基于模糊聚类的脑图像分割与识别研究

发布时间：2018-02-02 15:49

本文关键词： 模糊聚类算法 FLICM SCoW SVM 改进粒子群　出处：《西南科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着临床医学和计算机科学的相互渗透与发展,医学图像处理能避开主观因素的影响,尽早准确地诊断出肿瘤种类及恶性程度,有助于对患者制定有效的治疗方案,因此基于图像处理的脑图像分割与识别显得尤为重要。论文主要对模糊聚类分割与SVM识别及相关算法进行研究。针对模糊聚类算法对脑图像中的噪声、伪影等极为敏感的问题,提出了基于邻域信息的FLICM改进算法。论文在梯度下降法推导的FLICM算法基础上,通过引入像素的灰度相关性,融合像素的空间信息改进了目标函数。经实验测试表明,相比较其他的模糊聚类分割算法,该改进算法的分割效果更好,抗噪性能明显得到增强。为提高FLICM算法的分割效率,论文提出了结合SCoW的FLICM改进算法。首先对图像进行SCoW预处理并通过阈值法进行修正细分割,压缩预处理数据;然后提取各超像素块的均值特征作为聚类算法的输入;最后基于改进的FLICM算法完成图像分割。经实验测试表明,该改进算法的分割效果更好,运行效率平均提高了30%。由于SVM的性能很大程度上取决于模型参数的选择,同时粒子群算法存在过早收敛以致陷入局部最优等问题,致使基于粒子群优化的SVM的分类结果不够好,因此论文研究通过引入变异机制和动态参数设置的改进粒子群优化SVM训练模型,根据脑肿瘤的形状和边缘特征,实现二次SVM的脑肿瘤识别。经实验测试表明,本文算法的收敛速度更快、分类精度更高。
[Abstract]:With the mutual penetration and development of clinical medicine and computer science, medical image processing can avoid the influence of subjective factors and accurately diagnose tumor types and malignancy as soon as possible. It is helpful to establish an effective treatment plan for patients. Therefore, the segmentation and recognition of brain image based on image processing is particularly important. This paper mainly studies the fuzzy clustering segmentation, SVM recognition and related algorithms, aiming at the noise in the brain image based on the fuzzy clustering algorithm. Artifact and other extremely sensitive issues, this paper proposes an improved FLICM algorithm based on neighborhood information. Based on the FLICM algorithm derived by gradient descent method, the gray correlation of pixels is introduced. The spatial information of the fused pixels improves the objective function. The experimental results show that compared with other fuzzy clustering segmentation algorithm, the improved algorithm has better segmentation effect. The anti-noise performance is obviously enhanced. In order to improve the segmentation efficiency of FLICM algorithm. In this paper, an improved FLICM algorithm combined with SCoW is proposed. Firstly, the image is preprocessed by SCoW and modified by threshold method to compress the preprocessed data. Then the mean feature of each super-pixel block is extracted as the input of the clustering algorithm. Finally, the image segmentation is completed based on the improved FLICM algorithm. The experimental results show that the improved algorithm has better segmentation effect. Because the performance of SVM depends largely on the choice of model parameters, particle swarm optimization algorithm has some problems such as premature convergence and local optimization. As a result, the classification results of SVM based on PSO are not good enough, so this paper studies the improved PSO SVM training model by introducing mutation mechanism and dynamic parameter setting. According to the shape and edge features of brain tumors, the recognition of brain tumors by quadratic SVM is realized. The experimental results show that the proposed algorithm has faster convergence speed and higher classification accuracy.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】