality of image will inevitably be affected.Therefore,
In step 202, by original image I0The image of up-sampling interpolation acquisition is carried out as the low frequency part L for being amplified image1。
It under the control of CPU 110, proceeds in step 203, original image I is obtained by GPU 1200High frequency section H0。
In step 203, by GPU 120 according to scheduled reduction ratio 1/a, to original image I0It is reduced, obtains original image I0's
Downscaled images, downscaled images are also known as down-sampling.By GPU 120 according to magnification ratio a, downscaled images are up-sampled,
To obtain low-frequency image L0.It is then up-sampled due to first carrying out down-sampling to original image, compared with original image, low frequency figure
As L0Reduce high frequency section, therefore low-frequency image L0For relative to original image I0Low frequency part.By GPU120 by original image
I0Subtract low-frequency image L0Obtain high frequency section H0。
Under the control of CPU 110, proceeds in step 204, drawn by low frequency part of the CPU 110 to enlarged drawing
Block, to match the low frequency part of original image.In step 204, the low frequency part of enlarged drawing can be divided into predefined size
Block of pixels.In an embodiment of the present invention, the block of pixels of predefined size can be 5 × 5 pixel region, however model of the invention
Enclose the pixel region for being not limited to the size.Then search and the most like original image pixel of low frequency part block of pixels in original image
Block, so as to obtain coordinate of the low frequency part block of pixels in original image.
Under the control of CPU 110, proceed in step 205, by CPU 110 according to low frequency part block of pixels in original image
Coordinate as in, in high frequency section H0It is middle to obtain high frequency imaging block of pixels corresponding with original image block of pixels.
CPU 110 is according to low frequency part block of pixels and original image block of pixels, original image block of pixels and high frequency imaging block of pixels
Corresponding relationship, establish the corresponding relationship of low frequency part block of pixels Yu high frequency imaging block of pixels.
In step 206, high frequency imaging block of pixels is added to the low of enlarged drawing according to position corresponding relationship by CPU 110
Frequency part, to obtain the increased image of resolution ratio.
In an embodiment of the present invention, method 200 shown in Fig. 2 also optionally includes step 207, in step 207, is led to
It crosses CPU 110 and judges whether the increased image of the resolution ratio meets resolution requirement, if the increased image of the resolution ratio meets
Resolution requirement, then method 200 terminates and exports the image for meeting resolution requirement, otherwise image increased for the resolution ratio
Step 201-206 is repeated, until image obtained meets resolution requirement.
Next, the method that raising image resolution ratio of the invention is described in further detail in conjunction with specific embodiments.
Fig. 3 shows the example schematic diagram for improving image resolution ratio according to one embodiment of present invention.As shown in figure 3, right
Original image I0Carry out the low frequency part L that up-sampling interpolation obtains enlarged drawing1.According to scheduled reduction ratio 1/a, to original image
I0Down-sampling is carried out, original image I is obtained0Downscaled images downscaled images are up-sampled further according to magnification ratio a, thus
Obtain low-frequency image L0.By original image I0Subtract low-frequency image L0Obtain high frequency section H0.According to low frequency part block of pixels and original image
As the corresponding relationship of block of pixels, original image block of pixels and high frequency imaging block of pixels, low frequency part block of pixels and high frequency imaging are established
The corresponding relationship of block of pixels.By high frequency section H0It is added to and is amplified the low frequency part L of image1, to obtain high resolution graphics
Picture.
In one particular embodiment of the present invention, handled image uses YCbCr space encoding scheme.For
YCbCr space, Y refer to luminance component, and Cb refers to chroma blue component, and Cr refers to red chrominance component.Y of the human eye to video
Component is more sensitive, therefore in real image treatment process, processing method used by Y-component may differ from the place of CbCr component
Reason method.In the present embodiment, CbCr component directly carries out bi-cubic interpolation.And the raising image resolution ratio being described below in detail
Method be directed to Y-component numerical value, that is, below calculate in appearance parameter Ik、Lk、HkAll refer to the numerical value in the space Y.In reality
Treatment process in, CPU 110 receive image after, CPU 110 first determines whether the encoding scheme of the image, if the image is
RGB coded image then needs to instruct GPU 120 that RGB coded image is converted into YCbCr coded image.
Table 1 is shown to original image I0Carry out the process of classification scaling.
In table 1, U indicates up-sampling treatment, and D indicates down-sampling processing.
