CN103617626A - Central processing unit (CPU) and ground power unit (GPU)-based remote-sensing image multi-scale heterogeneous parallel segmentation method - Google Patents
Central processing unit (CPU) and ground power unit (GPU)-based remote-sensing image multi-scale heterogeneous parallel segmentation method Download PDFInfo
- Publication number
- CN103617626A CN103617626A CN201310686365.XA CN201310686365A CN103617626A CN 103617626 A CN103617626 A CN 103617626A CN 201310686365 A CN201310686365 A CN 201310686365A CN 103617626 A CN103617626 A CN 103617626A
- Authority
- CN
- China
- Prior art keywords
- gpu
- segmentation
- cpu
- image
- image blocks
- 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.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a high-efficiency image segmentation method aiming at remote-sensing image multi-scale segmentation. The method comprises the following steps: reading original remote-sensing image information to be segmented, setting segmentation parameters and formulating an optimized segmentation strategy in a CPU segmentation start control stage; segmenting by adopting a two-layer parallel computing architecture, wherein a first layer is used for solving limitation of a storage space and guaranteeing load balancing between the CPU and GPU based on a coarse-grained parallel architecture between image blocks, and a second layer is used for guaranteeing load balancing inside the CPU and GPU based on fine-grained parallel architecture of pixels; and finally, carrying out a CPU result saving stage. The method has the advantages that the segmentation speed is high, the segmentation precision is high, setting of multiple segmentation parameters is supported, segmentation of ultra-large remote-sensing images is supported, export of multiple segmentation results is supported, and the like.
Description
Technical field
The present invention relates to a kind of remote sensing image Parallel segmentation method, relate in particular to a kind of method that adopts the multiple dimensioned Parallel segmentation of remote sensing of region growing algorithm realization based on CPU+GPU heterogeneous platform.
Background technology
Current remote sensing image spatial resolution and spectral resolution become more and more higher, and remote sensing image acquisition capability is far longer than the processing power of image simultaneously, are badly in need of more image processing algorithm consuming time thoroughly to improve.Image Segmentation is an important step during OO remote sensing image information extracts, and its splitting speed directly has influence on the efficiency of information extraction.Conventional serial partitioning algorithm is subject to the restriction of cpu frequency, inefficiency; Parallel segmentation algorithm under multi-core CPU platform is subject to the impact of CPU framework and core amounts, and processing speed still exists very large room for promotion.
Summary of the invention
In order to address the above problem, the present invention proposes the multiple dimensioned isomery Parallel segmentation of a kind of remote sensing image based on CPU+GPU method, comprises the following steps:
Step 001, the original remote sensing image to be split such as reads, and obtains image Back ground Information and basic statistics information, obtains the calculated performance information of operation computer hardware CPU+GPU simultaneously;
Step 002, arranges partitioning parameters;
Step 003, the set partitioning parameters of the information of obtaining according to step 001 and step 002 is formulated optimum segmentation strategy;
Step 004, realize ground floor parallel computation framework: by Image Segmentation, be the impartial image blocks of some sizes, and untreated image blocks is loaded in CPU internal memory, undivided image blocks is loaded in GPU video memory, and start the thread of cutting apart of CPU and GPU, realize the Parallel segmentation between image blocks;
Step 005, realize second layer parallel computation framework: regard each pixel in image blocks as an independently object, each cuts apart pixel of processing that thread is corresponding, and in continuous iteration cutting procedure, the object that meets region growing condition is merged into an object;
Step 006, when the quantity of object can not change again, current image blocks is cut apart completely, segmentation result in GPU video memory is sent to CPU internal memory, and the result in CPU internal memory is saved in file, starts cutting apart of next image blocks simultaneously;
Step 007, repeating step 005, step 006 and step 007, until that all image blocks are cut apart is complete;
Step 008, to the aftertreatment of image blocks result;
Step 009, merges the result of all image blocks, and imaged object is carried out to unified numbering, and CPU obtains the information of each object, and all information is saved as to corresponding destination file.
Further, in step 002, computer hardware resource configuration when user can arrange Image Segmentation, if user abandons arranging, cutting procedure will be used the available resources of computing machine maximum.
Further, in step 008, comprise whether exist imaged object unmatched situation, if exist, secondary splitting is carried out in the region of unmatched imaged object if detecting image blocks border, perform step 005, step 006 and step 007.
