CN106023212A - Super-pixel segmentation method based on pyramid layer-by-layer spreading clustering - Google Patents
Super-pixel segmentation method based on pyramid layer-by-layer spreading clustering Download PDFInfo
- Publication number
- CN106023212A CN106023212A CN201610348500.3A CN201610348500A CN106023212A CN 106023212 A CN106023212 A CN 106023212A CN 201610348500 A CN201610348500 A CN 201610348500A CN 106023212 A CN106023212 A CN 106023212A
- Authority
- CN
- China
- Prior art keywords
- pyramid
- cluster
- layer
- pixel
- distance
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20156—Automatic seed setting
Abstract
The invention discloses a super-pixel segmentation method based on pyramid layer-by-layer spreading clustering. According to the method, a structure based on a pyramid layer-by-layer spreading clustering center is used to implement clustering, a compactness sensitive minimal barrier distance serves as measurement distance of clustering, a super-pixel segmentation result is obtained, and the minimal barrier distance is adjusted to generate super-pixels of different compact degrees. The method can be used to realize accurate super-pixel segmentation in high efficient; based on seed point clustering, the new structure based on the pyramid layer-by-layer spreading clustering center is used, the compactness sensitive minimal barrier distance serves as the measurement distance of clustering, and the accurate super-pixel segmentation result can be obtained; and the minimal barrier distance is adjusted to serve as one parameter of measured distance of clustering, and the super-pixels of different compact degrees are generated flexibly.
Description
Technical field
The invention belongs to digital image processing techniques field, particularly relate to a kind of based on pyramid successively propagation clustering super
Pixel dividing method.
Background technology
Super-pixel segmentation is since Ren and Malik in 2003 proposes, and one be increasingly becoming in computer vision field is not
The basic module that can or lack.Its research purpose is to be split input picture by technological means, and guarantee simultaneously is partitioned into
Block size is basically identical, and the edge of block is consistent with contour of object.It is " super that these results being partitioned into i.e. title are mentioned
Pixel ", the atomic unit that super-pixel is split frequently as image, semantic, or the input of other computer vision algorithms make, the most permissible
It is greatly enhanced precision and the efficiency of other algorithms.Through the development of more than ten years, super-pixel segmentation problem has had comparison many
Correlational study, but super-pixel segmentation yet suffers from some inconvenience in actual applications.These inconvenience are mainly reflected in super-pixel
In shape, regular, compact super-pixel is very important in a lot of computer vision algorithms make.Super-pixel is split according to being adopted
The difference of method, substantially can be divided into two kinds: one is based on energy optimizing model, one is to cluster based on seed points.
Super-pixel segmentation problem is modeled an object function by super-pixel partitioning algorithm based on energy optimizing model, utilizes numerical value side
Method optimizes this object function.It generally can obtain preferable global segmentation result, but the super-pixel shape of this kind of method generation
Often very irregular, mostly has the biggest time cost simultaneously, causes in actual applications as other computer vision algorithms make
Pretreatment time, it is impossible to meet the requirement of efficiency.Based on seed points cluster super-pixel partitioning algorithm by full figure with a seed
Point set is combined into initial, clusters full images vegetarian refreshments, and cluster result is i.e. super-pixel segmentation result.The usual efficiency of this type of method
The highest, the shape of super-pixel is the compactest simultaneously, but segmentation precision is generally difficult to meet actual requirement.
There is the normal very irregular of super-pixel shape produced in existing superpixel segmentation method, the cycle is longer, inefficient,
Segmentation precision is relatively low.
Summary of the invention
It is an object of the invention to provide a kind of superpixel segmentation method based on pyramid successively propagation clustering, it is intended to solve
Certainly there is the normal very irregular of super-pixel shape produced in existing superpixel segmentation method, and the cycle is longer, inefficient, segmentation essence
Spend relatively low problem.
