CN102903087B - Based on the SAR image denoising method of GPU programming - Google Patents
Based on the SAR image denoising method of GPU programming Download PDFInfo
- Publication number
- CN102903087B CN102903087B CN201210414683.6A CN201210414683A CN102903087B CN 102903087 B CN102903087 B CN 102903087B CN 201210414683 A CN201210414683 A CN 201210414683A CN 102903087 B CN102903087 B CN 102903087B
- Authority
- CN
- China
- Prior art keywords
- image
- denoising
- pixel
- process unit
- graphic process
- 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
Landscapes
- Image Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a kind of SAR image denoising method based on GPU programming, mainly solve prior art and the slower problem of denoising hourly velocity is carried out to extensive SAR image.Implementation step is: (1) is treated denoising image I and carried out continuation and obtain two noisy images; (2) region of memory is set in graphic process unit GPU, will treats that denoising image I and two noisy image is stored in the internal memory of graphic process unit GPU; (3) in graphic process unit GPU, use multiple thread parallel to calculate the weights treated between each pixel in denoising image I and other pixels; (4) exploitation right value matrix, calculates the retroreflection coefficient treating each pixel in denoising image I; (5) extraction of square root operation is carried out to retroreflection matrix of coefficients, obtain final denoising result.The present invention compared with prior art, improves the speed of extensive SAR image being carried out to PPB denoising, meets the requirement of real-time to pictures subsequent process, can be used for carrying out quick denoising to extensive SAR image.
Description
Technical field
The present invention relates to image processing field, be specifically related to the denoising of SAR image, can be used for the processing speed improving image denoising.
Background technology
In recent years, domestic and international researchist has done a large amount of correlative study work in image denoising field, propose the method for a lot of image denoising, in these methods, great majority research is additive white Gaussian noise model, and remaining is then design for LDPC code specially.Almost seldom there is denoising method can be applicable to different noise models.Such as:
Neighborhood Filtering method, utilize the difference between grey scale pixel value to define the weights between pixel in neighborhood and central pixel point, edge and the texture information of image can be kept while denoising, one of problem of the method be the single pixel being subject to noise gray-scale value between difference often can accurately not embody similarity between real pixel.Secondly, the pixel similar to a certain pixel differs in the neighborhood of establishing a capital and being present in around current pixel, also may be present in whole image space.
The non-local mean Image denoising algorithm proposed by Buades etc. is a significant improvement to traditional Neighborhood Filtering method.First, it considers the Self-similar Feature of image, breaches the restriction that Neighborhood Filtering only carries out part filter, can find similar pixel in the larger context.Secondly, similar pixel is defined as the pixel with same vicinity pattern by the method, utilize the information in the window of fixed size around pixel to represent the feature of this pixel, this method for expressing is more reliable than the similarity only utilizing the information of single pixel itself to obtain.
PPB Image denoising algorithm is an expansion to non-local mean Image denoising algorithm, this method proposes one and depends on the more general of noise profile model and have more the similarity criterion of statistics according to property.It is a maximum weight possibility predication problem the procedure definition of denoising, the mode of weights data-driven here draws, namely these weights can be improved according to the similarity between the noise block obtained in the similarity between noise block and front once estimation by the mode of iteration.The method especially obviously improves the effect of image denoising when processing the SAR image of low signal-to-noise ratio processing noise image.
Because PPB Image denoising algorithm has good edge hold facility, embody obvious superiority when denoising being carried out to SAR image simultaneously, be therefore widely applied.But existing Image denoising algorithm is all generally serial algorithm, along with the resolution of remote sensing images improves constantly, image data amount is also increasing, when processing these super large images, the greatest drawback of this method is that speed is slow, can not meet the requirement of real-time to pictures subsequent process.
Summary of the invention
The object of the invention is to overcome above-mentioned problematic shortcoming, propose a kind of SAR image denoising method based on GPU programming, to improve the speed of extensive SAR image being carried out to PPB denoising, meet the requirement of real-time to pictures subsequent process.
