CN107085827B - Super-resolution image restoration method based on hardware platform - Google Patents
Super-resolution image restoration method based on hardware platform Download PDFInfo
- Publication number
- CN107085827B CN107085827B CN201710286366.3A CN201710286366A CN107085827B CN 107085827 B CN107085827 B CN 107085827B CN 201710286366 A CN201710286366 A CN 201710286366A CN 107085827 B CN107085827 B CN 107085827B
- Authority
- CN
- China
- Prior art keywords
- result
- image
- iteration
- convolution
- hardware platform
- 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
- 238000000034 method Methods 0.000 title claims abstract description 37
- 239000000872 buffer Substances 0.000 claims description 23
- 238000007667 floating Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 8
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000012827 research and development Methods 0.000 abstract 1
- 238000004364 calculation method Methods 0.000 description 10
- 230000000875 corresponding effect Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000003139 buffering effect Effects 0.000 description 5
- 238000003384 imaging method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- PCTMTFRHKVHKIS-BMFZQQSSSA-N (1s,3r,4e,6e,8e,10e,12e,14e,16e,18s,19r,20r,21s,25r,27r,30r,31r,33s,35r,37s,38r)-3-[(2r,3s,4s,5s,6r)-4-amino-3,5-dihydroxy-6-methyloxan-2-yl]oxy-19,25,27,30,31,33,35,37-octahydroxy-18,20,21-trimethyl-23-oxo-22,39-dioxabicyclo[33.3.1]nonatriaconta-4,6,8,10 Chemical compound C1C=C2C[C@@H](OS(O)(=O)=O)CC[C@]2(C)[C@@H]2[C@@H]1[C@@H]1CC[C@H]([C@H](C)CCCC(C)C)[C@@]1(C)CC2.O[C@H]1[C@@H](N)[C@H](O)[C@@H](C)O[C@H]1O[C@H]1/C=C/C=C/C=C/C=C/C=C/C=C/C=C/[C@H](C)[C@@H](O)[C@@H](C)[C@H](C)OC(=O)C[C@H](O)C[C@H](O)CC[C@@H](O)[C@H](O)C[C@H](O)C[C@](O)(C[C@H](O)[C@H]2C(O)=O)O[C@H]2C1 PCTMTFRHKVHKIS-BMFZQQSSSA-N 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012804 iterative process Methods 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/60—Memory management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
Abstract
The invention discloses a super-resolution image restoration method based on a hardware platform. The method is realized based on a Zynq SOC hardware platform and adopts the idea of parallel processing. Firstly, image expansion is carried out on an original image, then the expanded image is cached into a cache buff, the cache buff is realized by utilizing a Block RAM in SoC, a series of operations in an image restoration algorithm are carried out while caching, and an operation result is cached into the corresponding cache buff at the same time. The invention has the innovative points that the operation steps in the algorithm are not sequentially executed but are executed in parallel, the parallel operation is carried out while caching the operation intermediate value, the execution efficiency of the algorithm is improved, the hardware resources only consume the internal storage resources of the PL part of the Zynq chip, the consumption of the hardware storage resources is greatly reduced, and the invention is beneficial to the research and development of equipment with small size, low power consumption and low cost.
Description
Technical Field
The invention relates to a super-resolution image restoration technology, in particular to a super-resolution image restoration technology realized based on a hardware platform.
Background
At present, various imaging devices are widely applied to various fields, but due to the limitation of various conditions, the imaging quality is often unsatisfactory, so people hope to find a method capable of improving the image quality urgently. Since the optical system is affected by diffraction of the optical system, image information other than the transfer function cutoff frequency cannot be acquired. How image information other than the cutoff frequency can be observed becomes a hot trend of research.
At present, the super-resolution image restoration method mainly comprises the following steps: spectrum extrapolation, POCS, neural network reconstruction, Bayes analysis, and the like. The algorithm with strong super-resolution restoration capability is a Poisson-ML image restoration method based on Bayes analysis, and is called PML algorithm for short. However, the algorithm cannot achieve a good restoration effect on a noisy image. The learners propose a super-resolution image restoration algorithm based on a Poisson-Markov field on the basis of a PML algorithm. The algorithm has strong image restoration capability and good denoising effect. However, the method involves convolution operation, correlation operation and some multiplication and division operations, the amount of calculation is large, and iteration is needed for obtaining a restored image with high quality many times, so that the method implemented by software is not suitable for an imaging system requiring real-time processing. In order to perform image restoration using this method in an imaging system requiring high real-time performance, hardware implementation is required. However, since a large number of intermediate operation values need to be stored in the process of multiple iterations, it is a great challenge to the storage resources of the hardware.