The present invention is achieved in that a kind of superpixel segmentation method based on pyramid successively propagation clustering, described base
Superpixel segmentation method in pyramid successively propagation clustering is carried out by structure based on pyramid successively propagation clustering center
Cluster and using the sensitive minimum distance of obstacle of compactedness as the distance measure of cluster, obtains the segmentation result of super-pixel, and energy
By the minimum distance of obstacle of regulation, generate the super-pixel with different compactedness degree.
Further, described superpixel segmentation method based on pyramid successively propagation clustering also includes carrying out input picture
Smothing filtering, obtains the image of smoothing processing.
Further, described pyramidal zoom factor selects 0.1～0.9.
Further, the space constraint item that described minimum distance of obstacle introduces, it may be assumed that
F (I, τ)=B (I, τ)+α × d (τ0, τt)；
Wherein, F (I, τ) is i.e. distance measure, and I is image, and τ is a 2D path in image.B (I, τ) is minimum obstacle
Distance, d (τ0, τt) it is the starting point Euclidean distance to terminal of path τ, α is the compactedness factor, by regulation α balance space Europe
Formula distance and the weight relationship of minimum distance of obstacle.
Further, the cluster centre of every layer is the weighted center correspondence position at this layer of last layer cluster result, weighting
Mode is the distance power as each pixel of the cluster centre utilizing each pixel its nearest neighbours obtained in cluster process
Weight, is weighted summation, and obtains the new cluster centre of next layer after being multiplied by the inverse of pyramid zoom factor.
The cluster centre of every layer is the weighted center of last layer cluster result correspondence position in this layer.Last layer clusters
The weighted center of result is specifically:
Wherein s represents that certain clusters, and c (s) represents this cluster centre clustered, and p (i) represents each picture belonging to this cluster
Vegetarian refreshments position, w (i) represents the weighting weight of each pixel, compact used here as the corresponding cluster centre of each pixel
Property sensitive minimum distance of obstacle as its weight, Ψ is normalization factor；
Finally carry out the weighted center of upper strata cluster result obtaining this layer of new cluster by pyramid scaling scaling
Center, is specifically multiplied by pyramid zoom factor by the weighted center of upper strata cluster result.
Further, described superpixel segmentation method based on pyramid successively propagation clustering comprises the following steps:
Step one, uses the BoxFilter of 5x5 that input picture I is carried out smothing filtering, obtains Is；
Step 2: with IsImage pyramid is constructed for the bottomK ∈ 1,2 ..., v}, v are the pyramid numbers of plies, will
The length and width of every first order image are reduced into original 0.5, respectively obtain v width image；WhereinIt is the artwork of the bottom,It is to push up most
Layer；
Step 3: at the top layer of image pyramidOn be distributed as the equally distributed seed points of law generation with grid；
Step 4: this layer of pixel is clustered around cluster centre, and utilize the minimum distance of obstacle of compactedness sensitivity
As the distance measure of cluster, specifically, a space constraint item is introduced for minimum distance of obstacle, it may be assumed that
F (I, τ)=B (I, τ)+α × d (τ0, τt)；
Wherein, F (I, τ) is i.e. the distance measure proposed, and I is image, and τ is a 2D path in image, and B (I, τ) is
Little distance of obstacle, d (τ0, τt) it is the starting point Euclidean distance to terminal of path τ, α is the compactedness factor；
Step 5: to described image pyramidThe most successively cluster, the cluster centre of every layer
It it is the weighted center correspondence position at this layer of last layer cluster result；Concrete weighting scheme is to utilize to obtain in cluster process
The distance of the cluster centre of each pixel its nearest neighbours arrived, as the weight of each pixel, is weighted summation, and is multiplied by gold
The new cluster centre of next layer is obtained after the inverse of word tower zoom factor；
Step 6: to the pyramid bottomCluster result in each class do different labellings after, be i.e. super-pixel
Segmentation result.