Especially, segmentation strategy described in step 003 comprises Task Assigned Policy, Memory Allocation Strategy, thread dividing strategy, the formulating method of the segmentation strategy of described optimum comprises: transfer to CPU to process data loading, thread scheduling, logic control task, Parallel segmentation task transfers to multi-core CPU and GPU to process simultaneously; One of dynamic assignment can be preserved the internal memory of cutting apart side information data or the video memory of total data and its correspondence of an image blocks; Under the condition allowing at storage space, image blocks is arranged to maximum; Reduce the number of transmissions and the transmission quantity of data PCI-E passage between CPU and GPU; To cut apart the maximum refinement of thread.
Especially, in step 005, comprise:
Step 501, is initialized as an object by each pixel in image blocks, calculates the attribute information of each object, and described attribute information comprises the information of object spectral signature information, space characteristics information, adjacent object;
Step 502, calculates the numbering of the object in four adjacent directions of each pixel, and it is in 4 array that order is according to the rules kept at size by the numbering of adjacent object, when a direction does not exist adjacent object, with 0, represents, the numbering of image is since 1;
Step 503, the heterogeneous degree of adding up each object and adjacent object, finds the adjacent object with minimum heterogeneous degree, when the heterogeneous degree of this minimum is less than cut size, using the adjacent object with the heterogeneous degree of this minimum as best adjacent object, otherwise be 0 by the numbering assignment of best adjacent object;
Step 504, the numbering of all best adjacent object being greater than to 0 object increases, merge current object and best adjacent object, comprise the numbering of best adjacent object is revised as to the numbering of current object and calculates the attribute that merges rear new object the attribute information of replacing current object;
Step 505, calculates in four adjacent directions of each pixel whether have other objects, as long as there are one or more other objects, current pixel point is frontier point.
Especially, in step 501, each pixel will be preserved the object number of four adjacent directions of each pixel point, whether current point is the location index information of next pixel point in frontier point and this object.
Especially, in step 503, the computing formula of described heterogeneous degree is:
Wherein: h represents heterogeneous degree; F1c, f2c be the eigenwert of indicated object 1, object 2 respectively; The heterogeneous degree of hdiff is poor; C represents c wave band; Wc represents the weights of c wave band; The pixel quantity of n1, n2 submeter indicated object 1, object 2; Hmc, h1c, h2c represent respectively to merge the heterogeneous degree of rear object, object 1, object 2.
Further, in step 005, also comprise step 506, size and the dimension of design GPU grid and piece, comprise: allow the number of threads of cutting apart in each GPU piece be 32 integral multiple, after determining the size of GPU piece, according to the size of the image of cutting apart, determine the size of GPU grid, its computing formula is:
Wherein: Sgrid represents the size of GPU grid; Simg represents image picture element quantity; Sblock represents the size of GPU piece.
Especially, the described partitioning parameters in step 002 comprises: each is cut apart yardstick, wave band weights, shape weights and spectrum weights, cuts apart pattern, smoothness weights and degree of compacting weights.
Especially, method of the present invention adopts GPU asynchronous operation method, comprises that GPU in step 004 cuts apart the asynchronous communication that GPU segmentation result in the asynchronous starting of thread and step 006 is back to CPU.
Beneficial effect: the invention provides a kind of efficient image division method for multi-scale segmentation of remote sensing images, adopt CPU and GPU to mix accelerates simultaneously, realization is at the image Parallel segmentation algorithm of heterogeneous platform, given full play to the max calculation ability of CPU and GPU, have that splitting speed is fast, segmentation precision good, support multiple partitioning parameters setting, support super large remote sensing image cut apart, support the multiple advantages such as multiple segmentation result derivation.Meanwhile, the method has adopted the two-layer framework of cutting apart, for the isomery parallelization of other remote sensing images provides a referential universal solution.
Above-mentioned explanation is only the general introduction of technical solution of the present invention, in order to better understand technological means of the present invention, and can be implemented according to the content of instructions, and for above and other object of the present invention, feature and advantage can be become apparent, below especially exemplified by preferred embodiment, and coordinate accompanying drawing, be described in detail as follows.
Accompanying drawing explanation
The method flow diagram of Fig. 1 Parallel segmentation method of the present invention.
Fig. 2 is the method flow diagram of second layer parallel organization.
Embodiment
For making those skilled in the art understand better technical scheme of the present invention, below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.
The method of the present embodiment mainly comprises: one, CPU is cut apart and started the control stage; Two, ground floor parallel organization is cut apart; Three, second layer parallel organization is cut apart; Four, the CPU saving result stage, below with reference to Fig. 1, from this four-stage, the method for the present embodiment is elaborated.
One, CPU is cut apart and is started the control stage.
