WO2018086433A1 - 一种医学图像分割方法 - Google Patents
一种医学图像分割方法 Download PDFInfo
- Publication number
- WO2018086433A1 WO2018086433A1 PCT/CN2017/105431 CN2017105431W WO2018086433A1 WO 2018086433 A1 WO2018086433 A1 WO 2018086433A1 CN 2017105431 W CN2017105431 W CN 2017105431W WO 2018086433 A1 WO2018086433 A1 WO 2018086433A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- harmony
- new
- search
- image
- memory
- 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.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Definitions
- the invention relates to an image segmentation method, in particular to a medical image segmentation method, and belongs to the field of image processing.
- CT Computed Tomography
- MRI Megectic Resonance Imaging
- clustering is used in many practical problems such as data mining, image processing, bioinformatics, decision making and planning because of its simplicity, easy operation and good robustness.
- Clustering methods are also one of the methods often used for image segmentation. The main idea of this kind of method is to divide the features into multiple categories according to the characteristics in the image data according to some clustering algorithm ideas. The image features in the same class are similar, but the differences between the same classes are larger, and finally the clustering will be The class result is mapped into the original image to implement segmentation.
- Image clustering has become one of the important methods of image segmentation. It is very common to use cluster analysis to classify objects in medical images. The purpose of medical image segmentation is to obtain meaningful clusters in medical images, which lays the foundation for medical image recognition and mining. Conventional clustering can only be applied to structured, small-scale, low-dimensional data sets.
- Medical image segmentation can better provide scientific reference for medical staff to judge and diagnose the cause of the disease, which will greatly reduce the misdiagnosis rate caused by insufficient visual resolution of human beings or subjective clinical experience of medical personnel, and further improve medical images. Utilization rate. Proper clustering of medical images can also effectively manage a large image database. With the development of network technology, these useful image information can be shared with more peers.
- Partitioning clustering and hierarchical clustering are commonly used clustering algorithms.
- the partitioning clustering is based on the method of optimizing the objective function to realize the division between different categories.
- the commonly used method of dividing clustering is K. Mean (K-means), fuzzy C-means (FCM), nuclear fuzzy C-means (KFCM), expected maximum (EM), and so on.
- K-means Mean
- FCM fuzzy C-means
- KFCM nuclear fuzzy C-means
- EM expected maximum
- the main idea of hierarchical clustering is to cluster data according to the similarity between the elements to be clustered, according to the top-down or bottom-up algorithmic idea, by splitting or condensing between classes.
- Commonly used hierarchical clustering methods include single-linkage, complete-linkage, and average-linkage clustering methods.
- the standard community partitioning method is also a trend of clustering algorithm research.
- Shi and Malik proposed the concept of standard community partitioning, which refers to maximizing similar regions in a community, rather than similar parts. minimize.
- Maji et al. also proposed a biased standard community partitioning method, in which the method includes modifying the community in the standard method to the merged prophetic condition, and the improvement can be used for multi-feature clustering.
- C-GeNCut considers the color, brightness, texture and other characteristics of the image.
- the intermediate contour clue method is used to calculate the fitness between two pixels, and finally the image segmentation is realized.
- the methods proposed by Kernighan and Lin mainly use greedy algorithms to classify the original network according to the principle of optimizing the edges within the community and between communities.
- the spectral grading method based on the eigenvalues of Laplace graphs can be divided into sub-groups based on the similarity or strength of the connections between the nodes through the traditional algorithm of the community structure in the social network, and then the image segmentation can be realized.
- the above prior art only seeks to divide the optimal solution of the cluster as a whole, but fails to consider whether the whole and the part are optimal, nor can it describe the multi-target clustering method, and can not effectively solve most of the problems.
- the single objective function, single threshold, and the problem of component clustering are not considered in the clustering algorithm.
- the object of the present invention is to provide a medical image segmentation method that fully considers the whole image and each component and makes them optimal, in view of the limitations of the prior art described above, thereby effectively solving most clustering.
- the algorithm only has a single target optimization function and does not consider the defects of the component clustering effect, significantly improving the anti-noise ability and making clustering.
- the processed image is closer to the objective and real situation of each object, providing a more accurate image basis for medical diagnosis.
- the medical image segmentation method of the present invention performs the following basic steps after the computer reads the original image:
- Step A Set parameters: search scale N, cluster number K, optimal iteration number T in each cluster, harmony search algorithm parameter memory value matrix HMCR, fine-tuning probability PAR, pitch fine-tuning bandwidth bw, maximum iteration number Tmax;
- the basic principle is that the Harmony Search Algorithm (CHS) simulates the music creation in which the musicians rely on their own memories, by repeatedly adjusting the pitch of the orchestra instrument, and finally to a wonderful harmony state process, and the harmony search algorithm will tone the instrument.
- the harmony is analogous to the solution vector of the optimization problem.