The superpixel segmentation method based on pyramid successively propagation clustering that the present invention provides, can be efficiently obtained by accurately
Super-pixel segmentation, and can by regulation one parameter, be flexibly generated the super-pixel with different compactedness degree；Based on kind
Son point cluster, by a kind of new structure based on pyramid successively propagation clustering center, coordinates the minimum sensitive with compactedness
Distance of obstacle, as the distance measure of cluster, can obtain the segmentation result of accurate super-pixel, and can be by one ginseng of regulation
Number, is flexibly generated the super-pixel with different compactedness degree.
The present invention has the advantage that compared to prior art
1) compared to super-pixel partitioning algorithm based on energy optimizing model, the present invention is highly efficient.Based on energy-optimised
Although the method for model can produce preferable global segmentation result, but during optimizing, can spend bigger time cost.Phase
Instead, the present invention can be during pyramid be successively propagated, and gradually Optimized Segmentation result highly shortened the operation time, with
Time, also can produce the precise results similar with based on energy optimizing method segmentation effect.
2) compared to other super-pixel partitioning algorithms clustered based on seed points, the present invention has highly efficient, the cleverest
The feature lived.Different from other methods based on seed points cluster, invention introduces pyramid successively mechanism of transmission, it is to avoid
In other methods based on seed points cluster, iteration updates the process of cluster centre, is effectively improved operational efficiency, the most not
Loss segmentation precision.On the other hand, in cluster process, we use the minimum distance of obstacle of compactedness sensitivity as cluster
Distance measure, this estimates can balanced division precision and super-pixel regular shape degree effectively.
Accompanying drawing explanation
Fig. 1 is the superpixel segmentation method flow chart based on pyramid successively propagation clustering that the embodiment of the present invention provides.
Fig. 2 is the grid distribution schematic diagram that the embodiment of the present invention provides.
Detailed description of the invention
In order to make the purpose of the present invention, technical scheme and advantage clearer, below in conjunction with embodiment, to this
Bright it is further elaborated.Should be appreciated that specific embodiment described herein, and need not only in order to explain the present invention
In limiting the present invention.
Below in conjunction with the accompanying drawings the application principle of the present invention is explained in detail.
As it is shown in figure 1, the superpixel segmentation method based on pyramid successively propagation clustering of the embodiment of the present invention include with
Lower step:
S101: input piece image, uses smoothing filter that the image of input is carried out process and obtains；
S102: construct image pyramid for the bottom with the image of smoothing processing；
S103: generate the seed points being evenly distributed on the top layer of the image pyramid obtained, and be combined into this seed points collection
The cluster centre of this layer；
S104: utilize the minimum distance of obstacle distance measure as cluster of compactedness sensitivity, around cluster centre to this
Layer pixel clusters；
S105: the most successively cluster described image pyramid, on the cluster centre of every layer is
The weighted center of one layer of cluster result is at the correspondence position of this layer；
The cluster result of S106: the bottom is as super-pixel segmentation result.
The cluster centre of every layer is the weighted center of last layer cluster result correspondence position in this layer.Last layer clusters
The weighted center of result is specifically:
Wherein s represents that certain clusters, and c (s) represents this cluster centre clustered, and p (i) represents each picture belonging to this cluster
Vegetarian refreshments position, w (i) represents the weighting weight of each pixel, compact used here as the corresponding cluster centre of each pixel
Property sensitive minimum distance of obstacle as its weight, Ψ is normalization factor；
Finally carry out the weighted center of upper strata cluster result obtaining this layer of new cluster by pyramid scaling scaling
Center, is specifically multiplied by pyramid zoom factor by the weighted center of upper strata cluster result.
Below in conjunction with specific embodiment, the application principle of the present invention is further described.
Embodiment 1:
Input piece image, said method comprising the steps of:
Input picture is labeled as I by S1.
S2 use smoothing filter carries out process to I and obtains Is。
S3 is with IsImage pyramid is constructed for the bottomK ∈ 1,2 ..., v}, v are the pyramid numbers of plies.
The top layer of the image pyramid that S4 obtains in S3 stepThe seed points that upper generation is evenly distributed, and with this seed points
Collection is combined into the cluster centre of this layer.