For grid G0, image effect is original image I0, according to scheduled reduction ratio 1/a, to original image I0Adopt
Sample up-samples downscaled images further according to magnification ratio a, to obtain low-frequency image L0.By original image I0Subtract low frequency
Image L0Obtain high frequency section H0。
From grid G1To grid Gl, difference is up-sampled step by step to image.Assuming that by G0To GlAmplification coefficient be α, lead to
Cross G1To GlTotal l grades of up-sampling difference is realized α times and is amplified.For example, it is assumed that amplification coefficient α=3, can be used 5:4,5:4,4:3
Difference is up-sampled with 3:2, that is, G0To G1For 5 → 4 up-samplings, G1To G2For 5 → 4 up-samplings, G2To G3For 4 → 3 up-samplings, G3
To G4For 3 → 2 up-samplings, obtained final amplification coefficient is 3+1/8, close to 3 times of amplification coefficient.In its of the invention
In its embodiment, other magnification ratios are can be used in up-sampling differences at different levels, can total amplifying stage determine according to actual needs
Number and from G0To GlAmplification coefficient used in every level-one.For example, it is assumed that expected from G0To GlAmplification coefficient be 3, it is available
Pyatyi 5 → 4 up-samples difference, and the amplification coefficient finally actually obtained is 1.255=3.05.
In one embodiment of the invention, N → N+1 (N is integer) up-sampling is such as given a definition:
Original image I is divided into N parts:
It is now to IkAbout ukIt is enlarged into ↑ Ik,
After amplification, finally obtained enlarged drawing is ↑ I
U hereink(i)=uk(- i) is constant.It in an embodiment of the present invention, can be according to specific amplification coefficient and image
Constant u used in each tomographic image of hierarchical arrangementk。
In one embodiment of the invention, N+1 → N down-sampling is such as given a definition:
Wherein, p=n mod N, q=(n-p)/N, dpThere is N kind to follow the example of: d0,d1,···,dN-1.In implementation of the invention
In example, d0,d1,···,dN-1For constant, can be used according to specific amplification coefficient and each tomographic image of image hierarchical arrangement
Constant d0,d1,···,dN-1。
For example, N=2, at this point, there are two types of value methods by p.
About constant dp,ukSetting method, can be obtained by some test samples.
Now, table 1 is returned to, for grid Gl,
Il(p)=Ll(p)+Hl(p)=Ll(p)+H0(q(p)) [5]
If by G0To GlAmplification coefficient be α, p is GlIn centre coordinate be (x, y) 5 × 5 block of pixels, due to original image
As I0With enlarged drawing LlSize it is different, it is therefore desirable to enlarged drawing block of pixels p is put into original image I0In scan for, search
Rope and the most like original image block of pixels of the enlarged drawing block of pixels.By original image I0Region to be searched setting centered on (x/
α, y/ α), 10 × 10 block of pixels of size, obtained in the region to be searched and p most like original image block of pixels q (p), therefore p
With q (p) it is corresponding be all 5 × 5 fritter.Then it repeats the above process, other block of pixels of enlarged drawing is scanned for
Match and be superimposed, finally obtains high-definition picture.
In an embodiment of the present invention, amplified step by step using N → N+1, be the advantage is that in progressive amplification process to the greatest extent
The holding precision of images more than possible amplifies the gained precision of images more compared with the method for disposably amplifying image in place step by step
It is high.
It deforms, makes due in image amplification, may result in rotation, offset, distortion of each element etc. in image
At in the subsequent problem for being superimposed and occurring alignment inaccuracy in splicing.In order to solve this problem, Fig. 4 is shown as improving
Image matching accuracy searches for the schematic diagram of respective pixel block in the image of different frequency.
To original image I0It is obscured, then down-sampling, is obtained
L-l=(I0*Bl)↓sl [6]
Wherein, sl represents the multiple of scaling, and Bl is fuzzy kernel.
Referring to Fig. 4, with L-lThe low frequency part of corresponding amplification sl multiple is Ll, in L-lWith LlBetween be original image I0
=L.
To match LlIn low frequency block A, first by CPU 110, in original image L0In search it is corresponding with low frequency block A
Block A ', then, in L-lIn search the block B ' most like with block A ', then by the height of corresponding with block B ' piece of B in original image L
Frequency partial stack is to low frequency block A.The search range during Block- matching can be increased by the above process, so that matching is more smart
Really.
In an embodiment of the present invention, the splicing between the block to ensure each high frequency is coherent as far as possible, can pass through k
Nearest neighbor algorithm increases the library of matched block of pixels.