Technical scheme of the present invention is that the calculating Unified Device framework CUDA utilizing NVIDIA company to propose realizes, and concrete steps are as follows:
(1) treat denoising image I and carry out continuation;
1a) treat denoising image I according to dimension D=2 to surrounding continuation, obtain the first noisy image I after continuation
1;
1b) treat denoising image I according to size W=7 to surrounding continuation, obtain the second noisy image I after continuation
2;
(2) region of memory A is set in graphic process unit GPU, denoising image I, the first noisy image I will be treated respectively
1with the second noisy image I
2be stored in the internal memory A of graphic process unit GPU;
(3) weight w (s, t) treated between each pixel in denoising image I and other pixels is calculated:
3a) region of memory B is set in graphic process unit GPU;
3b) utilize the in the internal memory A of processor GPU first noisy image I
1with the second noisy image I
2, in graphic process unit GPU, the middle weights R treated in denoising image I between each pixel and other pixels is calculated by multiple thread parallel
s, t:
Wherein, A
s,krepresent the pixel value of a kth adjoint point in the search window centered by pixel s, k ∈ [0, m], m are maximum iteration time;
3c) by step 3b) the middle weight matrix that obtains is stored in the region of memory of CPU, to each element in middle weight matrix successively by row with by column count prefix and, complete the renewal to middle weight matrix;
3d) by step 3c) the middle weight matrix that obtains is stored in the internal memory B of graphic process unit GPU, calculated the weight w (s, t) treated between each pixel in denoising image I and other pixels by multiple thread parallel in graphic process unit GPU:
Wherein, R
s, tfor treating the middle weights in denoising image I between pixel s and pixel t, h is weights coefficient, and exp () represents exponential function;
3e) circulation performs step 3b) ~ step 3d), until k=m, then weight matrix is stored in the internal memory of CPU;
(4) step 3e is utilized) the weight matrix w (s, t) that obtains, calculate the retroreflection coefficient treating each pixel in denoising image I
Wherein, A
trepresent the pixel value treating pixel t in denoising image I;
(5) extraction of square root operation is carried out to the retroreflection matrix of coefficients that step (4) obtains, final denoising result can be obtained.
The present invention has the following advantages compared with prior art:
1. the present invention owing to arranging region of memory in graphic process unit GPU, and this region of memory is set to a page locking page in memory, substantially increase the speed reading and writing image in the internal memory of graphic process unit GPU, indirectly improve the speed of extensive SAR image being carried out to denoising.
2. the present invention calculates owing to using multiple thread parallel in graphic process unit GPU the weights treated in denoising image I between each pixel and other pixels, greatly saves computing time, meets the requirement of real-time to pictures subsequent process.
Accompanying drawing explanation
Fig. 1 is realization flow figure of the present invention;
Fig. 2 is the piece image that the present invention emulates;
Fig. 3 is the second width image that the present invention emulates;
Embodiment
With reference to Fig. 1, performing step of the present invention comprises as follows:
Step 1. is treated denoising image I and is carried out continuation;
1a) treat denoising image I according to dimension D=2 to surrounding continuation, obtain the first noisy image I after continuation
1;
1b) treat denoising image I according to size W=7 to surrounding continuation, obtain the second noisy image I after continuation
2.
Step 2. arranges region of memory A in graphic process unit GPU, and invoke memory copy function cudaHostAlloc () will treat denoising image I, the first noisy image I respectively
1with the second noisy image I
2be stored in the internal memory A of graphic process unit GPU.