Step 001, the original remote sensing image to be split such as read, obtain image Back ground Information and basic statistics information, comprise wave band quantity, Pixel size, basic statistics information of image etc., obtain the calculated performance information of operation computer hardware CPU+GPU simultaneously, comprise the hardware informations such as video memory of CPU computing power, GPU computing power, internal memory, GPU.
Step 002, arranges partitioning parameters, comprises that each is cut apart yardstick, wave band weights, shape weights and spectrum weights, cuts apart pattern, the multiple partitioning parameters such as smoothness weights and degree of compacting weights.
" cut apart yardstick " and in image analysing computer, can be understood as the level of abstraction of people to atural object.The yardstick of cutting apart of image is the segmentation threshold of considering multiple factors, cuts apart the size that yardstick determines image, conventionally cuts apart yardstick larger, and imaged object is also larger.Various " weights " refer to the ratio that each several part parameter accounts in cutting procedure, and wherein shape weights and spectrum weights, smoothness weights and degree of compacting weights are two pairs of parameters independently separately, two pairs of weights and be 1.
" smoothness (Smoothness) " represents the smooth degree of imaged object boundary, is conventionally expressed as long (the being girth) l in border of imaged object and the deviation between long (the being rectangular perimeter) b of minimum outsourcing square boundary.In Raster Images, all imaged object existence condition l >=b.The computing formula of smoothness is as follows:
" compactness (Compactness) " represents the compact degree of imaged object in spatial domain, and being conventionally described as long (the being girth) l of imaged object actual boundary is the interior pixel number of object with object size n() root mean square between compactness deviation.The computing formula of compactness is as follows:
In this step, user can be in senior setting, the ratio of CPU and GPU calculated amount when Image Segmentation is set, and the size of CPU and GPU maximum free memory when cutting apart, and some other senior settings.If skipped senior setting, cutting procedure can be used the available resources of computing machine maximum.
Step 003, the set partitioning parameters of the information of obtaining by step 001 and step 002 is calculated optimum segmentation strategy, comprises Task Assigned Policy, Memory Allocation Strategy, image blocks allocation strategy, thread dividing strategy etc.
Above-mentioned strategy need be followed the most basic principle: realize Image Segmentation the most fast, concrete formulating method comprises: transfer to CPU to process the tasks such as data loading, thread scheduling, logic control, Parallel segmentation task transfers to multi-core CPU and GPU to process simultaneously; One of dynamic assignment can be preserved the internal memory of cutting apart side information data or the video memory of total data and its correspondence of an image blocks; Under the condition allowing at storage space, image blocks is arranged to maximum; Reduce the number of transmissions and the transmission quantity of data PCI-E passage between CPU and GPU; To cut apart the maximum refinement proof load of thread balanced.
Two, realize ground floor parallel computation framework, the coarse grain parallelism between CPU and GPU.
Step 004, realize ground floor parallel computation framework: by Image Segmentation, be the impartial image blocks of some sizes, and described image blocks is loaded into successively in the global storage of calculator memory and video memory, be in particular, untreated image blocks is loaded in internal memory, undivided image blocks is loaded in video memory, and starts the thread of cutting apart of CPU and GPU, realize the Parallel segmentation between image blocks.
This process belongs to the coarse grain parallelism between a kind of image blocks.Image block adopts a kind of " buffer zone formula " method of partition, can effectively reduce minute unmatched phenomenon of block boundary direct object.Ground floor parallel organization can solve the restriction of storage space and guarantee the load balancing between CPU and GPU.
Three, realize second layer parallel computation framework, the stage of CPU and GPU executed in parallel partitioning algorithm, the fine grained parallel between pixel.
Step 005, realize second layer parallel computation framework: regard each pixel in image blocks as an independently object, each cuts apart pixel of processing that thread is corresponding, and in continuous iteration cutting procedure, the object that meets region growing condition is merged into an object.
Step 006, when the quantity of object can not change again, current image blocks is cut apart completely, segmentation result in GPU video memory is sent to CPU internal memory, and the result in CPU internal memory is saved in file, starts cutting apart of next image blocks simultaneously.
Step 007, repeating step 005, step 006 and step 007, until that all image blocks are cut apart is complete.
Step 008, to the aftertreatment of image blocks result.
Especially, method of the present invention adopts GPU asynchronous operation method, comprises that GPU in step 004 cuts apart the asynchronous communication that GPU segmentation result in the asynchronous starting of thread and step 006 is back to CPU.
This layer of parallel architecture guaranteed the load balancing of CPU and GPU inside.
Owing to cutting apart, can produce on border the unmatched situation of imaged object after image block, therefore whether there is imaged object unmatched situation if in step 008, also will detect image blocks border, if exist, secondary splitting is carried out in the region of unmatched imaged object, perform step 005 and 006.