If p ∈ GlIt is with (px,py) centered on N × N block of pixels, q=q (p) ∈ G0It is with (qx,qy) centered on N × N
Block of pixels, then norm are as follows:
Better result can be obtained by carrying out slight normalization to the norm:
Wherein
Truncate is truncation funcation:
Distance (p, q) indicates in G0 layers of q block of pixels and the difference size of corresponding Gl layers of p block of pixels,
That is, the similarity of two block of pixels.For multi-layer image, multiple Distance (p, q) can be obtained.By these Distance (p,
Q) it is ranked up by sequence from small to large, being worth the smallest Distance (p, q) indicates that the difference between two block of pixels is minimum.
The repetition of the block message of image is mainly reflected in translation, scaling, rotation.The description of front mainly for translation and
Scaling, if it is considered that the repeatability of rotation, then can provide matched quality.The form of dot chart when due to image, as long as
Consider eight kinds of rotations of image L0 (qx+iqy+j) to J (qx+iqy+j):
J(qx+ i, qy+ j)=L0(qx+ ai, qy+ bj), or
J(qx+ i, qy+ j)=L0(qx+ aj, qy+b·i) [11]
Wherein a=± 1, b=± 1.
Then, above-mentioned normalization norm is modified are as follows:
Wherein γ is non-negative parameter.
K optimal block of pixels can be obtained by front norm, for example, can use k=9 for 5 × 5 block of pixels.To ensure
The quality of block of pixels, it is assumed that optimal piece at most k corresponding to each piece, are then to abandon wherein when there are more candidate blocks
A part.
Need to meet following two item referring to Fig. 5 to ensure that the splicing between each high-frequency pixels block is coherent as far as possible
Part:
1. k candidate block of pixels is obtained, if used low frequency block of pixels is M × M, M > N.It requires corresponding to them
Low frequency part have continuity, this continuity is of overall importance.
2. pair spliced N × N high-frequency pixels block, it is desirable that there are continuity, i.e. the boundary A of its extension in boundary each other
× A, B × B will as far as possible with corresponding overlapping margins.This continuity is local.
It is described below and how continuity is realized by iterative method.
Firstly, it is tentative optimal to obtain one by k candidate block of pixels, and these block of pixels are stitched together and are obtained
An initial high frequency imaging Image (0) is obtained, carries out induction below:
1. setting the image Image (n) for having obtained n-th, image Image (n+1) is calculated now.
To the pixel block analysis every time centered on (px, py) in image Image (n), it is now to from k candidate pixel
Block { J } finds block of pixels P, and wherein block of pixels P meets the following conditions: the boundary Ω of block of pixels extension, Ξ will as far as possible and image
The value of the corresponding region of Image (n) is consistent:
Wherein
ISPreviousPatch is Boolean type, and whether represent is same with block of pixels selected by image Image (n)
It is a.
For H1 (Ω) only in the Ω of region, region Ω does not include N × N block of pixels at center.
Block of pixels J contains its block of pixels position (qx,qy) and swing (a, b) and its class in formula [11]
Type.
Definition about H1 (Ξ) is similar with H1 (Ω), difference be H1 (Ξ) only in the Ξ of region, without include region Ω and
N × N block of pixels.θ, α, β are specific parameter.
2. these block of pixels are spliced into Image (n+1)
3. being repeated up to resulting image, there is no variations, or reach the maximum times of iteration.
Final high frequency imaging is obtained by above-mentioned induction.
Each embodiment can be provided as may include be stored thereon with one or more machines of machine-executable instruction can
The computer program product of medium is read, these instructions are by the one of computer, computer network or other electronic equipments etc.
When a or multiple machines execute, one or more machines can be caused to execute the operation of each embodiment according to the present invention.Machine
It is (read-only to deposit that readable medium can include but is not limited to floppy disk, CD, CD-ROM (aacompactadisk read onlyamemory) and magneto-optic disk, ROM
Reservoir), RAM (random access memory), EPROM (Erasable Programmable Read Only Memory EPROM), EEPROM (electrically erasable
Read-only memory), magnetically or optically card, flash memory or other kinds of medium/machine suitable for storing machine-executable instruction
Device readable medium.
Furthermore, it is possible to download each embodiment as computer program product, wherein can be via communication link (for example, adjusting
Modulator-demodulator and/or network connection) by carrier wave or one or more data-signal handles of the realization of other propagation mediums and/or modulation
Program is transferred to requesting computer (for example, client computer) from remote computer (for example, server).Therefore, used herein
Machine readable media may include such carrier wave, but that is not required.
This is indicated to the reference of " a kind of embodiment ", " one embodiment ", " example embodiment ", " various embodiments " etc.