S5 utilizes the minimum distance of obstacle distance measure as cluster of compactedness sensitivity, around cluster centre to this layer of picture
Element clusters.
S6 is to described image pyramidS5 step is the most successively utilized to cluster, the cluster of every layer
Center is the weighted center correspondence position at this layer of last layer cluster result.
S7 using the cluster result of the bottom in S6 step as super-pixel segmentation result.
1) the smoothing filter main purpose in S2 step is to filter granularity noise, it is however generally that quickly BoxFilter
Can meet requirement, other such as gaussian filtering, bilateral filterings etc. the most all can be applicable to this.
2) at the structural map of S3 step as during pyramidal, pyramidal zoom factor can be fixed as 0.5, it is possible to
Selecting the number between 0.1～0.9 with more actual application, zoom factor is the highest means that the pyramid number of plies is the most.
3) the super-pixel number ultimately generated depends on the number generating seed points in S4 step.It is evenly distributed kind in generation
During son point, the distribution of general seed points is chosen as grid distribution, it is possible to according to different application demands, adjusts seed points
The regularity of distribution.
4) distance measure in S5 step cluster process, can use different distance measures according to the difference of application,
The minimum distance of obstacle that the compactedness proposed in the present invention and use is sensitive, can be by one parameter balance space length of regulation
With the weight relationship of color distance, by regulating it, the super-pixel with different compactedness degree can be flexibly generated.
5), in S6 step, when calculating the weighted center of last layer cluster result, difference can be selected according to different needs
Weight Algorithm.
Embodiment 2
After obtaining a width input picture I, using the method for the present invention to process it, it specifically comprises the following steps that
Step 1: use the BoxFilter of 5x5 that input picture I is carried out smothing filtering, obtain Is。
Step 2: with IsImage pyramid is constructed for the bottomK ∈ 1,2 ..., v}, v are the pyramid numbers of plies, will be every
The length and width of first order image are reduced into original 0.5, respectively obtain v width image.WhereinIt is the artwork of the bottom,It is to push up most
Layer.
Step 3: at the top layer of image pyramidOn be distributed as the equally distributed seed points of law generation, grid with grid
The spacing of distribution depends on the super-pixel number wishing to ultimately generate, grid schematic diagram such as Fig. 2；
Step 4: this layer of pixel is clustered around cluster centre, and utilize the minimum distance of obstacle of compactedness sensitivity to make
Distance measure for cluster.Specifically, a space constraint item is introduced for minimum distance of obstacle, it may be assumed that
F (I, τ)=B (I, τ)+α × d (τ0, τt)；
Wherein, F (I, τ) is i.e. the distance measure that the present invention proposes, and I is image, and τ is a 2D path in image.B (I,
τ) it is minimum distance of obstacle, d (τ0, τt) it is the starting point Euclidean distance to terminal of path τ, α is the compactedness factor, by adjusting
Joint α can be with balance space Euclidean distance and the weight relationship of minimum distance of obstacle.
Step 5: to described image pyramidS5 step is the most successively utilized to cluster, every layer
Cluster centre is the weighted center correspondence position at this layer of last layer cluster result.Concrete weighting scheme is to utilize in cluster
During the distance of the cluster centre of each pixel its nearest neighbours that obtains as the weight of each pixel, be weighted summation,
And after being multiplied by the inverse of pyramid zoom factor, obtain the new cluster centre of next layer.
Step 6: to the pyramid bottomCluster result in each class do different labellings after, be i.e. that super-pixel is divided
Cut result.
The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all essences in the present invention
Any amendment, equivalent and the improvement etc. made within god and principle, should be included within the scope of the present invention.
Claims (7)
1. a superpixel segmentation method based on pyramid successively propagation clustering, it is characterised in that described based on pyramid by
The superpixel segmentation method of Es-region propagations cluster carries out clustering and with tightly by structure based on pyramid successively propagation clustering center
The minimum distance of obstacle of gathering property sensitivity, as the distance measure of cluster, obtains the segmentation result of super-pixel, and can be by regulating
Little distance of obstacle, generates the super-pixel with different compactedness degree.
