CN102314688B - Image segmentation method and image set segmentation method - Google Patents

Image segmentation method and image set segmentation method Download PDF

Info

Publication number
CN102314688B
CN102314688B CN 201110267364 CN201110267364A CN102314688B CN 102314688 B CN102314688 B CN 102314688B CN 201110267364 CN201110267364 CN 201110267364 CN 201110267364 A CN201110267364 A CN 201110267364A CN 102314688 B CN102314688 B CN 102314688B
Authority
CN
China
Prior art keywords
image
cpu
gpu
parameter
initial
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.)
Expired - Fee Related
Application number
CN 201110267364
Other languages
Chinese (zh)
Other versions
CN102314688A (en
Inventor
金海�
郑然�
汪聪
冯晓文
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huazhong University of Science and Technology
Original Assignee
Huazhong University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huazhong University of Science and Technology filed Critical Huazhong University of Science and Technology
Priority to CN 201110267364 priority Critical patent/CN102314688B/en
Publication of CN102314688A publication Critical patent/CN102314688A/en
Application granted granted Critical
Publication of CN102314688B publication Critical patent/CN102314688B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Analysis (AREA)

Abstract

The invention provides an image segmentation method, which includes the following steps: an image is read by a CPU (central processing unit), the parameters of the image are initialized, memory spaces needed by the CPU and a GPU (graphics processing unit) are allocated, and the image is transmitted to the GPU; the GPU calculates the image according to the mean shift algorithm to obtain initial RGB (red, blue and green) color information and initial coordinate information, and transmits the initial RGB color information and the initial coordinate information back to the CPU; according to the initial RGB color information and the initial coordinate information, the CPU clusters the pixels of the imager, so that a plurality of regions can be generated, the statistics of the parameters of the regions is carries out, and the parameters of the regions are transmitted to the GPU; according to the fuzzy C-means algorithm, the GPU calculates the parameters of the regions to obtain the final segmentation result of the image, and transmits the final segmentation result to the CPU; and the CPU outputs the final segmentation result of the image. The method is characterized in that: universality is high, the segmentation speed is high, and the image can be segmented into the number of regions needed by a user.

Description

Image partition method and image set dividing method
Technical field
The present invention relates to a kind of image partition method, be specifically related to a kind of vision-mix dividing method based on mean shift algorithm and FCM Algorithms.
Background technology
Image Segmentation Technology is one of important research content of Computer Image Processing and visionics, and focus and difficult point for studying.It is the important committed step of pattern-recognition and graphical analysis, and the quality of segmentation effect directly affects the processing of successive image.Therefore widely used various image partition methods all have length consuming time, cut apart the shortcoming that efficient is low, segmentation effect is poor now, and can not be widely used in needs the system that processes in real time, such as based on the image search engine of picture material etc.
Summary of the invention
In view of this, one object of the present invention is to provide a kind of image partition method, the advantage that it has weak point consuming time, cuts apart the efficient height, segmentation effect is good.
Another object of the present invention is to provide a kind of image set dividing method, the advantage that it has weak point consuming time, cuts apart the efficient height, segmentation effect is good.
A kind of image partition method comprises the steps:
CPU reads an image, and the parameter of initialisation image is distributed CPU and the needed memory headroom of GPU, and image is passed to GPU;
GPU calculates image according to mean shift algorithm, obtaining initial RGB colouring information and initial coordinate information, and passes it back CPU;
CPU carries out cluster according to initial RGB colouring information and initial coordinate information to the pixel of image, thereby forms a plurality of zones, the parameter of statistical regions, and send it to GPU;
GPU calculates the parameter in zone according to FCM Algorithms, with the final segmentation result of acquisition image, and sends it to CPU;
The final segmentation result of CPU output image.
CPU is poly-according to the initial RGB colouring information pixel RGB colouring information is identical and adjacent with initial coordinate information to be a class.
The parameter in zone comprises the interior pixel number in number, zone in zone and the color average in zone.
The final segmentation result of image comprises final RGB colouring information and final coordinate information.
A kind of image set dividing method comprises the steps:
CPU receives a plurality of images, and determines the quantity of a plurality of images;
CPU reads i image, and the parameter of initialisation image is distributed CPU and the needed memory headroom of GPU, and i image passed to GPU;
CPU judges whether i+1 is less than or equal to the quantity of a plurality of images;
If, then GPU calculates i image according to mean shift algorithm, to obtain initial RGB colouring information and initial coordinate information, and pass it back CPU, CPU reads i+1 image simultaneously, the parameter of i+1 image of initialization is distributed CPU and the needed memory headroom of GPU, and i+1 image passed to GPU;
CPU carries out cluster according to initial RGB colouring information and initial coordinate information to the pixel of i image, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU, GPU calculates i+1 image according to mean shift algorithm simultaneously, obtaining initial RGB colouring information and initial coordinate information, and pass it back CPU;
GPU calculates the parameter in zone according to FCM Algorithms, to obtain the final segmentation result of i image, and send it to CPU, CPU carries out cluster according to initial RGB colouring information and initial coordinate information to the pixel of i+1 image simultaneously, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU;
CPU judges whether i+2 is less than or equal to the quantity of a plurality of images;
If, then CPU exports the final segmentation result of i image, read i+2 image, the parameter of i+2 image of initialization is distributed CPU and the needed memory headroom of GPU, and i+2 image passed to GPU, GPU calculates the parameter in zone according to FCM Algorithms simultaneously, obtaining the final segmentation result of i+1 image, and send it to CPU
CPU judges whether i+3 is less than or equal to the quantity of a plurality of images;
If, then CPU exports the final segmentation result of i+1 image, read i+3 image, the parameter of i+3 image of initialization, distribute CPU and the needed memory headroom of GPU, and i+3 image passed to GPU, GPU calculates i+2 image according to mean shift algorithm, obtaining initial RGB colouring information and initial coordinate information, and pass it back CPU;
I=i+2 is set;
Repeat CPU and according to initial RGB colouring information and initial coordinate information the pixel of i image is carried out cluster, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU, GPU calculates i+1 image according to mean shift algorithm simultaneously, obtaining initial RGB colouring information and initial coordinate information, and pass it step of CPU back.
Image set dividing method of the present invention also comprises step:
If i+1 is not the quantity that is less than or equal to a plurality of images, then GPU calculates i image according to mean shift algorithm, obtaining initial RGB colouring information and initial coordinate information, and passes it back CPU;
CPU carries out cluster according to initial RGB colouring information and initial coordinate information to the pixel of i image, thereby forms a plurality of zones, the parameter of statistical regions, and send it to GPU;
GPU calculates the parameter in zone according to FCM Algorithms, obtaining the final segmentation result of i image, and sends it to CPU;
CPU exports the final segmentation result of i image.
Image set dividing method of the present invention also comprises step:
If i+2 is not the quantity that is less than or equal to a plurality of images, then CPU exports the segmentation result of i image, and simultaneously GPU calculates the parameter in zone according to FCM Algorithms, obtaining the final segmentation result of i+1 image, and sends it to CPU;
I=i+1 is set;
CPU exports the final segmentation result of i image.
Image set dividing method of the present invention also comprises step:
If i+3 is not the quantity that is less than or equal to a plurality of images, then CPU exports the final segmentation result of i+1 image, simultaneously GPU calculates i+2 image according to mean shift algorithm, obtaining initial RGB colouring information and initial coordinate information, and passes it back CPU;
I=i+2 is set;
CPU carries out cluster according to initial RGB colouring information and initial coordinate information to the pixel of i image, thereby forms a plurality of zones, the parameter of statistical regions, and send it to GPU;
GPU calculates the parameter in zone according to FCM Algorithms, obtaining the final segmentation result of i image, and sends it to CPU;
CPU exports the final segmentation result of i image.
CPU is poly-according to the initial RGB colouring information pixel RGB colouring information is identical and adjacent with initial coordinate information to be a class.
The parameter in zone comprises the interior pixel number in number, zone in zone and the color average in zone, and the final segmentation result of image comprises final RGB colouring information and final coordinate information.
Method of the present invention has the following advantages: the combination of (1) mean shift algorithm and FCM Algorithms, so that the present invention can both reach preferably segmentation effect to image and image set; (2) splitting speed is fast, and the asynchronism and concurrency execution pattern that adopts CPU to combine with GPU takes full advantage of the computation capability of GPU and the control ability of CPU, reasonable distribution task, maximum using hardware; (3) number of regions that image segmentation is become the user need is inputted the cluster centre parameter during FCM Algorithms cluster, just image segmentation is become the number of regions of this number, and this is very important to image-region feature application that needs extract fixed number.
Description of drawings
Fig. 1 is the process flow diagram of image partition method of the present invention;
Fig. 2 is the process flow diagram of image set dividing method of the present invention.
Embodiment
The present invention is described in detail below in conjunction with the drawings and specific embodiments.
As shown in Figure 1, image partition method of the present invention may further comprise the steps:
(1) CPU reads an image, and the parameter of initialisation image is distributed CPU and the needed memory headroom of GPU, and image is passed to GPU;
(2) GPU calculates image according to mean shift algorithm, obtaining initial RGB colouring information and initial coordinate information, and passes it back CPU;
(3) CPU carries out cluster according to initial RGB colouring information and initial coordinate information to the pixel of image, thereby forms a plurality of zones, the parameter of statistical regions, and send it to GPU;
(4) GPU calculates the parameter in zone according to FCM Algorithms, with the final segmentation result of acquisition image, and sends it to CPU;
(5) the final segmentation result of CPU output image.
In step (1), the RGB data-switching of image is become LUV spatial mode data, store in the memory headroom.
The parameter of initialisation image comprises: color space radius, coordinate space radius and regional minimum number of pixels, the image of transmission comprise original view data and weight data.
In step (2), the Thread Count of an image block is set to nThread (value 64,128 or 256), and the calculating of a pixel of each thread process dynamically distributes the image block number according to the image size.The image block number is: (width*height)/nThread+1, and width representative image width wherein, height representative image height.
Pixel y of each thread execution I, jDrift calculate, computing formula is:
y ( i , j ) + d = Σ k = 1 m { g ( | y ( i , j ) , s - x k , s h s | 2 ) g ( | y ( i , j ) , r - x k , r h r | 2 ) w ( x k ) x k } Σ k = 1 m { g ( | y ( i , j ) , s - x k , s h s | 2 ) g ( | y ( i , j ) , r - x k , r h r | 2 ) w ( x k ) }
Wherein d was for carrying out the number of times of drift calculating, and m is Fuzzy Exponential, y (i, j), sExpression pixel y I, jSpatial component, y (i, j), rExpression pixel y I, jColor component, x kTo fall into y I, jCentered by, the space radius is h sThe point of scope, h s, h rThe forms radius of spatial component and color component, x K, rThe color component of expression pixel k, x K, sThe spatial component of expression pixel k, w (x k) the expression weight.Circulation is calculated with this formula, only arrives | y (i, j)+d-y (i, j)+d-1|＜ε is finally circulation just.After all threads were all finished cycle calculations, GPU was with data back CPU.
In step (3), CPU is poly-according to RGB colouring information and the coordinate information pixel that the RGB colouring information is identical and adjacent to be a class.
The parameter in zone comprises the interior pixel number in number, zone in zone and the color average in zone, and the segmentation result of image comprises RGB colouring information and coordinate information.
In this step, eight neighborhoods of each pixel of CPU recursive search are labeled as the same area with colouring information same pixel point.The number of statistical regions reorganizes data, and each zone is made as a whole its color value of getting and carried out next step calculating.GPU distributes respectively memory headrooms such as being subordinate to matrix, cluster centre, raw data, distance matrix, copies data to GPU from the CPU end.
In step (4), FCM Algorithms mainly comprises following process:
(i) initialization degree of membership matrix
The degree of membership matrix size is the two-dimentional floating type array of c*n, and wherein c is the cluster centre number, and n is the pixel number.The number of threads that each image block is set is nThread (getting 64,128 or 256), and the number of image block is n/256+1.Each thread at first generates c the floating number between the 0-1 at random for each pixel.Then carry out the normalization operation, ask first the cumulative and sum of c number c, use sum cExcept every number is then finished the work of initialization degree of membership matrix.
(ii) according to degree of membership matrix computations cluster centre
Computing formula is:
Figure BDA0000090403090000061
0＜i＜c wherein, u is the degree of membership matrix, and n is the pixel number, and m is Fuzzy Exponential, x jBe j pixel.
The computing of degree of membership index of matrix, the number of threads that each image block is set are nThread (getting 64,128 or 256), and the number of image block is n/nThread+1, and each thread is responsible for the exponent arithmetic of a pixel.
Matrix multiplication operation for reducing the repetition multiply operation of molecule, is done the matrix multiplication operation to the degree of membership matrix behind the exponent arithmetic and data matrix first.
Cumulative reduction computing, the number of threads that each image block is set is nThread (getting 64,128 or 256), the initial number of image block is n/ (2*nThread)+1, behind each thread execution one sub-addition, the Thread Count reduction finally only has an image block to calculate final result.For raising the efficiency, when thread less than or equal to 32 the time, to its loop unrolling.
Molecule and denominator are done division, i.e. renewable cluster centre.
(iii) value calculation judges that whether it is less than a threshold value.If less than then turning (v), otherwise turn (iv).
Computing formula is:
Figure BDA0000090403090000062
Wherein U is the degree of membership matrix, and C is cluster centre, and d is distance matrix, and m is Fuzzy Exponential.
(iv) distance is calculated, and upgrades the degree of membership matrix, then turns (ii)
Calculate each as several points to a few moral distances of the Ou Ji of c cluster centre, the number of threads that each image block is set is that nThread (gets 64,128 or 256), the initial number of image block is n/ (2*nThread)+1, each thread is responsible for calculating a pixel to the distance of c cluster centre, and its value is saved in the distance matrix.
Upgrade the degree of membership matrix, computing formula is:
Figure BDA0000090403090000071
Wherein d is distance, and m is Fuzzy Exponential.The number of threads that each image block is set is nThread (getting 64,128 or 256), and the initial number of image block is n/ (2*nThread)+1, and each thread is responsible for calculating c square of a pixel from the calculating of value.
(v) according to the final ownership of degree of membership matrix computations data
The number of threads that each image block is set is that nThread (gets 64,128 or 256), the initial number of image block is n/ (2*nThread)+1, and each thread is responsible for the value maximum which cluster centre pixel of comparison is under the jurisdiction of, and this pixel mark is finally belonged to this type of.
As shown in Figure 2, image set dividing method of the present invention may further comprise the steps:
1.CPU receive a plurality of images, and determine the quantity of a plurality of images;
2.CPU read i image, the parameter of initialisation image is distributed CPU and the needed memory headroom of GPU, and i image passed to GPU;
Whether be less than or equal to the quantity of a plurality of images 3.CPU judge i+1; If so, then enter step 4, if not, step 16 then entered;
4.GPU according to mean shift algorithm i image calculated, to obtain initial RGB colouring information and initial coordinate information, and pass it back CPU, CPU reads i+1 image simultaneously, the parameter of i+1 image of initialization, distribute CPU and the needed memory headroom of GPU, and i+1 image passed to GPU;
5.CPU according to initial RGB colouring information and initial coordinate information the pixel of i image is carried out cluster, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU, GPU calculates i+1 image according to mean shift algorithm simultaneously, obtaining initial RGB colouring information and initial coordinate information, and pass it back CPU;
6.GPU according to FCM Algorithms the parameter in zone is calculated, to obtain the final segmentation result of i image, and send it to CPU, CPU carries out cluster according to initial RGB colouring information and initial coordinate information to the pixel of i+1 image simultaneously, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU;
Whether be less than or equal to the quantity of a plurality of images 7.CPU judge i+2, if so, then change step 8 over to, if not, then change step 12 over to;
8.CPU export the final segmentation result of i image, read i+2 image, the parameter of i+2 image of initialization, distribute CPU and the needed memory headroom of GPU, and i+2 image passed to GPU, GPU calculates the parameter in zone according to FCM Algorithms simultaneously, to obtain the final segmentation result of i+1 image, and send it to CPU
Whether be less than or equal to the quantity of a plurality of images 9.CPU judge i+3, if so, then change step 10 over to, if not, then change step 14 over to;
10.CPU export the final segmentation result of i+1 image, read i+3 image, the parameter of i+3 image of initialization, distribute CPU and the needed memory headroom of GPU, and i+3 image passed to GPU, GPU calculates i+2 image according to mean shift algorithm, obtaining initial RGB colouring information and initial coordinate information, and passes it back CPU;
11. i=i+2 is set, then forwards step 5 to;
12.CPU export the final segmentation result of i image, simultaneously GPU calculates the parameter in zone according to FCM Algorithms, obtaining the final segmentation result of i+1 image, and sends it to CPU;
13. i=i+1 is set, then forwards step 19 to;
14.CPU export the final segmentation result of i+1 image, simultaneously GPU calculates i+2 image according to mean shift algorithm, obtaining initial RGB colouring information and initial coordinate information, and passes it back CPU;
15. i=i+2 is set, then forwards step 17 to.
16.GPU according to mean shift algorithm i image calculated, obtaining initial RGB colouring information and initial coordinate information, and passes it back CPU;
17.CPU according to initial RGB colouring information and initial coordinate information the pixel of i image is carried out cluster, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU;
18.GPU according to FCM Algorithms the parameter in zone is calculated, obtaining the final segmentation result of i image, and is sent it to CPU;
19.CPU export the final segmentation result of i image;
Example:
In order to verify feasibility of the present invention and validity, under the experimental configuration environment shown in the lower tabulation 1, carry out the computer program of writing, invention to be tested, operation result is as shown in table 2.
Table 1: experimental configuration environment
Figure BDA0000090403090000091
In table 2, the parameter of mean shift algorithm is (x, y), and wherein x represents the locational space radius, and y represents the color space radius.By testing the working time of 100 width of cloth images, find that speed-up ratio can reach more than 150 times, visible the present invention has good operational efficiency.
Table 2: operation result
The above only is the specific embodiment of the present invention; should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the principle of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (6)

1. an image partition method is characterized in that, comprises the steps:
CPU reads an image, and the parameter of the described image of initialization is distributed CPU and the needed memory headroom of GPU, and described image is passed to GPU;
GPU calculates described image according to mean shift algorithm, obtaining initial RGB colouring information and initial coordinate information, and passes it back CPU;
CPU carries out cluster according to described RGB colouring information and described coordinate information to the pixel of described image, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU, the parameter in zone comprises the interior pixel number in number, zone in zone and the color average in zone;
GPU calculates the parameter in described zone according to FCM Algorithms, obtaining the final segmentation result of described image, and sends it to CPU;
CPU exports the segmentation result of described image.
2. image partition method according to claim 1 is characterized in that, CPU is poly-according to RGB colouring information and the coordinate information pixel that the RGB colouring information is identical and adjacent to be a class.
3. image partition method according to claim 1 is characterized in that, the final segmentation result of described image comprises final RGB colouring information and final coordinate information.
4. an image set dividing method is characterized in that, comprises the steps:
CPU receives a plurality of images, and determines the quantity of described a plurality of images;
CPU reads i image, and the parameter of described i the image of initialization is distributed CPU and the needed memory headroom of GPU, and i image passed to GPU;
CPU judges whether i+1 is less than or equal to the quantity of described a plurality of images;
If, then GPU calculates i image according to mean shift algorithm, to obtain initial RGB colouring information and initial coordinate information, and pass it back CPU, CPU reads i+1 image simultaneously, the parameter of i+1 image of initialization is distributed CPU and the needed memory headroom of GPU, and i+1 image passed to GPU;
CPU carries out cluster according to described initial RGB colouring information and described initial coordinate information to the pixel of i image, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU, GPU calculates i+1 image according to mean shift algorithm simultaneously, obtaining initial RGB colouring information and initial coordinate information, and pass it back CPU, the parameter in zone comprises the color average in pixel number in the number, zone in zone and zone;
GPU calculates the parameter in described zone according to FCM Algorithms, to obtain the final segmentation result of i image, and send it to CPU, CPU carries out cluster according to RGB colouring information and the coordinate information of described i+1 image to the pixel of i+1 image simultaneously, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU;
CPU judges whether i+2 is less than or equal to the quantity of described a plurality of images;
If, then CPU exports the final segmentation result of i image, read i+2 image, the parameter of i+2 image of initialization is distributed CPU and the needed memory headroom of GPU, and i+2 image passed to GPU, GPU calculates the parameter in the zone of described i+1 image according to FCM Algorithms simultaneously, obtaining the final segmentation result of i+1 image, and send it to CPU
CPU judges whether i+3 is less than or equal to the quantity of described a plurality of images;
If, then CPU exports the final segmentation result of i+1 image, read i+3 image, the parameter of i+3 image of initialization, distribute CPU and the needed memory headroom of GPU, and i+3 image passed to GPU, GPU calculates i+2 image according to mean shift algorithm simultaneously, obtaining initial RGB colouring information and initial coordinate information, and pass it back CPU;
I=i+2 is set;
Turn back to described CPU and according to described initial RGB colouring information and described initial coordinate information the pixel of i image is carried out cluster, thereby form a plurality of zones, the parameter of statistical regions, and send it to GPU, GPU calculates i+1 image according to mean shift algorithm simultaneously, obtaining initial RGB colouring information and initial coordinate information, and pass it step of CPU back;
If i+3 is not the quantity that is less than or equal to described a plurality of images, then CPU exports the final segmentation result of i+1 image, simultaneously GPU calculates i+2 image according to mean shift algorithm, obtaining initial RGB colouring information and initial coordinate information, and passes it back CPU;
I=i+2 is set;
CPU carries out cluster according to described initial RGB colouring information and described initial coordinate information to the pixel of i image, thereby forms a plurality of zones, the parameter of statistical regions, and send it to GPU;
GPU calculates the parameter in the zone of described i image according to FCM Algorithms, obtaining the final segmentation result of i image, and sends it to CPU;
CPU exports the final segmentation result of i image, and process ends;
If i+2 is not the quantity that is less than or equal to described a plurality of images, then CPU exports the final segmentation result of i image, GPU calculates the parameter in the zone of described i+1 image according to FCM Algorithms simultaneously, obtaining the final segmentation result of i+1 image, and send it to CPU;
I=i+1 is set;
CPU exports the final segmentation result of i image, and process ends.
5. image set dividing method according to claim 4 is characterized in that, CPU is poly-according to the initial RGB colouring information pixel RGB colouring information is identical and adjacent with initial coordinate information to be a class.
6. image set dividing method according to claim 4 is characterized in that,
The final segmentation result of described image comprises final RGB colouring information and final coordinate information.
CN 201110267364 2011-09-09 2011-09-09 Image segmentation method and image set segmentation method Expired - Fee Related CN102314688B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 201110267364 CN102314688B (en) 2011-09-09 2011-09-09 Image segmentation method and image set segmentation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 201110267364 CN102314688B (en) 2011-09-09 2011-09-09 Image segmentation method and image set segmentation method

Publications (2)

Publication Number Publication Date
CN102314688A CN102314688A (en) 2012-01-11
CN102314688B true CN102314688B (en) 2013-01-23

Family

ID=45427826

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 201110267364 Expired - Fee Related CN102314688B (en) 2011-09-09 2011-09-09 Image segmentation method and image set segmentation method

Country Status (1)

Country Link
CN (1) CN102314688B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103617626A (en) * 2013-12-16 2014-03-05 武汉狮图空间信息技术有限公司 Central processing unit (CPU) and ground power unit (GPU)-based remote-sensing image multi-scale heterogeneous parallel segmentation method
CN103744736B (en) * 2014-01-09 2018-10-02 深圳Tcl新技术有限公司 The method and Linux terminal of memory management
CN105261004B (en) * 2015-09-10 2018-03-06 西安电子科技大学 Fuzzy C-mean algorithm image partition method based on average drifting and neighborhood information
CN114339412B (en) * 2021-12-30 2024-02-27 咪咕文化科技有限公司 Video quality enhancement method, mobile terminal, storage medium and device
CN114494253B (en) * 2022-04-14 2022-07-15 江苏智云天工科技有限公司 Method and device for industrial quality inspection

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7697756B2 (en) * 2004-04-09 2010-04-13 Siemens Medical Solutions Usa, Inc. GPU accelerated multi-label image segmentation (MLS)
CN101916432B (en) * 2010-07-29 2012-05-02 武汉大学 Parallelization method for image segmentation algorithm based on mean shift
CN101964911B (en) * 2010-10-09 2012-10-17 浙江大学 Ground power unit (GPU)-based video layering method

Also Published As

Publication number Publication date
CN102314688A (en) 2012-01-11

Similar Documents

Publication Publication Date Title
CN111242208B (en) Point cloud classification method, segmentation method and related equipment
CN102314688B (en) Image segmentation method and image set segmentation method
Sun et al. Automatic remotely sensed image classification in a grid environment based on the maximum likelihood method
Chen A tutorial on kernel density estimation and recent advances
CN104240244B (en) A kind of conspicuousness object detecting method based on communication mode and manifold ranking
CN103020993B (en) Visual saliency detection method by fusing dual-channel color contrasts
CN103593855B (en) The image partition method of cluster is estimated based on particle group optimizing and space length
CN107909015A (en) Hyperspectral image classification method based on convolutional neural networks and empty spectrum information fusion
CN102999926B (en) A kind of image vision significance computational methods merged based on low-level image feature
CN103839269A (en) Image segmentation method based on quaternion and fuzzy C-means clustering
CN102982544B (en) Many foreground object image interactive segmentation method
CN103049241A (en) Method for improving computation performance of CPU (Central Processing Unit) +GPU (Graphics Processing Unit) heterogeneous device
CN102013093A (en) High resolution remote sensing image segmentation method based on Gram-Schmidt fusion and locally excitatory globally inhibitory oscillator networks (LEGION)
CN108596885B (en) CPU + FPGA-based rapid SAR image change detection method
CN105373517A (en) Spark-based distributed matrix inversion parallel operation method
CN103942774A (en) Multi-target collaborative salient-region detection method based on similarity propagation
CN103400368A (en) Parallel rapid SAR image segmentation method based on graph theory and superpixel
CN102622761B (en) Image segmentation method based on similarity interaction mechanism
CN101520899B (en) Method for parallel reconstruction of cone beam CT three-dimension images
CN103886335A (en) Polarized SAR image classifying method based on fuzzy particle swarms and scattering entropy
KR20110037183A (en) Power controllable computer system combining neuro-fuzzy system and parallel processor, method and apparatus for recognizing objects using the computer system in images
CN104299201B (en) Image reconstruction method based on heredity sparse optimization
CN109410136B (en) Color homogenizing method and processing device based on shortest transmission path
CN107423771A (en) A kind of two phase method for detecting change of remote sensing image
CN105046632A (en) Space-time efficient binary image binary logic operation method

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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20130123

Termination date: 20150909

EXPY Termination of patent right or utility model