(each) embodiment of the invention of sample description may include specific feature, structure or characteristic, but not each embodiment must
Including specific feature, structure or characteristic.Further, some embodiments can have relative to spy described in other embodiments
It is some, whole in sign, or without any of these features.
In the description and in the claims which follow, term " coupling " and its derivative can be used." coupling " is used to refer to two
A or more element cooperating interacts, but can have or do not have intermediary's physics or electrical group between them
Part.
As used in the claims, unless otherwise indicated, ordinal adjectives " first ", " second ", " third " etc.
Deng being used to describe common element, only instruction refers to the different instances of analogous element, and not expected implies the member described in this way
Element necessarily is in given sequence, either time, space, classification or in any other manner.
The description of attached drawing and front gives the example of each embodiment.Those skilled in the art will appreciate that described
One of element or multiple can be combined into individual feature element well.It is alternatively possible to a little elements be split into more
A functional element.A kind of element from embodiment can be added to another embodiment.Such as, thus it is possible to vary it is described herein
The sequence of processing, and it is not limited to mode described herein.Moreover, it is not necessary to realize the dynamic of any flow chart by shown order
Make；Also everything is not necessarily needed to be implemented.Moreover, can be parallel with other movements independent of those of other movements movement
It executes.The range of each embodiment is never limited to these specific examples.Regardless of whether clearly providing in the description, such as material
Numerous changes of the structure and size of material and the difference used etc. are all possible.The range of each embodiment is at least following power
It is broad as benefit requirement is specified.
The foregoing describe several embodiments of the invention.However, the present invention can be embodied as other concrete forms without carrying on the back
From its spirit or essential characteristics.Described embodiment should all be to be considered merely as illustrative and not restrictive in all respects.
Therefore, the scope of the present invention is by the appended claims rather than foregoing description limits.Fall into the equivalent scheme of claims
All changes in meaning and scope are covered by the range of claims.
Claims (9)
1. a kind of Image Intensified System, comprising:
Central processing unit, the central processing unit receive image；
Graphics processing unit；
And memory,
Wherein, under the control of the central processing unit, the graphics processing unit: according to scheduled magnification ratio, to figure
The Y-component of picture amplifies, the low frequency part of the Y-component as enlarged drawing；Obtain the radio-frequency head of the Y-component of described image
Point；
The central processing unit: drawing block to the low frequency part of the Y-component of the enlarged drawing, to match Y points of described image
The block of pixels of amount；Using the block of pixels of the Y-component of described image, the picture of the low frequency part of the Y-component of the enlarged drawing is established
The corresponding relationship of the block of pixels of plain block and high frequency section；The block of pixels of high frequency section is added to the Y-component of the enlarged drawing
Low frequency part, obtain the increased image of resolution ratio Y-component；Divided using the CbCr component after interpolation with the increased Y of resolution ratio
Amount synthesis new images.
2. the system as claimed in claim 1, which is characterized in that the central processing unit is sentenced after receiving described image
The coded format of disconnected described image instructs the graphics processing unit to compile RGB if described image is RGB coded image
Code image is converted into YCbCr coded image.
3. system as claimed in claim 2, which is characterized in that the graphics processing unit carries out the CbCr component of image double
Cube interpolation.
4. the system as claimed in claim 1, which is characterized in that the central processing unit judges the increased figure of the resolution ratio
Seem it is no meet resolution requirement, if being unsatisfactory for the resolution requirement, using the increased image of the resolution ratio as just
Beginning image is supplied to the central processing unit.
5. the system as claimed in claim 1, which is characterized in that amplified image including right according to scheduled magnification ratio
The Y-component progress of described image is repeatedly amplified step by step.
6. system as claimed in claim 5, which is characterized in that the amplification amplified step by step every time in the multiple amplification step by step
Ratio is N+1/N, and N is integer.
7. the system as claimed in claim 1, which is characterized in that the central processing unit utilizes the Y-component of described image
Block of pixels establishes the corresponding relationship packet of the block of pixels of the low frequency part of the Y-component of the enlarged drawing and the block of pixels of high frequency section
It includes:
Described image is obscured, then down-sampling, obtains downscaled images, the minification and enlarged drawing of the downscaled images
The magnification ratio of picture is corresponding,
The second block of pixels corresponding with the first block of pixels in the low frequency part of the enlarged drawing is searched in described image,
The third block of pixels most like with the second block of pixels is searched in the downscaled images,
Wherein the high frequency section of the 4th block of pixels corresponding with third block of pixels in described image is added to the first block of pixels.
8. the system as claimed in claim 1, which is characterized in that the central processing unit is increased and spy by nearest neighbor algorithm
The block of pixels of matched high frequency section is set in positioning.
9. system as claimed in claim 8, which is characterized in that the central processing unit:
The block of pixels with matched k candidate high frequency section of specific position is obtained by nearest neighbor algorithm；
The block of pixels of the high frequency section candidate from k obtains a tentative optimal block of pixels；
Spliced using tentative optimal block of pixels, to obtain initial high frequency imaging；
Final high frequency imaging is obtained by induction, the induction includes:
1) for the high frequency imaging N obtained, high frequency imaging N+1 is calculated based on high frequency imaging N, in high frequency imaging N (px,
Py the pixel block analysis centered on) finds optimal block of pixels P from k candidate block of pixels, so that block of pixels P meets following item
Part: the boundary of block of pixels extension is as consistent with the value of the corresponding region of high frequency imaging N as possible,
2) splice high frequency imaging N+1 using optimal block of pixels,
3) it repeats step 1) and 2) there is no variation up to resulting image, or reach the maximum times of iteration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611055338.2A CN106780418B (en) | 2016-11-25 | 2016-11-25 | A kind of Image Intensified System |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611055338.2A CN106780418B (en) | 2016-11-25 | 2016-11-25 | A kind of Image Intensified System |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106780418A CN106780418A (en) | 2017-05-31 |
| CN106780418B true CN106780418B (en) | 2019-05-31 |
Family
ID=58913028
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611055338.2A Active CN106780418B (en) | 2016-11-25 | 2016-11-25 | A kind of Image Intensified System |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106780418B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103985085A (en) * | 2014-05-26 | 2014-08-13 | 三星电子（中国）研发中心 | Image super-resolution amplifying method and device |
| CN105590303A (en) * | 2014-10-20 | 2016-05-18 | Tcl集团股份有限公司 | Method and system for increasing image resolution |
-
2016
- 2016-11-25 CN CN201611055338.2A patent/CN106780418B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103985085A (en) * | 2014-05-26 | 2014-08-13 | 三星电子（中国）研发中心 | Image super-resolution amplifying method and device |
| CN105590303A (en) * | 2014-10-20 | 2016-05-18 | Tcl集团股份有限公司 | Method and system for increasing image resolution |
Non-Patent Citations (1)
| Title |
|---|
| 《Image and Video upscaling from local self-examples》;GILAD FREEDMAN 等;《ACM Transactions on Graphics》;20110430;第30卷(第2期);第1-11页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106780418A (en) | 2017-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Deep learning for image super-resolution: A survey | |
| Huang et al. | Wavelet-srnet: A wavelet-based cnn for multi-scale face super resolution | |
| US9734566B2 (en) | Image enhancement using semantic components | |
| Kim et al. | Progressive face super-resolution via attention to facial landmark | |
| US11537873B2 (en) | Processing method and system for convolutional neural network, and storage medium | |
| CN110119780B (en) | Hyper-spectral image super-resolution reconstruction method based on generation countermeasure network | |
| Li et al. | From beginner to master: A survey for deep learning-based single-image super-resolution | |
| Cai et al. | FCSR-GAN: Joint face completion and super-resolution via multi-task learning | |
| Deng et al. | Lau-net: Latitude adaptive upscaling network for omnidirectional image super-resolution | |
| CN106548452B (en) | A kind of image-enhancing equipment and method | |
| CN111696035A (en) | Multi-frame image super-resolution reconstruction method based on optical flow motion estimation algorithm | |
| Guo et al. | Dual-view attention networks for single image super-resolution | |
| Pan et al. | Towards bidirectional arbitrary image rescaling: Joint optimization and cycle idempotence | |
| Guan et al. | Srdgan: learning the noise prior for super resolution with dual generative adversarial networks | |
| CN108876716A (en) | Super resolution ratio reconstruction method and device | |
| Luo et al. | Video super-resolution using multi-scale pyramid 3d convolutional networks | |
| Tarasiewicz et al. | A graph neural network for multiple-image super-resolution | |
| Guo et al. | Deep illumination-enhanced face super-resolution network for low-light images | |
| Gong et al. | Learning deep resonant prior for hyperspectral image super-resolution | |
| Liu et al. | Facial image inpainting using multi-level generative network | |
| Liu et al. | Efficient video super-resolution via hierarchical temporal residual networks | |
| Ding et al. | Single image super-resolution via dynamic lightweight database with local-feature based interpolation | |
| CN106780418B (en) | A kind of Image Intensified System | |
| Wang et al. | Face super-resolution via hierarchical multi-scale residual fusion network | |
| Mengbei et al. | Overview of research on image super-resolution reconstruction |
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 |