As shown in Figure 2, step 005 can be subdivided into following subprocess:
Step 501, is initialized as an object by each pixel in image blocks, calculates the attribute information of each object.Attribute information comprises the information (numbering of adjacent object, with the quantity of the common edge of adjacent object etc.) of object spectral signature information (average, standard deviation), space characteristics information (pixel quantity, girth, outsourcing rectangle), adjacent object.In addition each pixel will be preserved a part of information (comprising that whether the numbering of four adjacent directions (neighbours territory) object of each pixel point, current point be the location index of next pixel point in frontier point, this object etc.).
Step 502, calculate the numbering of the object on the neighbours territory of each pixel, it is in 4 array that order according to the rules (adopting order upper, left, down, right in the present embodiment) is kept at size by the numbering of adjacent object, when a direction does not exist adjacent object, with 0, represent, the numbering of image is since 1.
Step 503, the heterogeneous degree of adding up each object and adjacent object, finds the adjacent object with minimum heterogeneous degree, when the heterogeneous degree of this minimum is less than cut size, using the adjacent object with the heterogeneous degree of this minimum as best adjacent object, otherwise be 0 by the numbering assignment of best adjacent object.Complicated for high-resolution remote sensing image terrestrial object information, the method for the present embodiment adopts " minimum heterogeneous degree is poor " as the criterion of region growing, and heterogeneous degree has fully taken into account the parameters such as the gray feature of imagery zone and shape facility, and its computing formula is as follows:
Wherein: h represents heterogeneous degree; F1c, f2c be the eigenwert of indicated object 1, object 2 respectively; The heterogeneous degree of hdiff is poor; C represents c wave band; Wc represents the weights of c wave band; The pixel quantity of n1, n2 submeter indicated object 1, object 2; Hmc, h1c, h2c represent respectively to merge the heterogeneous degree of rear object, object 1, object 2.
Step 504, the numbering of all best adjacent object being greater than to 0 object increases, merge current object and best adjacent object, comprise the numbering of best adjacent object is revised as to the numbering of current object and calculates the attribute that merges rear new object the attribute information of replacing current object.
Step 505, calculates in four adjacent directions of each pixel whether have other objects, as long as there are one or more other objects, current pixel point is frontier point.
Step 506, size and the dimension of design GPU grid (Grid) and piece (Block).The Grid of appropriate design GPU and the size of Block and dimension can effectively improve the efficiency of cutting apart of image.In the dividing method of this embodiment, following following criterion designs: allow the number of threads of cutting apart in each block be 32 integral multiple, conventionally the quantity of cutting apart thread remains between 64 ~ 512, after determining the size of block, just can according to the size of the image of cutting apart, determine the size of grid, its computing formula is:
Wherein: Sgrid represents the size of GPU grid; Simg represents image picture element quantity; Sblock represents the size of GPU piece.
Four, the CPU saving result stage.
Step 009, merges the result of all image blocks, and imaged object is carried out to unified numbering, and CPU obtains the information of each object, and all information is saved as to corresponding destination file.
Expansion for the ease of follow-up work, the present embodiment is preserved the result for different file layouts by different segmentation results, comprising: the segmentation result picture file directly perceived of the imaged object vector border file of GeoTiff form object number raster file, Shapefile form, XML form object properties file, BMP and JPG form etc.
The efficient image division method for multi-scale segmentation of remote sensing images that the present embodiment provides, adopt CPU and GPU to mix accelerates simultaneously, realized the image Parallel segmentation algorithm at heterogeneous platform, given full play to the max calculation ability of CPU and GPU, have that splitting speed is fast, segmentation precision good, support multiple partitioning parameters setting, support super large remote sensing image cut apart, support the multiple advantages such as multiple segmentation result derivation.
For validity and the feasibility of checking the present embodiment method, in following hardware environment, for different partitioning parameters and remote sensing image, carry out the computer program of writing according to the present embodiment method, obtain following the result.
Main hardware environment
Main software environment
Image size with cut apart the efficiency table of comparisons
Known by the above results is analyzed, with respect to the serial algorithm that utilizes merely CPU to carry out, isomery Parallel segmentation can be obtained the speed-up ratio more than order of magnitude.For its acceleration specific energy of undersized image, reach more than 10 times, and along with its speed-up ratio of increase of image size is also increasing, until be of a size of the image acceleration effect of 6000*6000, have very significantly growth.Its speed-up ratio of image for 6000* 6000 has reached more than 20 times.By speed-up ratio rising tendency, can predict the increase along with image size, the speed-up ratio of isomery Parallel segmentation also can increase within the specific limits.The hardware device of the present embodiment test is comparatively backward, adopts better processor can obtain higher speed-up ratio.