The image restoration method provided by the invention is a Poisson-Markov field-based super-resolution image restoration algorithm realized on a hardware platform, and the parallel processing is performed on partial calculation and the storage of the calculation result by utilizing the parallel processing idea, so that the execution efficiency of the algorithm can be improved, and the consumption of hardware storage resources in the iteration process is greatly reduced. The method is beneficial to the research of low power consumption, miniaturization and low cost of the equipment.
Disclosure of Invention
The invention aims to provide a super-resolution image restoration method based on a hardware platform, which can restore information except the cut-off frequency of an image, enable the image to obtain higher resolution and improve the quality of the image. The image restoration method is realized based on a Zynq hardware platform, the PL part of a Zynq chip completes an image restoration algorithm, the parallel processing mode is adopted to perform parallel processing on the calculation steps in the algorithm, and meanwhile, the storage resources in the PL are used for caching the calculation intermediate value.
The technical solution for realizing the purpose of the invention is as follows: a super-resolution image restoration method based on a hardware platform is realized by describing a mathematical model as formula (1), wherein an operator is a convolution operator, and the operator is a convolution operator
H is convolution kernel, g (i, j) is expanded image, fn-1(i, j) is the result of the (n-1) th iteration, fnAnd (i, j) is the result of the nth iteration, n is the iteration number, i, j is the image row and column index, A, B is the floating point integer expansion coefficient, generally A can be 14, and B can be 10.
According to the formula (1), the image restoration method comprises the steps of image expansion, convolution operation and related operation, and the steps of image expansion, convolution operation and related operation are executed in parallel; wherein the content of the first and second substances,
the image expanding step comprises the following steps: expanding the edge of the original image according to the size of the convolution kernel function template; caching the expanded image data stream into an image caching space;
the convolution operation step comprises:
(a1) reading a last iteration result from an iteration result cache space, and performing convolution operation on the last iteration result and a convolution kernel function; for the first iteration, the last iteration result is an expanded image;
(a2) dividing the expanded image with the result obtained in the step (a1) to obtain a convolution operation result of the round, and caching the convolution operation result into a convolution result cache space;
the correlation operation step comprises:
(b1) reading a convolution operation result corresponding to the current iteration from a convolution result cache space, and performing correlation operation on the convolution operation result and a convolution kernel function;
(b2) reading the last iteration result from the iteration result cache space, multiplying the last iteration result with the result obtained in the step (b1) to obtain the iteration result of the current round, and caching the iteration result into the iteration result cache space; for the first iteration, the last iteration result is an extended image.
Preferably, in the image expanding step, for a convolution kernel function template with the size of M × M, expanding the four edges of the original image by M-1 pixels respectively.
Preferably, the image buffer space size is ((N-1) × (M +1) +2) × (W + M-1), the convolution result buffer space size is N × (M +1) × (W + M-1), and the iteration result buffer space size is N × (M +2) × (W + M-1), where W is the width of the expanded image and N is the set total number of iterations.
Compared with the prior art, the invention has the following remarkable advantages: 1. the method improves the operation efficiency of the algorithm and can be used for real-time processing of the infrared video images. 2. The storage resources consumed by hardware are greatly reduced, so that the internal storage resources of the PL part of the Zynq chip can meet the storage requirement of the algorithm, and the low-power consumption, miniaturization and low-cost design of equipment is facilitated. The PL part of the Zynq chip is essentially FPGA, so the invention is not only suitable for the Zynq hardware platform, but also suitable for the FPGA hardware platform.
Drawings
FIG. 1 is a schematic diagram of an image expansion of the present invention.
FIG. 2 is a schematic diagram of a buffer Buff _ g according to the present invention, which is used for buffering the extended image.
FIG. 3 is a schematic diagram of the buffer Buff _ b (n-1) of the present invention, which is used for buffering the convolution result of the (n-1) th iteration.
FIG. 4 is a schematic diagram of buffer Buff _ b (n) for buffering convolution results of the nth iteration according to the present invention.