- the algorithm introduces two main parameters, the memory value matrix HMCR, and the fine-tuning probability PAR. The idea of solving the optimization problem can be applied to the clustering algorithm.
- Step B Using the K-means algorithm to cluster the medical images, and divide them into K clusters to complete the initial segmentation of the image;
- the K-means algorithm first needs to assume the initial center of the class of the K class, then calculate the distance from each point to the cluster center and classify it as the class with the smallest distance, then recalculate each cluster center and iterate. The above steps are until the convergence condition.
- the number K of clusters can be obtained by itself according to the actual situation.
- Step C According to the initial segmentation result of the K-means algorithm in step B, the segmented each cluster is optimized by using a harmony search algorithm (CHS);
- CHS harmony search algorithm
- the Harmony Search Algorithm first generates an initial solution of the number of HMS equal to the harmony scale and puts it into the harmony memory; then, randomly searches for a new solution in the harmony memory, and the probability of generating a new connection is taken from the memory.
- the value matrix HMCR is determined.
- the new solution obtained from the harmony memory is locally perturbed by the fine tuning probability PAR.
- judge the value of the new solution objective function Whether it is better than the worst solution in the harmony memory, and if so, update the sound library and iterate until it reaches the predetermined number of iterations Tmax.
- HMS harmonies x 1 , x 2 , ..., x HMS are randomly generated into the harmony memory, and the acoustic memory is as follows:
- each tone of the new harmony x' i passes the following three
- the mechanism produces: (1) learning and sound memory; (2) pitch fine-tuning; (3) randomly selecting pitch.
- the variables are generated as follows:
- rand represents a uniformly distributed random number on (0, 1).
- Step D If the optimal solution is found in a certain cluster, stop the search in this cluster and record its optimal solution X best ; use the niche technology to generate a niche radius R with the optimal solution X best as the center. With X best as the center, R is the area formed by the radius as a taboo area (to prevent subsequent search points from entering the area for repeated search);
- a niche is an organizational structure in a particular environment.
- Niche technology divides each generation into several categories. Each class selects a number of individuals with greater fitness as a representative of a class to form a group, and then in the population, and between different populations, hybridization, Mutation produces a new generation of individual groups. At the same time, the pre-selection mechanism and the crowding mechanism or sharing mechanism are used to complete the task.
- Niche technology is often used when solving multi-peak problems and can find all the optimal solutions.
- Step E When the termination condition is not satisfied, N new search points are randomly generated in the re-search space, and if the new search point is in the “taboo area”, re-generate, and the process proceeds to step C to continue execution;
- Step F If the optimal solution is found in all the clusters or the number of iterations is greater than T, the current search is stopped, and according to the search result, the image is divided into images to complete the image segmentation, and the segmented image is output.
- the medical image segmentation method of the present invention is directed to the problem that the traditional clustering method only seeks to segment the overall optimal solution.
- the harmony search algorithm Based on the harmony search algorithm, the multi-cluster multiple clustering strategy is adopted, and the image is initially segmented first, and then The harmony search algorithm seeks to optimize each part of the segmentation, so that it is possible to find multiple parts while achieving the optimal solution, and this method does not cause conflicts between the parts.
- Fully considering the characteristics of neighborhood information and interval it can effectively segment medical images with high noise, which has higher accuracy, significantly improved anti-noise ability, and can retain true image information and gather for medical images. Classes have important practical significance and application value.
- Figure 1 is a flow chart of an embodiment of the present invention
- FIG. 2a and b are original diagrams and clustering result diagrams of the embodiment of Fig. 1, respectively.
- a brain MRI map is selected for analysis, and a brain MRI map 2a originally containing noise points is selected to illustrate the corresponding result after the implementation of the present invention.
- the specific steps are as follows:
- the medical image is segmented by K-means algorithm and divided into 3 clusters.
- each segmentation is optimized by using the harmony search algorithm (CHS) to find the optimal solution of each cluster;
- CHS harmony search algorithm
- the initial cluster number obtained in this embodiment is three types, and the optimization results of the initial cluster are [13, 82], [63, 183], [120, 146], respectively.
- the present invention is directed to the general clustering algorithm having only a single target optimization function and not considering the defects of the component clustering effect.
- the general clustering method has a single objective function, a single threshold, and does not consider the component clustering effect, based on Harmony search algorithm, will
- the K-means algorithm optimizes each cluster and image as a whole, and realizes harmony search in each region.
- the “taboo zone” is reasonably set to achieve more effective optimal search. It is practicable through the above examples.