2. superpixel segmentation method based on pyramid successively propagation clustering as claimed in claim 1, it is characterised in that described
Superpixel segmentation method based on pyramid successively propagation clustering also includes input picture is carried out smothing filtering, obtains smooth place
The image of reason.
3. superpixel segmentation method based on pyramid successively propagation clustering as claimed in claim 1, it is characterised in that described
Pyramidal zoom factor selects 0.1～0.9.
4. superpixel segmentation method based on pyramid successively propagation clustering as claimed in claim 1, it is characterised in that described
The space constraint item that minimum distance of obstacle introduces, it may be assumed that
F (I, τ)=B (I, τ)+α × d (τ0, τt)；
Wherein, F (I, τ) is i.e. distance measure, and I is image, and τ is a 2D path in image；B (I, τ) is minimum distance of obstacle,
d(τ0, τt) be the starting point Euclidean distance to terminal of path τ, α is the compactedness factor, by regulation α balance space European away from
From the weight relationship with minimum distance of obstacle.
5. superpixel segmentation method based on pyramid successively propagation clustering as claimed in claim 1, it is characterised in that every layer
Cluster centre be the weighted center correspondence position at this layer of last layer cluster result, weighting scheme is to utilize at cluster process
In the distance of the cluster centre of each pixel its nearest neighbours that obtains as the weight of each pixel, be weighted summation, and take advantage of
To obtain the new cluster centre of next layer after the inverse of pyramid zoom factor.
6. superpixel segmentation method based on pyramid successively propagation clustering as claimed in claim 5, it is characterised in that every layer
Cluster centre be the weighted center of last layer cluster result correspondence position in this layer, in the weighting of last layer cluster result
The heart is specifically:
Wherein s represents that certain clusters, and c (s) represents this cluster centre clustered, and p (i) represents each pixel belonging to this cluster
Position, w (i) represents the weighting weight of each pixel, and the compactedness used here as the corresponding cluster centre of each pixel is quick
The minimum distance of obstacle of sense is as its weight, and Ψ is normalization factor；
Finally carry out the weighted center of upper strata cluster result obtaining this layer of new cluster centre by pyramid scaling scaling,
Specifically the weighted center of upper strata cluster result is multiplied by pyramid zoom factor.
7. superpixel segmentation method based on pyramid successively propagation clustering as claimed in claim 1, it is characterised in that described
Superpixel segmentation method based on pyramid successively propagation clustering comprises the following steps:
Step one, uses the BoxFilter of 5x5 that input picture I is carried out smothing filtering, obtains Is；
Step 2: with IsImage pyramid is constructed for the bottomV is the pyramid number of plies, will be each
The length and width of level image are reduced into original 0.5, respectively obtain v width image；WhereinIt is the artwork of the bottom,It it is top；
Step 3: at the top layer of image pyramidOn be distributed as the equally distributed seed points of law generation with grid；
Step 4: this layer of pixel is clustered around cluster centre, and utilize the minimum distance of obstacle conduct of compactedness sensitivity
The distance measure of cluster, specifically, introduces a space constraint item for minimum distance of obstacle, it may be assumed that
F (I, τ)=B (I, τ)+α × d (τ0+τt)；
Wherein, F (I, τ) is i.e. the distance measure proposed, and I is image, and τ is a 2D path in image, and B (I, τ) is minimum barrier
Hinder distance, d (τ0, τt) it is the starting point Euclidean distance to terminal of path τ, α is the compactedness factor；