Step 3. calculates the weight w (s, t) treated between each pixel in denoising image I and other pixels:
3a) region of memory B is set in graphic process unit GPU;
3b) utilize the in the internal memory A of processor GPU first noisy image I
1with the second noisy image I
2, in graphic process unit GPU, the middle weights R treated in denoising image I between each pixel and other pixels is calculated by multiple thread parallel
s, t:
Wherein, A
s, krepresent the pixel value of a kth adjoint point in the search window centered by pixel s, k ∈ [0, m], m is maximum iteration time, realize the synchronous of multiple thread in graphic process unit GPU in ending place invokes thread synchronous function cudaThreadSynchronize () of each thread, namely ensure that in graphic process unit GPU, all threads all preserve result of calculation again according to after above-mentioned formulae discovery;
3c) by step 3b) the middle weight matrix that obtains is stored in the region of memory of CPU, to each element in middle weight matrix successively by row with by column count prefix and, complete the renewal to middle weight matrix;
3d) by step 3c) the middle weight matrix that obtains is stored in the internal memory B of graphic process unit GPU, calculated the weight w (s, t) treated between each pixel in denoising image I and other pixels by multiple thread parallel in graphic process unit GPU:
Wherein, R
s, tfor treating the middle weights in denoising image I between pixel s and pixel t, h is weights coefficient, exp () represents exponential function, realize the synchronous of multiple thread in graphic process unit GPU in ending place invokes thread synchronous function cudaThreadSynchronize () of each thread, namely ensure that in graphic process unit GPU, all threads all preserve result of calculation again according to after above-mentioned formulae discovery;
3e) circulation performs step 3b) ~ step 3d), until k=m, then weight matrix is stored in the internal memory of CPU.
Step 4. utilizes step 3e) the weight matrix w (s, t) that obtains, calculate the retroreflection coefficient treating each pixel in denoising image I
Wherein, A
trepresent the pixel value treating pixel t in denoising image I.
The retroreflection matrix of coefficients that step 5. pair step 4 obtains carries out extraction of square root operation, can obtain final denoising result.
Effect of the present invention can be further illustrated by following emulation experiment:
1. simulated conditions
In order to verify the superiority of the inventive method, the inventive method and prior art are carried out quantitative test to the computing time that SAR image is carried out needed for PPB denoising by one group of emulation experiment by us.
The test platform of emulation experiment is:
CPU:Six-Core AMD Operon (tm); Internal memory: 32G; Video card: NVIDIA Quadro FX 4800; Video card video memory: 1.5G.
2 emulation content and results
Emulation experiment 1, by method of the present invention and prior art, the SAR image to the area, airport shown in Fig. 2 carries out PPB denoising respectively, and does quantitative test to computing time, and image size is 2000 × 2000, and Comparative result is as shown in table 1.
Emulation experiment 2, by method of the present invention and prior art, the SAR image to the area portion region, airport shown in Fig. 3 carries out PPB denoising respectively, and does quantitative test to computing time, and image size is 300 × 300, and Comparative result is as shown in table 2.
Table 1 two kinds of methods carry out PPB denoising contrast computing time to area, airport SAR image
As can be seen from Table 1, the inventive method substantially increases the speed of extensive SAR image being carried out to PPB denoising, solves prior art and carries out the slow-footed problem of PPB denoising to extensive SAR image, meet the requirement of real-time to pictures subsequent process.
Table 2 two kinds of methods carry out PPB denoising contrast computing time to area portion region, airport SAR image
Even if the inventive method is carried out PPB denoising for the SAR image of small-scale and has also been embodied advantage in speed as can be seen from Table 2.
Compared by the data by table 1 and table 2 and be not difficult to draw the following conclusions: the inventive method calculates owing to using multiple thread parallel in graphic process unit GPU, improve the speed of extensive SAR image being carried out to PPB denoising, meet the requirement of real-time to pictures subsequent process.