Be understandable that, above embodiment is only used to principle of the present invention is described and the illustrative embodiments that adopts, yet the present invention is not limited thereto.For those skilled in the art, without departing from the spirit and substance in the present invention, can make various modification and improvement, these modification and improvement are also considered as protection scope of the present invention.
Claims (10)
1. the multiple dimensioned isomery Parallel segmentation of the remote sensing image based on a CPU+GPU method, is characterized in that, comprises the following steps:
Step 001, the original remote sensing image to be split such as reads, and obtains image Back ground Information and basic statistics information, obtains the calculated performance information of operation computer hardware CPU+GPU simultaneously;
Step 002, arranges partitioning parameters;
Step 003, the set partitioning parameters of the information of obtaining according to step 001 and step 002 is formulated optimum segmentation strategy;
Step 004, realize ground floor parallel computation framework: by Image Segmentation, be the impartial image blocks of some sizes, and untreated image blocks is loaded in CPU internal memory, undivided image blocks is loaded in GPU video memory, and start the thread of cutting apart of CPU and GPU, realize the Parallel segmentation between image blocks;
Step 005, realize second layer parallel computation framework: regard each pixel in image blocks as an independently object, each cuts apart pixel of processing that thread is corresponding, and in continuous iteration cutting procedure, the object that meets region growing condition is merged into an object;
Step 006, when the quantity of object can not change again, current image blocks is cut apart completely, segmentation result in GPU video memory is sent to CPU internal memory, and the result in CPU internal memory is saved in file, starts cutting apart of next image blocks simultaneously;
Step 007, repeating step 005, step 006 and step 007, until that all image blocks are cut apart is complete;
Step 008, to the aftertreatment of image blocks result;
Step 009, merges the result of all image blocks, and imaged object is carried out to unified numbering, and CPU obtains the information of each object, and all information is saved as to corresponding destination file.
2. method according to claim 1, is characterized in that, in step 002, and computer hardware resource configuration when user can arrange Image Segmentation, if user abandons arranging, cutting procedure will be used the available resources of computing machine maximum.
3. method according to claim 1, it is characterized in that in step 008, comprise whether detect image blocks border exists the unmatched situation of imaged object, if exist, secondary splitting is carried out in the region of unmatched imaged object, perform step 005, step 006 and step 007.
4. method according to claim 1, it is characterized in that, segmentation strategy described in step 003 comprises Task Assigned Policy, Memory Allocation Strategy, thread dividing strategy, the formulating method of the segmentation strategy of described optimum comprises: transfer to CPU to process data loading, thread scheduling, logic control task, Parallel segmentation task transfers to multi-core CPU and GPU to process simultaneously; One of dynamic assignment can be preserved the internal memory of cutting apart side information data or the video memory of total data and its correspondence of an image blocks; Under the condition allowing at storage space, image blocks is arranged to maximum; Reduce the number of transmissions and the transmission quantity of data PCI-E passage between CPU and GPU; To cut apart the maximum refinement of thread.
5. method according to claim 1, is characterized in that, step 005 comprises:
Step 501, is initialized as an object by each pixel in image blocks, calculates the attribute information of each object, and described attribute information comprises the information of object spectral signature information, space characteristics information, adjacent object;
Step 502, calculates the numbering of the object in four adjacent directions of each pixel, and it is in 4 array that order is according to the rules kept at size by the numbering of adjacent object, when a direction does not exist adjacent object, with 0, represents, the numbering of image is since 1;
Step 503, the heterogeneous degree of adding up each object and adjacent object, finds the adjacent object with minimum heterogeneous degree, when the heterogeneous degree of this minimum is less than cut size, using the adjacent object with the heterogeneous degree of this minimum as best adjacent object, otherwise be 0 by the numbering assignment of best adjacent object;
Step 504, the numbering of all best adjacent object being greater than to 0 object increases, merge current object and best adjacent object, comprise the numbering of best adjacent object is revised as to the numbering of current object and calculates the attribute that merges rear new object the attribute information of replacing current object;
Step 505, calculates in four adjacent directions of each pixel whether have other objects, as long as there are one or more other objects, current pixel point is frontier point.
6. method according to claim 5, is characterized in that, in step 501, each pixel will be preserved the object number of four adjacent directions of each pixel point, whether current point is the location index information of next pixel point in frontier point and this object.