Step 5: to described image pyramidThe most successively cluster, on the cluster centre of every layer is
The weighted center of one layer of cluster result is at the correspondence position of this layer；Concrete weighting scheme utilizes and obtains in cluster process
The distance of the cluster centre of each pixel its nearest neighbours, as the weight of each pixel, is weighted summation, and is multiplied by pyramid
The new cluster centre of next layer is obtained after the inverse of zoom factor；
Step 6: to the pyramid bottomCluster result in each class do different labellings after, be i.e. super-pixel segmentation knot
Really.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610348500.3A CN106023212A (en) | 2016-05-24 | 2016-05-24 | Super-pixel segmentation method based on pyramid layer-by-layer spreading clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610348500.3A CN106023212A (en) | 2016-05-24 | 2016-05-24 | Super-pixel segmentation method based on pyramid layer-by-layer spreading clustering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106023212A true CN106023212A (en) | 2016-10-12 |
Family
ID=57094035
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610348500.3A Pending CN106023212A (en) | 2016-05-24 | 2016-05-24 | Super-pixel segmentation method based on pyramid layer-by-layer spreading clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106023212A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765267A (en) * | 2018-05-25 | 2018-11-06 | 青岛九维华盾科技研究院有限公司 | A kind of digital camouflage generation method and system based on error diffusion dither algorithm |
| CN108875798A (en) * | 2018-05-29 | 2018-11-23 | 电子科技大学 | A kind of super-pixel grade feature extracting method based on spatial pyramid pond |
| CN108932729A (en) * | 2018-08-17 | 2018-12-04 | 安徽大学 | A kind of minimum distance of obstacle Weight tracking method |
| CN109907824A (en) * | 2019-03-11 | 2019-06-21 | 杭州市红十字会医院 | A kind of intelligence needle knife treatment system |
| CN110738688A (en) * | 2019-10-25 | 2020-01-31 | 中国人民解放军国防科技大学 | novel infrared ultra-weak moving target detection method |
| CN116028838A (en) * | 2023-01-09 | 2023-04-28 | 广东电网有限责任公司 | Clustering algorithm-based energy data processing method and device and terminal equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102509097A (en) * | 2011-09-29 | 2012-06-20 | 北京新媒传信科技有限公司 | Method and device for image segmentation |
| CN103353987A (en) * | 2013-06-14 | 2013-10-16 | 山东大学 | Superpixel segmentation method based on fuzzy theory |
| CN103984953A (en) * | 2014-04-23 | 2014-08-13 | 浙江工商大学 | Cityscape image semantic segmentation method based on multi-feature fusion and Boosting decision forest |
| CN104376330A (en) * | 2014-11-19 | 2015-02-25 | 西安电子科技大学 | Polarization SAR image ship target detection method based on superpixel scattering mechanism |
| CN104680538A (en) * | 2015-03-09 | 2015-06-03 | 西安电子科技大学 | SAR image CFAR target detection method on basis of super pixels |
-
2016
- 2016-05-24 CN CN201610348500.3A patent/CN106023212A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102509097A (en) * | 2011-09-29 | 2012-06-20 | 北京新媒传信科技有限公司 | Method and device for image segmentation |
| CN103353987A (en) * | 2013-06-14 | 2013-10-16 | 山东大学 | Superpixel segmentation method based on fuzzy theory |
| CN103984953A (en) * | 2014-04-23 | 2014-08-13 | 浙江工商大学 | Cityscape image semantic segmentation method based on multi-feature fusion and Boosting decision forest |
| CN104376330A (en) * | 2014-11-19 | 2015-02-25 | 西安电子科技大学 | Polarization SAR image ship target detection method based on superpixel scattering mechanism |
| CN104680538A (en) * | 2015-03-09 | 2015-06-03 | 西安电子科技大学 | SAR image CFAR target detection method on basis of super pixels |
Non-Patent Citations (5)
| Title |
|---|
| HAJAR MOMENI等: ""Fast face recognition using a combination of image pyramid and hierarchical clustering algorithms "", 《INTERNATIONAL CONFERENCE ON WIRELESS COMMUNICATIONS & SIGNAL PROCESSING, 2009. WCSP 2009》 * |
| 宋琪等: ""一种基于分形的金字塔快速图象分割方法"", 《计算机应用与软件》 * |
| 王小乐等: ""一种处理障碍约束的聚类算法"", 《计算机应用》 * |