Claims (3)
1., based on a SAR image denoising method for GPU programming, comprise the following steps:
(1) treat denoising image I and carry out continuation;
1a) treating denoising image I according to being of a size of 2 to surrounding continuation, obtaining the first noisy image I after continuation
1;
1b) treating denoising image I according to being of a size of 7 to surrounding continuation, obtaining the second noisy image I after continuation
2;
(2) region of memory A is set in graphic process unit GPU, denoising image I, the first noisy image I will be treated respectively
1with the second noisy image I
2be stored in the internal memory A of graphic process unit GPU;
(3) weight w (s, t) treated between each pixel in denoising image I and other pixels is calculated:
3a) region of memory B is set in graphic process unit GPU;
3b) utilize the in the internal memory A of processor GPU first noisy image I
1with the second noisy image I
2, in graphic process unit GPU, the middle weights R treated in denoising image I between each pixel and other pixels is calculated by multiple thread parallel
s,t:
Wherein, A
s,krepresent the pixel value of a kth adjoint point in the search window centered by pixel s, k ∈ [0, m], m are maximum iteration time;
Realize the synchronous of multiple thread in graphic process unit GPU in ending place invokes thread synchronous function cudaThreadSynchronize () of each thread, namely ensure that in graphic process unit GPU, all threads are all according to above-mentioned middle weights R
s,tresult of calculation is preserved again after formulae discovery;
3c) by step 3b) the middle weight matrix that obtains is stored in the region of memory of CPU, to each element in middle weight matrix successively by row with by column count prefix and, complete the renewal to middle weight matrix;
3d) by step 3c) the middle weight matrix that obtains is stored in the internal memory B of graphic process unit GPU, calculated the weight w (s, t) treated between each pixel in denoising image I and other pixels by multiple thread parallel in graphic process unit GPU:
Wherein, R
s,tfor treating the middle weights in denoising image I between pixel s and pixel t, h is weights coefficient, and exp () represents exponential function;
3e) circulation performs step 3b) ~ step 3d), until k=m, then weight matrix is stored in the internal memory of CPU;
(4) step 3e is utilized) the weight matrix w (s, t) that obtains, calculate the retroreflection coefficient treating each pixel in denoising image I
Wherein, A
trepresent the pixel value treating pixel t in denoising image I;
(5) extraction of square root operation is carried out to the retroreflection matrix of coefficients that step (4) obtains, final denoising result can be obtained.
2. the SAR image denoising method based on GPU programming according to claim 1, the denoising image I that treats described in step (1) carries out continuation, is complete in page locking page in memory.
3. the SAR image denoising method based on GPU programming according to claim 1, will treat denoising image I, the first noisy image I described in step (2) respectively
1with the second noisy image I
2being stored in the internal memory of graphic process unit GPU, is that invoke memory copy function cudaHostAlloc () realizes.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210414683.6A CN102903087B (en) | 2012-10-25 | 2012-10-25 | Based on the SAR image denoising method of GPU programming |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210414683.6A CN102903087B (en) | 2012-10-25 | 2012-10-25 | Based on the SAR image denoising method of GPU programming |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102903087A CN102903087A (en) | 2013-01-30 |
CN102903087B true CN102903087B (en) | 2015-09-30 |
Family
ID=47575301
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210414683.6A Active CN102903087B (en) | 2012-10-25 | 2012-10-25 | Based on the SAR image denoising method of GPU programming |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102903087B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104240199A (en) * | 2014-09-02 | 2014-12-24 | 苏州阔地网络科技有限公司 | Method and device for eliminating noise of image |
CN104751453A (en) * | 2015-03-11 | 2015-07-01 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image changing and detecting method based on CUDA (Compute Unified Device Architecture) and steady WT (Wavelet Transform) |
CN104992421B (en) * | 2015-07-09 | 2018-08-17 | 西安电子科技大学 | A kind of parallel optimization method of the Image denoising algorithm based on OpenCL |