7. method according to claim 5, is characterized in that, in step 503, the computing formula of described heterogeneous degree is:
Wherein: h represents heterogeneous degree; f
1c, f
2cthe eigenwert of difference indicated object 1, object 2; h
diffheterogeneous degree is poor; C represents c wave band; w
cthe weights that represent c wave band; n
1, n
2the pixel quantity of submeter indicated object 1, object 2; h
mc, h
1c, h
2cthe heterogeneous degree that represents respectively object after merging, object 1, object 2.
8. method according to claim 5, it is characterized in that, step 005 also comprises step 506, size and the dimension of design GPU grid and piece, comprise: allow the number of threads of cutting apart in each GPU piece be 32 integral multiple, after determining the size of GPU piece, according to the size of the image of cutting apart, determine the size of GPU grid, its computing formula is:
Wherein: S
gridthe size that represents GPU grid; S
imgrepresent image picture element quantity; S
blockthe size that represents GPU piece.
9. method according to claim 1, is characterized in that, in step 002, described partitioning parameters comprises: each is cut apart yardstick, wave band weights, shape weights and spectrum weights, cuts apart pattern, smoothness weights and degree of compacting weights.
10. method according to claim 1, is characterized in that, described method adopts GPU asynchronous operation method, comprises that GPU in step 004 cuts apart the asynchronous communication that GPU segmentation result in the asynchronous starting of thread and step 006 is back to CPU.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310686365.XA CN103617626A (en) | 2013-12-16 | 2013-12-16 | Central processing unit (CPU) and ground power unit (GPU)-based remote-sensing image multi-scale heterogeneous parallel segmentation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310686365.XA CN103617626A (en) | 2013-12-16 | 2013-12-16 | Central processing unit (CPU) and ground power unit (GPU)-based remote-sensing image multi-scale heterogeneous parallel segmentation method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103617626A true CN103617626A (en) | 2014-03-05 |
Family
ID=50168330
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310686365.XA Pending CN103617626A (en) | 2013-12-16 | 2013-12-16 | Central processing unit (CPU) and ground power unit (GPU)-based remote-sensing image multi-scale heterogeneous parallel segmentation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103617626A (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104200482A (en) * | 2014-09-17 | 2014-12-10 | 武汉狮图空间信息技术有限公司 | SSC (scale-shape-compactness) based optimization method of multi-scale segmentation parameters |
CN104267940A (en) * | 2014-09-17 | 2015-01-07 | 武汉狮图空间信息技术有限公司 | Quick map tile generation method based on CPU+GPU |
CN105719333A (en) * | 2014-12-02 | 2016-06-29 | 上海联影医疗科技有限公司 | 3D image data processing method and 3D image data processing device |
CN105931256A (en) * | 2016-06-03 | 2016-09-07 | 中国地质大学(武汉) | CUDA (compute unified device architecture)-based large-format remote sensing image fast segmentation method |
CN106060581A (en) * | 2016-07-05 | 2016-10-26 | 广州华多网络科技有限公司 | Video real-time transmission data processing method, device and system |
CN106137235A (en) * | 2016-07-26 | 2016-11-23 | 中国科学院深圳先进技术研究院 | C-arm X-ray machine, control system and medical image system |
CN109523459A (en) * | 2017-09-19 | 2019-03-26 | 通用电气公司 | Method and system for reconstruction image |
CN110083446A (en) * | 2019-04-23 | 2019-08-02 | 北京吉威时代软件股份有限公司 | A kind of GPU parallel with remote sensing image real-time processing method and system under zero I/O mode |
CN110135569A (en) * | 2019-04-11 | 2019-08-16 | 中国人民解放军国防科技大学 | Heterogeneous platform neuron positioning three-level flow parallel method, system and medium |
CN110288602A (en) * | 2019-05-10 | 2019-09-27 | 香港理工大学深圳研究院 | Come down extracting method, landslide extraction system and terminal |
CN110415162A (en) * | 2019-07-22 | 2019-11-05 | 中国人民大学 | Towards the adaptive figure division methods of isomery fusion treatment device in big data |
CN111489281A (en) * | 2020-04-09 | 2020-08-04 | 常州奥创医疗科技有限公司 | Detection method based on GPU and CPU cooperative operation |