FIG. 5 is a schematic diagram of the buffer Buff _ a (n-1) according to the present invention, which is used for buffering the (n-1) th iteration result.
FIG. 6 is a schematic diagram of Buff _ a (n) buffers according to the present invention, for buffering the nth iteration result.
FIG. 7 is a schematic flow chart of the method of the present invention.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings and specific embodiments.
The invention relates to a super-resolution image restoration method based on a hardware platform, which has the following principle: in the process of acquiring the video image, the image quality cannot meet the observation requirement of people due to the influence of various factors. The super-resolution image restoration algorithm can restore information except the cut-off frequency of the image, so that higher image resolution is obtained, and the details of the image are enhanced. The arithmetic mathematical model is as formula (1):
n is iteration number, the larger the value of n is, the more the iteration number is, the better the quality of the obtained restored image is, but the larger the value of n is, the larger the calculation amount is, when the restoration effect verification is performed on Matlab by an algorithm, it can be found that a infrared image with the resolution of 320 x 240 can obtain higher image quality after being iterated for 20 times. In order to meet the requirement of real-time processing of video images, the restoration algorithm needs to be realized on a hardware platform, if the convolution operation is performed on a whole image and then the correlation operation is performed, a large amount of hardware resources are consumed for 20 times of iteration, and the internal storage resources of hardware cannot meet the storage requirement of calculating an intermediate value in the iteration process. Therefore, the invention utilizes the idea of parallel execution, processes the convolution operation and the related operation in each iteration in parallel, and caches only a part of the calculated intermediate values, thereby improving the operation efficiency and greatly reducing the consumption of hardware storage resources.
The invention relates to a super-resolution image restoration method based on a hardware platform, wherein a hardware implementation flow chart is shown in fig. 7, and main modules of the method comprise: the device comprises an image expanding module, a convolution operation module and a correlation operation module. The image expansion module is used for expanding the original image; the convolution operation module is used for convolution operation and corresponding multiplication and division operation in the iterative process; and the correlation operation module is used for correlation operation and corresponding multiplication and division operation in the iterative process. The three modules are processed in parallel.
As shown in fig. 7, a super resolution image restoration method implemented based on a hardware platform according to an embodiment of the present invention includes the following steps:
step 1: and (5) expanding the image. The specific implementation method is as shown in fig. 1, an original image with a resolution of 320 × 240 is in a black frame, the size of the kernel function template is 11 × 11, the size of the original image to be expanded is (11-1) × 2 ═ 20, that is, the edge of the original image needs to be expanded by 10 pixels each, the ABCD portion in the black frame in fig. 1 is the edge of the original image, and the ABCD outside the black frame is 10 pixels expanded by using the corresponding edge of the original image. The expanded image data stream is cached into a cache space Buff _ g, and the space size of Buff _ g is 350 × 230.
Step 2: performing convolution operation on the iteration result of the (n-1) th time, wherein the calculation formula is formula (2), wherein fn-1(i, j) is the result of the (n-1) th iteration, and h is the convolution kernel.
In the first iteration, fn-1And (i, j) is an expanded image g (i, j), so that while expanded image data is cached into the Buff _ g, the convolution kernel function h and the data in the Buff _ g perform convolution operation of the first iteration, and a convolution operation result is obtained before step 3 is executed.
In the nth iteration, fn-1And (i, j) is the result of the (n-1) th iteration operation, the result is stored in Buff _ a (n-1), when the (n-1) th iteration result is cached in Buff _ a (n-1), the convolution kernel function h simultaneously carries out the (n) th convolution operation with the data in Buff _ a (n-1), and step 3 is executed after the result of the convolution operation is obtained.
fn-1(i,j)*(h×214) Formula (2)
And step 3: and (3) dividing the expanded image by the nth convolution result, wherein the calculation formula is formula (3), and caching the calculated result into Buff _ b (n), wherein g (i, j) is the expanded image.
And 4, step 4: while the result obtained in step 3 is cached in Buff _ b (n), the data in Buff _ b (n) is correlated with the convolution kernel h, and the calculation formula is formula (4).
And 5: and (3) multiplying the n-1 th iteration result by the result obtained in the step (4), wherein the calculation formula is formula (5), and the calculated result is cached into Buff _ a (n), namely the nth iteration result.