| 靳明明: ""基于聚类算法胆结石CT图像分割的研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 黄鹏飞等: ""拉普拉斯加权聚类算法"", 《电子学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765267A (en) * | 2018-05-25 | 2018-11-06 | 青岛九维华盾科技研究院有限公司 | A kind of digital camouflage generation method and system based on error diffusion dither algorithm |
| CN108875798A (en) * | 2018-05-29 | 2018-11-23 | 电子科技大学 | A kind of super-pixel grade feature extracting method based on spatial pyramid pond |
| CN108875798B (en) * | 2018-05-29 | 2022-06-24 | 电子科技大学 | Super-pixel-level feature extraction method based on spatial pyramid pooling |
| CN108932729A (en) * | 2018-08-17 | 2018-12-04 | 安徽大学 | A kind of minimum distance of obstacle Weight tracking method |
| CN108932729B (en) * | 2018-08-17 | 2021-06-04 | 安徽大学 | Minimum obstacle distance weighted tracking method |
| CN109907824A (en) * | 2019-03-11 | 2019-06-21 | 杭州市红十字会医院 | A kind of intelligence needle knife treatment system |
| CN109907824B (en) * | 2019-03-11 | 2020-12-22 | 杭州市红十字会医院 | Intelligent needle-knife treatment system |
| CN110738688A (en) * | 2019-10-25 | 2020-01-31 | 中国人民解放军国防科技大学 | novel infrared ultra-weak moving target detection method |
| CN116028838A (en) * | 2023-01-09 | 2023-04-28 | 广东电网有限责任公司 | Clustering algorithm-based energy data processing method and device and terminal equipment |
| CN116028838B (en) * | 2023-01-09 | 2023-09-19 | 广东电网有限责任公司 | Clustering algorithm-based energy data processing method and device and terminal equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106023212A (en) | Super-pixel segmentation method based on pyramid layer-by-layer spreading clustering | |
| Wang et al. | Design flow of accelerating hybrid extremely low bit-width neural network in embedded FPGA | |
| Xin et al. | Centroidal power diagrams with capacity constraints: Computation, applications, and extension | |
| Yu et al. | Pu-net: Point cloud upsampling network | |
| TW201835817A (en) | Apparatus and method for designing super resolution deep convolutional neural networks | |
| WO2018032763A1 (en) | Method and device for generating thermodynamic diagram | |
| Xia et al. | SparkNoC: An energy-efficiency FPGA-based accelerator using optimized lightweight CNN for edge computing | |
| Kong et al. | Pixel-wise attentional gating for scene parsing | |
| CN104240299B (en) | Remeshing method based on maximal Poisson-disk sampling | |
| CN109299685A (en) | Deduction network and its method for the estimation of human synovial 3D coordinate | |
| CN105024886B (en) | A kind of Fast W eb service QoS Forecasting Methodologies based on user metadata | |
| CN108960251A (en) | A kind of images match description generates the hardware circuit implementation method of scale space | |
| Zhu et al. | Sensor network localization using sensor perturbation | |
| CN109948575A (en) | Eyeball dividing method in ultrasound image | |
| CN107133877B (en) | Method for mining overlapped communities in network | |
| CN109146792A (en) | Chip image super resolution ratio reconstruction method based on deep learning | |
| Wu et al. | CASR: a context-aware residual network for single-image super-resolution | |
| Zint et al. | Generation of block structured grids on complex domains for high performance simulation | |
| Sheng et al. | Lfnat 2023 challenge on light field depth estimation: Methods and results | |
| Luo et al. | Reconstruction of missing flow field from imperfect turbulent flows by machine learning | |
| CN108629405A (en) | The method and apparatus for improving convolutional neural networks computational efficiency | |
| CN109408870A (en) | A kind of topological net generation method and electronic equipment based on boundary constraint | |
| CN106648883B (en) | Dynamic reconfigurable hardware acceleration method and system based on FPGA | |
| CN105045906A (en) | Estimation method and device of click rate of delivery information | |
| Wang et al. | Mesh optimization based on the centroidal voronoi tessellation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161012 |