CN112149287A (en) * | 2020-09-08 | 2020-12-29 | 安徽中科龙安科技股份有限公司 | Traffic simulation road network graphical segmentation method and system oriented to load balancing |
CN112463905A (en) * | 2020-11-30 | 2021-03-09 | 湖北金拓维信息技术有限公司 | Vector data SHP file parallel writing method |
US11094067B2 (en) | 2014-12-02 | 2021-08-17 | Shanghai United Imaging Healthcare Co., Ltd. | Method and system for image processing |
CN116483587A (en) * | 2023-06-21 | 2023-07-25 | 湖南马栏山视频先进技术研究院有限公司 | Video super-division parallel method, server and medium based on image segmentation |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070040830A1 (en) * | 2005-08-18 | 2007-02-22 | Pavlos Papageorgiou | Volume rendering apparatus and process |
CN101354780A (en) * | 2007-07-26 | 2009-01-28 | Lg电子株式会社 | Graphic data processing apparatus and method |
CN101706950A (en) * | 2009-07-13 | 2010-05-12 | 中国科学院遥感应用研究所 | High-performance implementation method for multi-scale segmentation of remote sensing images |
CN101783028A (en) * | 2010-02-26 | 2010-07-21 | 清华大学 | Quick partition method of three-dimensional medical image on basis of video card parallel computing |
CN102314688A (en) * | 2011-09-09 | 2012-01-11 | 华中科技大学 | Image segmentation method and image set segmentation method |
-
2013
- 2013-12-16 CN CN201310686365.XA patent/CN103617626A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070040830A1 (en) * | 2005-08-18 | 2007-02-22 | Pavlos Papageorgiou | Volume rendering apparatus and process |
CN101354780A (en) * | 2007-07-26 | 2009-01-28 | Lg电子株式会社 | Graphic data processing apparatus and method |
CN101706950A (en) * | 2009-07-13 | 2010-05-12 | 中国科学院遥感应用研究所 | High-performance implementation method for multi-scale segmentation of remote sensing images |
CN101783028A (en) * | 2010-02-26 | 2010-07-21 | 清华大学 | Quick partition method of three-dimensional medical image on basis of video card parallel computing |
CN102314688A (en) * | 2011-09-09 | 2012-01-11 | 华中科技大学 | Image segmentation method and image set segmentation method |
Non-Patent Citations (6)
Title |
---|
MAHMOUD A. HOSSAM 等: "Hybrid Cluster of Multicore CPUs and GPUs for Accelerating Hyperspectral Image Hierarchical Segmentation", 《2013 IEEE ICCES》 * |
P. N. HAPP 等: "A PARALLEL IMAGE SEGMENTATION ALGORITHM ON GPUS", 《PROCEEDINGS OF THE 4TH GEOBIA》 * |
PATRICK NIGRI HAPP 等: "A Region Growing Segmentation Algorithm for GPUs", 《2013 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
于海洋 等: "基于LiDAR和航空影像的地震灾害倒塌建筑物信息提取", 《国土资源遥感》 * |
戈乐乐 等: "基于分水岭算法的高分辨率遥感图像多尺", 《HTTP://WWW.DOCIN.COM/P-129504965.HTML》 * |
肖汉: "基于CPU+GPU的影像匹配高效能异构并行计算研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104200482A (en) * | 2014-09-17 | 2014-12-10 | 武汉狮图空间信息技术有限公司 | SSC (scale-shape-compactness) based optimization method of multi-scale segmentation parameters |
CN104267940A (en) * | 2014-09-17 | 2015-01-07 | 武汉狮图空间信息技术有限公司 | Quick map tile generation method based on CPU+GPU |
CN105719333A (en) * | 2014-12-02 | 2016-06-29 | 上海联影医疗科技有限公司 | 3D image data processing method and 3D image data processing device |
US11094067B2 (en) | 2014-12-02 | 2021-08-17 | Shanghai United Imaging Healthcare Co., Ltd. | Method and system for image processing |
CN105719333B (en) * | 2014-12-02 | 2019-12-20 | 上海联影医疗科技有限公司 | Three-dimensional image data processing method and device |
CN105931256A (en) * | 2016-06-03 | 2016-09-07 | 中国地质大学(武汉) | CUDA (compute unified device architecture)-based large-format remote sensing image fast segmentation method |
CN106060581A (en) * | 2016-07-05 | 2016-10-26 | 广州华多网络科技有限公司 | Video real-time transmission data processing method, device and system |
CN106060581B (en) * | 2016-07-05 | 2019-05-07 | 广州华多网络科技有限公司 | A kind of realtime video transmission data processing method, apparatus and system |
CN106137235A (en) * | 2016-07-26 | 2016-11-23 | 中国科学院深圳先进技术研究院 | C-arm X-ray machine, control system and medical image system |
CN109523459A (en) * | 2017-09-19 | 2019-03-26 | 通用电气公司 | Method and system for reconstruction image |
CN109523459B (en) * | 2017-09-19 | 2024-04-23 | 通用电气公司 | Method and system for reconstructing an image |
CN110135569A (en) * | 2019-04-11 | 2019-08-16 | 中国人民解放军国防科技大学 | Heterogeneous platform neuron positioning three-level flow parallel method, system and medium |