The step 1 belongs to an image expansion module, the steps 2 and 3 belong to a convolution operation module, and the steps 4 and 5 belong to a correlation operation module. The three modules are executed in parallel.
The invention uses the internal storage resources of the PL part of the Zynq chip as a buffer space, which is called buff space. FIGS. 2-6 illustrate buff involved in the above steps, respectively. The BUFF _ g in fig. 2 is used to buffer the buffer BUFF of the extended image, and the yellow part is the kernel h and is also an 11 × 11 kernel template. Buff _ b (n-1) in FIG. 3 is used to buffer the convolution result in the (n-1) th iteration of a frame image. Buff _ b (n) in FIG. 4, used to buffer the convolution operation result in the nth iteration of a frame of image. Buff _ a (n-1) in FIG. 5, is used to buffer the (n-1) th iteration result of a frame image. Buff _ a (n) in fig. 6, which is used to buffer the nth iteration result of one frame of image.
The specific implementation steps of performing the first iteration on one frame image are as follows:
and S1, when the expanded image data stream is stored in BUFF _ g, carrying out convolution operation of first iteration on the kernel function h and data in BUFF _ g from left to right, executing a formula (6), executing a formula (7) on the obtained convolution operation result, and simultaneously storing the operation result in Buff _ b (1).
fn(i,j)*(h×214) Formula (6)
And S2, performing correlation operation on the kernel function h and the data in the Buff _ b (1) from left to right while performing convolution operation, executing a formula (8), and performing a formula (9) on the obtained correlation operation result, wherein the operation result is stored in the Buff _ a (1) at the same time. In the first iteration, f in the formulan(i, j) is an extended image g (i, j).
The (n) th iteration of a frame image is implemented by the following steps:
s1: when the (n-1) th iteration result is stored in Buff _ a (n-1), the kernel function h performs convolution operation result of the (n) th iteration with Buff _ a (n-1) from left to right, and the formula (10) is executed
fn-1(i,j)*(h×214) The convolution operation result obtained by the formula (10) is executed by the formula (11)
The result of the operation is buffered by line _ Buff and also stored in Buff _ b (n).
S2: and (3) carrying out convolution operation and correlation operation on the kernel function h and the data in the Buff _ b (n) from left to right at the same time, executing a formula (12), executing a formula (13) by using the obtained correlation operation result, and simultaneously storing the operation result into the Buff _ a (n).
The required size of the buffer BUFF _ g is 230 × 350. The buffer Buff _ a (n-1) and the buffer Buff _ a (n) are buffers corresponding to the (n-1) th iteration and the nth iteration in Buff _ a respectively, the required size of Buff _ a is 20 × 12 × 350, the buffer Buff _ b (n-1) and the buffer Buff _ b (n) are buffers corresponding to the (n-1) th iteration and the nth iteration in Buff _ b respectively, and the required size of Buff _ b is 20 × 13 × 350.
Claims (8)
1. A super-resolution image restoration method based on a hardware platform improves the resolution of an image by expanding and iterating for many times on the basis of an original image, each iteration comprises convolution operation and related operation steps, and the method is characterized in that: the image expansion, convolution operation and related operation steps are executed in parallel;
the image expanding step comprises the following steps: expanding the edge of the original image according to the size of the convolution kernel function template; caching the expanded image data stream into an image caching space;
the convolution operation step includes:
(a1) reading a last iteration result from an iteration result cache space, and performing convolution operation on the last iteration result and a convolution kernel function; for the first iteration, the last iteration result is an expanded image;
(a2) dividing the expanded image with the result obtained in the step (a1) to obtain a convolution operation result of the round, and caching the convolution operation result into a convolution result cache space;
the correlation operation step comprises:
(b1) reading a convolution operation result corresponding to the current iteration from a convolution result cache space, and performing correlation operation on the convolution operation result and a convolution kernel function;
(b2) reading the last iteration result from the iteration result cache space, multiplying the last iteration result with the result obtained in the step (b1) to obtain the iteration result of the current round, and caching the iteration result into the iteration result cache space; for the first iteration, the last iteration result is an extended image.
2. The super resolution image restoration method based on the hardware platform implementation of claim 1, wherein: in the image expanding step, for a convolution kernel function template with the size of M multiplied by M, expanding M-1 pixels on four edges of an original image respectively.
3. The super resolution image restoration method based on the hardware platform implementation of claim 2, wherein:
the size of the image cache space is ((N-1) × (M +1) +2) × (W + M-1), the size of the convolution result cache space is Nx (M +1) × (W + M-1), and the size of the iteration result cache space is Nx (M +2) × (W + M-1), wherein W is the width of the expanded image, and N is the set total iteration number.
4. The super resolution image restoration method based on the hardware platform implementation of claim 1, wherein: step (a1) according to the formula fn-1(i，j)*(h×2A) The last iteration result fn-1And (i, j) performing convolution operation on the (i, j) and a convolution kernel function h, wherein n is iteration times, i, j is row and column indexes of the image, and A is a floating point number integer expansion coefficient.
5. The super resolution image restoration method based on the hardware platform implementation of claim 4, wherein: step (a2) according to the formula
Combining the expansion image g (i, j) with the result f obtained in the step (a1)n-1(i，j)*(h×2A) And dividing, wherein B is a floating-point integer expansion coefficient.
6. The super resolution image restoration method based on the hardware platform implementation of claim 5, wherein: step (b1) according to the formula
The convolution operation result
And performing correlation operation with the convolution kernel function h.
7. The super resolution image restoration method based on the hardware platform implementation of claim 6, wherein: step (b2) according to the formula
The last iteration result fn-1(i, j) and the result obtained in step (b1)
Multiplication.
8. The super resolution image restoration method based on the hardware platform implementation of claim 1, wherein: internal storage resources of the PL part of the Zynq chip are used as buffer space.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710286366.3A CN107085827B (en) | 2017-04-27 | 2017-04-27 | Super-resolution image restoration method based on hardware platform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710286366.3A CN107085827B (en) | 2017-04-27 | 2017-04-27 | Super-resolution image restoration method based on hardware platform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107085827A CN107085827A (en) | 2017-08-22 |
| CN107085827B true CN107085827B (en) | 2020-06-16 |
Family
ID=59611552
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710286366.3A Active CN107085827B (en) | 2017-04-27 | 2017-04-27 | Super-resolution image restoration method based on hardware platform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107085827B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919295B (en) * | 2017-12-12 | 2022-10-28 | 北京大学深圳研究生院 | Embedded audio event detection method based on lightweight convolutional neural network |
| CN110288518B (en) * | 2019-06-28 | 2020-10-23 | 北京三快在线科技有限公司 | Image processing method, device, terminal and storage medium |
| CN112906567A (en) | 2020-05-05 | 2021-06-04 | 王建华 | Fingerprint sensing device and fingerprint identification method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1398118A (en) * | 2002-08-16 | 2003-02-19 | 清华大学 | Method based on slide window for estimating and equalizing channels of block signals containing pilot |
| CN101261176A (en) * | 2008-04-03 | 2008-09-10 | 华中科技大学 | Sequence image correction based pneumatic optical transmission effect evaluation method and apparatus |
| CN101949768A (en) * | 2010-08-20 | 2011-01-19 | 中国科学院光电技术研究所 | Processor of Hartmann -Shack front slope relative to point target and manufacture method |
| CN102197369A (en) * | 2008-10-08 | 2011-09-21 | Arm有限公司 | Apparatus and method for performing SIMD multiply-accumulate operations |
| CN105260773A (en) * | 2015-09-18 | 2016-01-20 | 华为技术有限公司 | Image processing device and image processing method |
| CN105814874A (en) * | 2013-12-11 | 2016-07-27 | 来特荣国际有限公司 | Super-resolution processing method for tv video images, super-resolution processing device for tv video images that is used in same method, first to fourteenth super-resolution processing programs, and first to fourth storage media |
| CN105869117A (en) * | 2016-03-28 | 2016-08-17 | 上海交通大学 | Method for accelerating GPU directed at deep learning super-resolution technology |
| CN105874775A (en) * | 2014-02-03 | 2016-08-17 | 来特荣国际有限公司 | Super resolution processing method, device, and program for single interaction multiple data (SIMD) type super parallel computation processing device, and storage medium |
-
2017