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
本发明涉及一种医学图像分割方法,属于图像处理领域。本发明的方法基于和声搜索算法,经图像预处理、提取图像特征、图像预分割、全局最优和声搜索、图像部分区域内最优和声搜索等步骤实现。理论和实践证明,本发明妥善解决了聚类算法中单目标函数、单阈值、未考虑组成部分聚类效果的问题。采用本发明的方法后,不仅可以高效完成医学图像的分割,而且不丢失重要图像特征,因此可以获得高质量的医学分割图像,满足医护人员的诊断阅读需求。
Description
本发明涉及一种图像分割方法,尤其是一种医学图像分割方法,属于图像处理领域。
随着计算机断层扫描(Computed Tomography,CT)、核磁共振图像(Megnectic Resonance Imaging,MRI)等人体医学成像技术的迅速发展,医学图像在临床医学诊断中发挥着越来越重要的作用。
作为非监督学习算法,聚类因其简单、易操作、鲁棒性较好等优势运用于很多实际问题当中,如数据挖掘、图像处理、生物信息学、决策与规划等等方面。聚类方法也是常常用于图像分割的方法之一。该类方法主要思想是根据图像数据中的特征,按照某种聚类算法思想,将特征分为多种类别,同一类中图像特征相近,而不相同类之间的差异较大,最终将聚类结果映射到原图像中实现分割。
图像聚类已成为图像分割的重要方法之一,运用聚类分析对医学图像中的物体进行分类是十分常见的方法。医学图像分割的目的是得到医学图像中有意义的簇,为医学图像识别与挖掘奠定基础。常规聚类法只能适用于结构化的、小规模的、低维的数据集。
医学图像分割可以更好的在医务人员判断和诊断病情病因时提供科学参考,从而会大大减少因为人类本身视力分辨力不足或是医疗人员主观上临床经验不足产生的误诊率,进一步提高医学图像的利用率。对医学图像进行合理聚类还可以有效地管理庞大的图像数据库,随着网络技术的发展,能够和更多的同行共享这些有用的图片信息。
划分聚类和层次聚类是常用的聚类算法。划分聚类是根据优化目标函数的方法实现不同类别之间的划分,常用的划分聚类的方法有K
均值(K-means)、模糊C均值(FCM)、核模糊C均值(KFCM)、期望最大值(EM)等等。层次聚类主要思想是根据待聚类元素之间的相似性,根据自上而下的或自下而上的算法思想,通过对类之间的分裂或者是凝聚的方法对数据进行聚类,常用的层次聚类方法有single-linkage,complete-linkage以及average-linkage聚类方法等。现阶段研究中,标准社团划分方法也是聚类算法研究的一个趋势,Shi和Malik于2000年提出了标准社团划分的概念,这个概念是指使得一个社团中相似区域最大化,而非相似的部分最小化。Maji等人也提出了有偏的标准社团划分方法,其中该方法包括将标准方法中的社团修正为合并的先知条件,这种改进可以用于多特征的聚类。基于遗传算法的社团划C-GeNCut,考虑到图像的颜色,亮度,纹理等多种特征,通过中间轮廓线索的方法计算两个像素之间的适应度,最终实现图像分割。Kernighan和Lin提出的方法主要采用贪婪算法,根据使社团内部及社团间的边最优化的原则对原始的网络进行分类。基于Laplace图特征值的谱平分法通过社会网络中社团结构的一类传统算法,基于各个节点之间连接的相似性或者强度,把网络自然划分为各个子群,进而可以实现图像分割。
归纳起来,上述现有技术只在寻求分割聚类整体的最优解,而没能考虑整体与部分是否都达到最优,也未能有多目标聚类方法的描述,也无法有效解决大部分聚类算法中单目标函数、单阈值、未考虑组成部分聚类效果的问题。
发明内容
本发明的目的在于:针对上述现有技术存在的局限性,提出一种充分考虑到图像整体和各个组成部分、并使它们都能达到最优的医学图像分割方法,从而有效解决大部分聚类算法只有单一的目标优化函数并且未考虑组成部分聚类效果的缺陷,显著提高抗噪能力,使聚类
处理后的图像更接近于每个对象的客观、真实情况,为医疗诊断提供更为精准的图像依据。
为了达到上述目的,本发明的医学图像分割方法在计算机读入原始图像后,进行如下基本步骤:
步骤A.设置参数:搜索规模N,聚类个数K,每个簇中优化迭代次数T,和声搜索算法参数记忆取值矩阵HMCR,微调概率PAR,音调微调带宽bw,最大迭代次数Tmax;
其基本原理为:和声搜索算法(CHS)模拟了音乐创作中乐师们凭借自己的记忆,通过反复调整乐队乐器的音调,最终到一个美妙的和声状态的过程,和声搜索算法将乐器音调的和声类比于优化问题的解向量。算法引入两个主要参数,即记忆取值矩阵HMCR,微调概率PAR。其解答优化问题的思想可以运用在聚类算法中。
步骤B.采用K-means算法对医学图像进行聚类,将其分为K个簇,完成对图像的初始分割;
K-means算法首先需要假定K类聚的类初始中心,之后计算每个点到聚类中心的距离并将它归为距离最小的点所在的类,然后重新计算各个聚类中心,并迭代进行以上步骤直到收敛条件为止。聚类数K可以根据实际情况自行获取。
步骤C.根据步骤B中K-means算法的初始分割结果,对分割出来的每一个簇采用和声搜索算法(CHS)进行优化;
和声搜索算法(CHS)首先产生数量等于和声规模的HMS的初始解,并放入和声记忆库中;然后,在和声记忆库内随机搜索新解,产生新接的概率由记忆取值矩阵HMCR决定。再以微调概率PAR对取自和声记忆库内的新解进行局部扰动。最后,判断新解目标函数值
是否优于和声记忆库内的最差解,若是,则更新和声库,并不断迭代,直至达到预定迭代次数Tmax为止。
具体算法如下:
随机生成HMS个和声x1,x2,...,xHMS放入和声记忆库,和声记忆库如下:
生成新的和声x’=(x’1,x’2,...x’N),新和声的每一个音调x’i(i=1,2,...N)通过以下三种机理产生:(1)学习和声记忆库;(2)音调微调;(3)随机选择音调。变量的生成方式如下:
其中rand表示(0,1)上的均匀分布随机数。
其次,如果新的和声x’i来自和声记忆库,要对其进行音调微调,具体操作如下:
然后,对上述和声新解进行评估,如果优于和声记忆库中的函数值最差的一个,则将新解更新至和声记忆库中。具体操作如下:
重复上述过程,直至迭代次数达到Tmax。
步骤D.如果在某一个簇中找到最优解,则停止在这一个簇的搜索,记录其最优解Xbest;以最优解Xbest为中心,利用小生境技术产生一个小生境半径R,以Xbest为中心,R为半径形成的区域作为禁忌区(防止后续搜索点进入该区域进行重复搜索);
小生境是指特定环境下的一种组织结构。小生境技术就是将每一代个体划分为若干类,每个类中选出若干适应度较大的个体作为一个类的优秀代表组成一个群,再在种群中,以及不同种群中之间,杂交,变异产生新一代个体群。同时采用预选择机制和排挤机制或分享机制完成任务。小生境技术经常在求解多峰值问题时使用,能找到全部的最优解。
步骤E.未满足终止条件时,重新再搜索空间中随机生成N个新的搜索点,如果新的搜索点在“禁忌区”中,则重新生成,转至步骤C继续执行;
步骤F.如果在所有簇中都找到了最优解或者迭代次数大于T,则停止当前搜索,并根据搜索结果,映射至图像中完成图像分割,输出分割后的图像。
本发明的医学图像分割方法针对传统聚类方法中只在寻求分割整体的最优解的问题,基于和声搜索算法,采用多簇多次聚类的策略,先对图像进行初始分割,再对分割出的每个部分中都采用和声搜索算法寻求最优,这样就有可能找到是多个部分同时达到最优解,并且这种方法不会造成各个部分之间的冲突。充分考虑了邻域信息、区间的性质等特性,可以针对噪声较大的医学图像进行有效地分割,其准确率更高,抗噪声能力显著增强,能够保留真实的图像信息,对医学图像的聚类具有重要现实意义和应用价值。
图1为本发明一个实施例的流程图;
图2a和b分别为图1实施例的原图和聚类结果图。
下面主要结合附图和具体实施例对医学图像分割方法作进一步的说明。
本实施例选取脑MRI图来进行分析,选取原始含有噪声点的脑MRI图2a来说明本发明实施后的相应结果,结合图1,具体步骤如下:
A.计算机读入脑MRI图的原始图像,设置搜索规模N=100,聚类个数K=3,每个簇中优化迭代次数T=100,和声搜索算法参数记忆取值矩阵HMCR=I,微调概率PAR=0.01,音调微调带宽bw=0.0001,最大迭代次数Tmax=200。
B.采用K-means算法对医学图像进行分割,将其分为3个簇。
C.根据B中分割结果,对分割出来的每一个簇采用都采用和声搜索算法(CHS)进行优化,找到每个簇的最优解;
本实施例得到的初始聚类数为三类,初始簇的优化结果分别为[13,82],[63,183],[120,146]。
D.根据最优解Xbest;以Xbest为中心,利用小生境技术产生一个小生境半径R,以Xbest为中心,R为半径形成的区域作为禁忌区。
E.重新再搜索空间中随机生成N个新的搜索点,如果新的搜索点在“禁忌区”中,则重新生成,转至步骤C继续执行。
F.不断搜索直到如所有簇中都找到了最优解或者迭代次数大于T,并根据搜索结果,映射至图像中完成图像分割,输出分割后的图像。
实施上述实验后,获取最终的聚类中心为[11,74],[34,138],[99,99],输出最终聚类结果,如图2b所示。
由此可见,本发明针对一般聚类算法只有单一的目标优化函数并且未考虑组成部分聚类效果的缺陷,一般聚类方法单目标函数、单阈值、未考虑组成部分聚类效果的问题,基于和声搜索算法,将
K-means算法初始分割后的各个簇和图像整体进行优化,在各个区域内实现和声搜索,并根据小生境技术合理设置“禁忌区”,实现更有效的最优搜索。通过上述实施例证明切实可行。
Claims (3)
- 一种医学图像分割方法,在计算机读入原始图像后,进行如下基本步骤:步骤A.设置参数:搜索规模N,聚类个数K,每个簇中优化迭代次数T,和声搜索算法参数记忆取值矩阵HMCR,微调概率PAR,音调微调带宽bw,最大迭代次数Tmax;步骤B.采用K-means算法对原始图像进行聚类,将其分为K个簇,完成对原始图像的初始分割;步骤C.根据步骤B中K-means算法的初始分割结果,对分割出来的每一个簇采用和声搜索算法进行优化;步骤D.如果在某一个簇中找到最优解,则停止在这一个簇的搜索,记录其最优解Xbest;以最优解Xbest为中心,利用小生境技术产生一个小生境半径R,以Xbest为中心,R为半径形成的区域作为禁忌区;步骤E.未满足终止条件时,重新再搜索空间中随机生成N个新的搜索点,如果新的搜索点在“禁忌区”中,则重新生成,转至步骤C继续执行;步骤F.如果在所有簇中都找到了最优解或者迭代次数大于T,则停止当前搜索,并根据搜索结果,映射至图像中完成图像分割,输出分割后的图像。
- 根据权利要求1所述的医学图像分割方法,其特征在于:所述步骤C和步骤D之间还具有以下步骤:对于分割出来的每一个簇,都采用和声搜索算法:首先产生数量等于和声规模的HMS的初始解,并放入和声记忆库中;然后,在和声记忆库内随机搜索新解,产生新接的概率由记忆取值矩阵HMCR决定;再以微调概率PAR对取自和声记忆库内的新解进行局部扰动;最后,判断新解目标函数值是否优于和声记忆库内的最差解,若是,则更新和声库,并不断迭代,直至达到预定迭代次数Tmax为止;具体算法如下:随机生成HMS个和声x1,x2,...,xHMS放入和声记忆库,和声记忆库如下:生成新的和声x’=(x’1,x’2,...x’N),新和声的每一个音调x’i(i=1,2,...N)通过以下三种机理产生:(1)学习和声记忆库、(2)音调微调、(3)随机选择音调;变量的生成方式如下:其中rand表示(0,1)上的均匀分布随机数;其次,如果新的和声x’i来自和声记忆库,要对其进行音调微调,具体操作如下:然后,对上述和声新解进行评估,重复上述过程,直至迭代次数达到Tmax。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610986747.8A CN106570873B (zh) | 2016-11-08 | 2016-11-08 | 一种医学图像分割方法 |
| CN201610986747.8 | 2016-11-08 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2018086433A1 true WO2018086433A1 (zh) | 2018-05-17 |
Family
ID=58540896
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2017/105431 Ceased WO2018086433A1 (zh) | 2016-11-08 | 2017-10-10 | 一种医学图像分割方法 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN106570873B (zh) |
| WO (1) | WO2018086433A1 (zh) |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872330A (zh) * | 2019-01-25 | 2019-06-11 | 安徽理工大学 | 一种改进狮群优化的二维Otsu快速图像分割方法 |
| CN109886979A (zh) * | 2019-02-27 | 2019-06-14 | 安徽理工大学 | 基于改进布谷鸟优化法的图像多阈值分割方法 |
| CN110097566A (zh) * | 2019-04-30 | 2019-08-06 | 合肥工业大学 | 一种图像阈值分割算法 |
| CN110147610A (zh) * | 2019-05-20 | 2019-08-20 | 杭州电子科技大学 | 一种用于河流突发水污染事件的溯源方法 |
| CN110956641A (zh) * | 2019-11-20 | 2020-04-03 | 南京拓控信息科技股份有限公司 | 一种基于化学反应优化的列车车轮踏面图像分割方法 |
| CN111563549A (zh) * | 2020-04-30 | 2020-08-21 | 广东工业大学 | 基于多任务进化算法的医学图像聚类方法 |
| CN111612782A (zh) * | 2020-05-28 | 2020-09-01 | 上海应用技术大学 | 一种颅脑医学图像配准方法 |
| CN111681245A (zh) * | 2020-06-17 | 2020-09-18 | 中原工学院 | 一种基于树结构的自适应权重k-means算法分割遥感图像的方法 |
| CN111724389A (zh) * | 2020-04-30 | 2020-09-29 | 北京天智航医疗科技股份有限公司 | 髋关节ct图像分割方法、装置、存储介质和计算机设备 |
| CN112287247A (zh) * | 2019-07-12 | 2021-01-29 | 南京邮电大学 | 基于Meanshift和K-means聚类的社交网络用户位置特征提取方法和装置 |
| CN112612011A (zh) * | 2020-12-24 | 2021-04-06 | 西安电子科技大学 | 基于kd-树空间搜索的dbscan生命探测雷达点迹凝聚方法和系统 |
| CN113011512A (zh) * | 2021-03-29 | 2021-06-22 | 长沙理工大学 | 基于rbf神经网络模型的交通生成预测方法及系统 |
| CN114782726A (zh) * | 2022-03-18 | 2022-07-22 | 桂林理工大学 | 一种基于蝙蝠算法的图像聚类分割方法 |
| CN115859105A (zh) * | 2022-12-02 | 2023-03-28 | 苏州镁伽科技有限公司 | 一种数据更新方法和装置 |
| CN116309398A (zh) * | 2023-03-02 | 2023-06-23 | 湘潭大学 | 一种多通道特征融合学习的印制电路板小目标缺陷检测方法 |
| CN116484065A (zh) * | 2023-06-19 | 2023-07-25 | 曲阜师范大学 | 基于物联网的定位数据优化存储方法及系统 |
| CN116703946A (zh) * | 2023-06-16 | 2023-09-05 | 安徽信息工程学院 | 一种基于深度学习的图像分割方法和系统 |
| CN117011307A (zh) * | 2023-07-03 | 2023-11-07 | 合肥工业大学 | 多项交互的模糊聚类的有效性指标图像分割方法及系统 |
| CN119205832A (zh) * | 2024-11-26 | 2024-12-27 | 龙兴(杭州)航空电子有限公司 | 基于空间熵的和声搜索的航空机载导航图像目标识别方法 |
| CN119397316A (zh) * | 2024-09-20 | 2025-02-07 | 泰安利丰源机械有限公司 | 基于激光熔覆的液压支架表面处理方法及系统 |
| CN121010619A (zh) * | 2025-10-27 | 2025-11-25 | 南京信息工程大学 | 一种多级图像分割方法 |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106570873B (zh) * | 2016-11-08 | 2019-10-25 | 江苏大学 | 一种医学图像分割方法 |
| CN107403435B (zh) * | 2017-07-28 | 2020-06-26 | 江苏大学 | 一种基于rgb颜色空间的彩色图像边缘提取方法 |
| CN107977990B (zh) * | 2018-01-27 | 2021-05-04 | 江西理工大学 | 基于正弦启发式搜索的图像配准方法 |
| CN108830289B (zh) * | 2018-04-28 | 2020-09-08 | 河南师范大学 | 一种基于改进的模糊c均值聚类的图像聚类方法及装置 |
| CN110443363B (zh) * | 2018-05-04 | 2022-06-07 | 北京市商汤科技开发有限公司 | 图像特征学习方法及装置 |
| CN108923708A (zh) * | 2018-07-16 | 2018-11-30 | 哈尔滨电机厂有限责任公司 | 一种基于非光滑控制理论的励磁系统调节器计算方法 |
| CN110738214A (zh) * | 2019-09-26 | 2020-01-31 | 天津大学 | 模糊c均值聚类光干涉条纹图二值化方法 |
| CN110838123B (zh) * | 2019-11-06 | 2022-02-11 | 南京止善智能科技研究院有限公司 | 一种室内设计效果图像光照高亮区域的分割方法 |
| CN111340824B (zh) * | 2020-02-26 | 2022-07-12 | 青海民族大学 | 一种基于数据挖掘的图像特征分割方法 |
| CN112734697B (zh) * | 2020-12-24 | 2024-05-28 | 沈阳东软智能医疗科技研究院有限公司 | 图像处理方法、装置、存储介质和电子设备 |
| CN113570555B (zh) * | 2021-07-07 | 2024-02-09 | 温州大学 | 基于改进蚱蜢算法的多阈值医学图像的二维分割方法 |
| CN114419020B (zh) * | 2022-01-26 | 2022-10-18 | 深圳大学 | 医学图像分割方法、装置、计算机设备及存储介质 |
| CN115473513B (zh) * | 2022-09-13 | 2025-06-27 | 福州大学 | 基于和声搜索算法的改进粒子滤波定位与建图方法 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120130929A1 (en) * | 2010-11-24 | 2012-05-24 | International Business Machines Corporation | Controlling quarantining and biasing in cataclysms for optimization simulations |
| CN104809737A (zh) * | 2015-05-13 | 2015-07-29 | 江西理工大学 | 一种基于双策略和声搜索算法的柚子图像分割方法 |
| CN106023162A (zh) * | 2016-05-12 | 2016-10-12 | 江西理工大学 | 基于集成策略和声搜索算法的草莓图像分割方法 |
| CN106570873A (zh) * | 2016-11-08 | 2017-04-19 | 江苏大学 | 一种医学图像分割方法 |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104715490B (zh) * | 2015-04-09 | 2017-05-03 | 江西理工大学 | 一种基于自适应步长和声搜索算法的脐橙图像分割方法 |
| CN104881871A (zh) * | 2015-05-22 | 2015-09-02 | 湖南大学 | 基于改进多目标和声搜索算法的交通图像分割方法 |
-
2016
- 2016-11-08 CN CN201610986747.8A patent/CN106570873B/zh active Active
-
2017
- 2017-10-10 WO PCT/CN2017/105431 patent/WO2018086433A1/zh not_active Ceased
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120130929A1 (en) * | 2010-11-24 | 2012-05-24 | International Business Machines Corporation | Controlling quarantining and biasing in cataclysms for optimization simulations |
| CN104809737A (zh) * | 2015-05-13 | 2015-07-29 | 江西理工大学 | 一种基于双策略和声搜索算法的柚子图像分割方法 |
| CN106023162A (zh) * | 2016-05-12 | 2016-10-12 | 江西理工大学 | 基于集成策略和声搜索算法的草莓图像分割方法 |
| CN106570873A (zh) * | 2016-11-08 | 2017-04-19 | 江苏大学 | 一种医学图像分割方法 |
Non-Patent Citations (3)
| Title |
|---|
| CUI, ZHAOHUA ET AL.: "Improved Fuzzy C-Means Clustering Combined with The Global Best Harmony Search Algorithm for Image Segmentation", JOURNAL OF IMAGE AND GRAPHICS, vol. 18, no. 9, 30 September 2013 (2013-09-30) * |
| MANDAVA, R. ET AL.: "Osteosarcoma Segmentation in MRI using Dynamic Harmony Search Based Clustering", INTERNATIONAL CONFERENCE OF SOFT COMPUTING AND PATTERN RECOGNITION (SO-CPAR, 13 January 2011 (2011-01-13), XP031848518 * |
| YONG, LONGQUAN ET AL.: "Global Harmony Search Algorithm for Absolute Value Equation with 2n Solutions", JOURNAL OF CHINESE COMPUTER SYSTEMS, 31 August 2014 (2014-08-31), pages 1862 - 1863 * |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872330A (zh) * | 2019-01-25 | 2019-06-11 | 安徽理工大学 | 一种改进狮群优化的二维Otsu快速图像分割方法 |
| CN109872330B (zh) * | 2019-01-25 | 2022-10-14 | 安徽理工大学 | 一种改进狮群优化的二维Otsu快速图像分割方法 |
| CN109886979A (zh) * | 2019-02-27 | 2019-06-14 | 安徽理工大学 | 基于改进布谷鸟优化法的图像多阈值分割方法 |
| CN110097566A (zh) * | 2019-04-30 | 2019-08-06 | 合肥工业大学 | 一种图像阈值分割算法 |
| CN110097566B (zh) * | 2019-04-30 | 2022-10-25 | 合肥工业大学 | 一种图像阈值分割方法及装置 |
| CN110147610A (zh) * | 2019-05-20 | 2019-08-20 | 杭州电子科技大学 | 一种用于河流突发水污染事件的溯源方法 |
| CN110147610B (zh) * | 2019-05-20 | 2023-02-07 | 杭州电子科技大学 | 一种用于河流突发水污染事件的溯源方法 |
| CN112287247A (zh) * | 2019-07-12 | 2021-01-29 | 南京邮电大学 | 基于Meanshift和K-means聚类的社交网络用户位置特征提取方法和装置 |
| CN112287247B (zh) * | 2019-07-12 | 2022-11-11 | 南京邮电大学 | 基于Meanshift和K-means聚类的社交网络用户位置特征提取方法和装置 |
| CN110956641A (zh) * | 2019-11-20 | 2020-04-03 | 南京拓控信息科技股份有限公司 | 一种基于化学反应优化的列车车轮踏面图像分割方法 |
| CN111724389A (zh) * | 2020-04-30 | 2020-09-29 | 北京天智航医疗科技股份有限公司 | 髋关节ct图像分割方法、装置、存储介质和计算机设备 |
| CN111724389B (zh) * | 2020-04-30 | 2023-12-12 | 北京天智航医疗科技股份有限公司 | 髋关节ct图像分割方法、装置、存储介质和计算机设备 |
| CN111563549A (zh) * | 2020-04-30 | 2020-08-21 | 广东工业大学 | 基于多任务进化算法的医学图像聚类方法 |
| CN111612782B (zh) * | 2020-05-28 | 2023-04-07 | 上海应用技术大学 | 一种颅脑医学图像配准方法 |
| CN111612782A (zh) * | 2020-05-28 | 2020-09-01 | 上海应用技术大学 | 一种颅脑医学图像配准方法 |
| CN111681245A (zh) * | 2020-06-17 | 2020-09-18 | 中原工学院 | 一种基于树结构的自适应权重k-means算法分割遥感图像的方法 |
| CN111681245B (zh) * | 2020-06-17 | 2023-03-14 | 中原工学院 | 一种基于树结构的自适应权重k-means算法分割遥感图像的方法 |
| CN112612011A (zh) * | 2020-12-24 | 2021-04-06 | 西安电子科技大学 | 基于kd-树空间搜索的dbscan生命探测雷达点迹凝聚方法和系统 |
| CN112612011B (zh) * | 2020-12-24 | 2023-12-26 | 西安电子科技大学 | 基于kd-树空间搜索的dbscan生命探测雷达点迹凝聚方法和系统 |
| CN113011512A (zh) * | 2021-03-29 | 2021-06-22 | 长沙理工大学 | 基于rbf神经网络模型的交通生成预测方法及系统 |
| CN114782726A (zh) * | 2022-03-18 | 2022-07-22 | 桂林理工大学 | 一种基于蝙蝠算法的图像聚类分割方法 |
| CN115859105A (zh) * | 2022-12-02 | 2023-03-28 | 苏州镁伽科技有限公司 | 一种数据更新方法和装置 |
| CN116309398A (zh) * | 2023-03-02 | 2023-06-23 | 湘潭大学 | 一种多通道特征融合学习的印制电路板小目标缺陷检测方法 |
| CN116703946A (zh) * | 2023-06-16 | 2023-09-05 | 安徽信息工程学院 | 一种基于深度学习的图像分割方法和系统 |
| CN116484065A (zh) * | 2023-06-19 | 2023-07-25 | 曲阜师范大学 | 基于物联网的定位数据优化存储方法及系统 |
| CN116484065B (zh) * | 2023-06-19 | 2023-08-25 | 曲阜师范大学 | 基于物联网的定位数据优化存储方法及系统 |
| CN117011307A (zh) * | 2023-07-03 | 2023-11-07 | 合肥工业大学 | 多项交互的模糊聚类的有效性指标图像分割方法及系统 |
| CN117011307B (zh) * | 2023-07-03 | 2025-12-26 | 合肥工业大学 | 多项交互的模糊聚类的有效性指标图像分割方法及系统 |
| CN119397316A (zh) * | 2024-09-20 | 2025-02-07 | 泰安利丰源机械有限公司 | 基于激光熔覆的液压支架表面处理方法及系统 |
| CN119205832A (zh) * | 2024-11-26 | 2024-12-27 | 龙兴(杭州)航空电子有限公司 | 基于空间熵的和声搜索的航空机载导航图像目标识别方法 |
| CN121010619A (zh) * | 2025-10-27 | 2025-11-25 | 南京信息工程大学 | 一种多级图像分割方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106570873A (zh) | 2017-04-19 |
| CN106570873B (zh) | 2019-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106570873B (zh) | 一种医学图像分割方法 | |
| CN113693563B (zh) | 一种基于超图注意力网络的脑功能网络分类方法 | |
| Saha et al. | Brain image segmentation using semi-supervised clustering | |
| CN113408605A (zh) | 基于小样本学习的高光谱图像半监督分类方法 | |
| CN116959572B (zh) | 基于图卷积神经网络的乳腺癌亚型分类方法和系统 | |
| CN102194133A (zh) | 基于数据聚类的适应性图像sift特征匹配方法 | |
| CN105139430A (zh) | 一种基于图熵的医学图像聚类方法 | |
| CN114638301A (zh) | 基于密度相似性的密度峰值聚类算法 | |
| CN106919950A (zh) | 概率密度加权测地距离的脑部mr图像分割方法 | |
| CN110706804B (zh) | 一种混合专家系统在肺腺癌分类中给的应用方法 | |
| CN114463320B (zh) | 一种磁共振成像脑胶质瘤idh基因预测方法及系统 | |
| CN118115787A (zh) | 一种基于图神经网络的全切片病理图像分类方法 | |
| Yang et al. | Segmentation of MRI brain images with an improved harmony searching algorithm | |
| CN120544189B (zh) | 一种子宫内膜癌细胞检测方法及系统 | |
| CN114399501A (zh) | 基于深度学习卷积神经网络的自动分割前列腺全腺体方法 | |
| CN120339773A (zh) | 一种基于图像识别的乳腺癌转移状态分析系统 | |
| CN113971441A (zh) | 一种基于样本包络多层聚类的数据集平衡化学习方法 | |
| CN103336781B (zh) | 一种医学图像聚类方法 | |
| Zhao et al. | Faster mean-shift: Gpu-accelerated embedding-clustering for cell segmentation and tracking | |
| CN118133881A (zh) | 一种用于抑郁症诊断的自适应图神经网络构建方法及应用 | |
| Hofmann et al. | An optimization approach to unsupervised hierarchical texture segmentation | |
| Lv et al. | Imbalanced Data Over-Sampling Method Based on ISODATA Clustering | |
| Zhan et al. | Medical image clustering algorithm based on graph entropy | |
| He et al. | CF-CAM: Cluster filter class activation mapping for reliable gradient-based interpretability | |
| CN115017988A (zh) | 一种用于状态异常诊断的竞争聚类方法 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 17869545 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 17869545 Country of ref document: EP Kind code of ref document: A1 |