CN110135569B (en) * | 2019-04-11 | 2021-09-21 | 中国人民解放军国防科技大学 | Heterogeneous platform neuron positioning three-level flow parallel method, system and medium |
CN110083446A (en) * | 2019-04-23 | 2019-08-02 | 北京吉威时代软件股份有限公司 | A kind of GPU parallel with remote sensing image real-time processing method and system under zero I/O mode |
CN110288602A (en) * | 2019-05-10 | 2019-09-27 | 香港理工大学深圳研究院 | Come down extracting method, landslide extraction system and terminal |
CN110415162A (en) * | 2019-07-22 | 2019-11-05 | 中国人民大学 | Towards the adaptive figure division methods of isomery fusion treatment device in big data |
CN111489281A (en) * | 2020-04-09 | 2020-08-04 | 常州奥创医疗科技有限公司 | Detection method based on GPU and CPU cooperative operation |
CN112149287A (en) * | 2020-09-08 | 2020-12-29 | 安徽中科龙安科技股份有限公司 | Traffic simulation road network graphical segmentation method and system oriented to load balancing |
CN112149287B (en) * | 2020-09-08 | 2024-06-18 | 安徽中科龙安科技股份有限公司 | Traffic simulation road network graphical segmentation method and system for load balancing |
CN112463905A (en) * | 2020-11-30 | 2021-03-09 | 湖北金拓维信息技术有限公司 | Vector data SHP file parallel writing method |
CN112463905B (en) * | 2020-11-30 | 2022-06-03 | 湖北金拓维信息技术有限公司 | Vector data SHP file parallel writing method |
CN116483587A (en) * | 2023-06-21 | 2023-07-25 | 湖南马栏山视频先进技术研究院有限公司 | Video super-division parallel method, server and medium based on image segmentation |
CN116483587B (en) * | 2023-06-21 | 2023-09-08 | 湖南马栏山视频先进技术研究院有限公司 | Video super-division parallel method, server and medium based on image segmentation |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103617626A (en) | Central processing unit (CPU) and ground power unit (GPU)-based remote-sensing image multi-scale heterogeneous parallel segmentation method | |
CN103365796B (en) | Reused by optimizing caching in the drafting of multiprocessor shared memory system upper volume | |
CN107657599B (en) | Parallel implementation method of remote sensing image fusion system based on mixed granularity division and dynamic load distribution | |
US9330492B2 (en) | Prediction method, generation method and storage medium | |
US20140267238A1 (en) | Conservative rasterization of primitives using an error term | |
US9424617B2 (en) | Graphics command generation device and graphics command generation method | |
EP3161793B1 (en) | Adaptive partition mechanism with arbitrary tile shape for tile based rendering gpu architecture | |
US20180018751A1 (en) | Graphics Processing Method, Apparatus, and System | |
TW201435795A (en) | Consistent vertex snapping for variable resolution rendering | |
CN105550974A (en) | GPU-based acceleration method of image feature extraction algorithm | |
KR100959349B1 (en) | A method for accelerating terrain rendering based on quadtree using graphics processing unit | |
CN109388486B (en) | Data placement and migration method for heterogeneous memory and multi-type application mixed deployment scene | |
TW201539376A (en) | System and method for cutting point clouds | |
GB2561807A (en) | Computing systems and methods for processing graphics data using cost indications for sets of tiles of a rendering space | |
WO2020125062A1 (en) | Image fusion method and related device | |
CN105931256A (en) | CUDA (compute unified device architecture)-based large-format remote sensing image fast segmentation method | |
CN104267940A (en) | Quick map tile generation method based on CPU+GPU | |
CN103400404A (en) | Method for efficiently rendering bitmap motion trail | |
US20130328876A1 (en) | Building kd-trees in a depth first manner on heterogeneous computer systems | |
CN105701760A (en) | Histogram real-time generation method of geographic raster data optional polygon area | |
CN109416688B (en) | Method and system for flexible high performance structured data processing | |
CN106484532B (en) | GPGPU parallel calculating method towards SPH fluid simulation | |
Martin et al. | Load-Balanced Isosurfacing on Multi-GPU Clusters. | |
Sun et al. | A 28nm 2D/3D unified sparse convolution accelerator with block-wise neighbor searcher for large-scaled voxel-based point cloud network | |
CN103810137A (en) | NCS algorithm parallelization method based on multiple FPGA platforms |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140305 |