- 2017-04-27 CN CN201710286366.3A patent/CN107085827B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1398118A (en) * | 2002-08-16 | 2003-02-19 | 清华大学 | Method based on slide window for estimating and equalizing channels of block signals containing pilot |
| CN101261176A (en) * | 2008-04-03 | 2008-09-10 | 华中科技大学 | Sequence image correction based pneumatic optical transmission effect evaluation method and apparatus |
| CN102197369A (en) * | 2008-10-08 | 2011-09-21 | Arm有限公司 | Apparatus and method for performing SIMD multiply-accumulate operations |
| CN101949768A (en) * | 2010-08-20 | 2011-01-19 | 中国科学院光电技术研究所 | Processor of Hartmann -Shack front slope relative to point target and manufacture method |
| CN105814874A (en) * | 2013-12-11 | 2016-07-27 | 来特荣国际有限公司 | Super-resolution processing method for tv video images, super-resolution processing device for tv video images that is used in same method, first to fourteenth super-resolution processing programs, and first to fourth storage media |
| CN105874775A (en) * | 2014-02-03 | 2016-08-17 | 来特荣国际有限公司 | Super resolution processing method, device, and program for single interaction multiple data (SIMD) type super parallel computation processing device, and storage medium |
| CN105260773A (en) * | 2015-09-18 | 2016-01-20 | 华为技术有限公司 | Image processing device and image processing method |
| CN105869117A (en) * | 2016-03-28 | 2016-08-17 | 上海交通大学 | Method for accelerating GPU directed at deep learning super-resolution technology |
Non-Patent Citations (3)
| Title |
|---|
| A Multiwindow Partial Buffering Scheme for FPGA-Based 2-D Convolvers;Hui Zhang et al;《IEEE Transactions on Circuits and Systems II: Express Briefs 》;20070228;第54卷(第2期);第200-204页 * |
| 基于Poisson-Markov场的超分辨力图像复原算法;苏秉华 等;《电子学报》;20030131;第31卷(第1期);第41-44页 * |
| 基于复合算子的红外复原图像弱小目标检测算法;杜瀚宇 等;《指挥信息系统与技术》;20170228;第8卷(第1期);第70-74页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107085827A (en) | 2017-08-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108537733B (en) | Super-resolution reconstruction method based on multi-path deep convolutional neural network | |
| Liu et al. | Image restoration using total variation with overlapping group sparsity | |
| Xu et al. | A distributed canny edge detector: algorithm and FPGA implementation | |
| Mahmoud et al. | Diffy: A Déjà vu-free differential deep neural network accelerator | |
| CN107085827B (en) | Super-resolution image restoration method based on hardware platform | |
| CN107274362B (en) | Optimization system and method for realizing oriented filtering by hardware | |
| CN112884851A (en) | Deep compression sensing network for expanding iterative optimization algorithm | |
| CN108765282B (en) | Real-time super-resolution method and system based on FPGA | |
| Zhao et al. | Comprehensive and delicate: An efficient transformer for image restoration | |
| Tan et al. | A real-time video denoising algorithm with FPGA implementation for Poisson–Gaussian noise | |
| Xiao et al. | FPGA-based scalable and highly concurrent convolutional neural network acceleration | |
| CN112966807A (en) | Convolutional neural network implementation method based on storage resource limited FPGA | |
| CN102970545A (en) | Static image compression method based on two-dimensional discrete wavelet transform algorithm | |
| CN113674172A (en) | Image processing method, system, device and storage medium | |
| CN104683818A (en) | Image compression method based on biorthogonal invariant set multi-wavelets | |
| CN106558021A (en) | Video enhancement method based on super-resolution technique | |
| CN111626943A (en) | Total variation image denoising method based on first-order forward and backward algorithm | |
| CN107680126B (en) | Random sampling consistency image matching denoising processing system and method | |
| Zhang et al. | Memory-efficient high-speed VLSI implementation of multi-level discrete wavelet transform | |
| Bevara et al. | VLSI implementation of high throughput parallel pipeline median finder for IoT applications | |
| Peng et al. | Low-light image enhancement based on FPGA | |
| CN111598781A (en) | Image super-resolution method based on hybrid high-order attention network | |
| SP et al. | Evaluating Winograd Algorithm for Convolution Neural Network using Verilog | |
| Gao et al. | A non-local means denoising co-processor with data reuse scheme and dual-clock domain for high resolution image sensor | |
| Duan et al. | An Energy-Efficient Image Denoising Accelerator with Depth-wise Separable Convolution and Fused-Layer Architecture |
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 |