CN111062878B (en) * | 2019-10-31 | 2023-04-18 | 深圳先进技术研究院 | Image denoising method and device and computer readable storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101482617A (en) * | 2009-01-16 | 2009-07-15 | 西安电子科技大学 | Synthetic aperture radar image denoising method based on non-down sampling profile wave |
CN102073982A (en) * | 2011-01-10 | 2011-05-25 | 西安电子科技大学 | Method for realizing acceleration of anisotropic diffusion filtration of overlarge synthetic aperture radar (SAR) image by graphic processing unit (GPU) |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8497865B2 (en) * | 2006-12-31 | 2013-07-30 | Lucid Information Technology, Ltd. | Parallel graphics system employing multiple graphics processing pipelines with multiple graphics processing units (GPUS) and supporting an object division mode of parallel graphics processing using programmable pixel or vertex processing resources provided with the GPUS |
-
2012
- 2012-10-25 CN CN201210414683.6A patent/CN102903087B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101482617A (en) * | 2009-01-16 | 2009-07-15 | 西安电子科技大学 | Synthetic aperture radar image denoising method based on non-down sampling profile wave |
CN102073982A (en) * | 2011-01-10 | 2011-05-25 | 西安电子科技大学 | Method for realizing acceleration of anisotropic diffusion filtration of overlarge synthetic aperture radar (SAR) image by graphic processing unit (GPU) |
Non-Patent Citations (4)
Title |
---|
CUDA架构下的快速图像去噪;李军等;《计算机工程与应用》;20091231;第45卷(第11期);第183-185、222页 * |
GPU-based Framework for Interactive Visualization of SAR Data;Martin Lambers等;《Geoscience and Remote Sensing Symposium,2007》;20070728;第4076-4079页 * |
Iterative Weighted Maximum Likelihood Denoising with Probabilistic Patch-Based Weights;Charles-Alban Deledalle等;《IEEE TRANSACTION ON IMAGE PROCESSING》;20091231;第18卷(第12期);第1-12页 * |
利用CUDA实现的基于GPU的SAR成像算法;柳彬等;《信息技术》;20091231(第11期);第63-65页 * |
Also Published As
Publication number | Publication date |
---|---|
CN102903087A (en) | 2013-01-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102117482B (en) | Non-local mean image denoising method combined with structure information | |
CN103093433B (en) | Natural image denoising method based on regionalism and dictionary learning | |
CN110084238B (en) | Finger vein image segmentation method and device based on LadderNet network and storage medium | |
CN102903087B (en) | Based on the SAR image denoising method of GPU programming | |
CN107085840B (en) | Based on partial fractional differential graph of equation as denoising method | |
CN105046672A (en) | Method for image super-resolution reconstruction | |
CN103854262A (en) | Medical image noise reduction method based on structure clustering and sparse dictionary learning | |
CN103279959B (en) | A kind of two-dimension analysis sparse model, its dictionary training method and image de-noising method | |
CN102821228B (en) | Low-rank video background reconstructing method | |
CN105427264A (en) | Image reconstruction method based on group sparsity coefficient estimation | |
CN103208104B (en) | A kind of image de-noising method based on nonlocal theory | |
CN103761717B (en) | A kind of urban water-body extracting method based on panchromatic remote sensing image | |
CN104899906A (en) | Magnetic resonance image reconstruction method based on adaptive orthogonal basis | |
CN104252704A (en) | Total generalized variation-based infrared image multi-sensor super-resolution reconstruction method | |
CN103971354A (en) | Method for reconstructing low-resolution infrared image into high-resolution infrared image | |
CN107133938A (en) | Robust image fusion method based on wavelet coefficient statistical model | |
Wang et al. | Intelligent nonconvex compressive sensing using prior information for image reconstruction by sparse representation | |
CN104463822A (en) | Multi-focus image fusing method and device based on multi-scale overall filtering | |
CN106940904A (en) | Attendance checking system based on recognition of face and speech recognition | |
Ono et al. | Image recovery by decomposition with component-wise regularization | |
CN102693530B (en) | Synthetic aperture radar (SAR) image despeckle method based on target extraction and speckle reducing anisotropic diffusion (SRAD) algorithm | |
CN104200472A (en) | Non-local wavelet information based remote sensing image change detection method | |
CN115578274A (en) | Double-branch image defogging method based on edge constraint | |
CN104537371A (en) | Fingerprint texture image denoising enhancing method and system | |
CN115184998A (en) | Rayleigh wave frequency dispersion curve automatic extraction method based on improved U-net neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |