论文总字数:31680字
摘 要
本文提出了一种基于脉冲耦合神经网络(Pulse Coupled Neural Network,PCNN)模型的新型图像分割算法。为了提高算法的自适应能力,对简化的PCNN模型采用粒子群算法(Particle Swarm Optimization,PSO)进行自动寻优,避免了人工确定参数带来的低效率问题。同时设计了一种新的分割框架,通过对图像进行平移不变剪切变换(Shift Invariant Shearlet Transform,SIST),得到其低频系数和高频系数,对低频系数进行PCNN预分割后,再对SIST逆变换(Inverse Shift Invariant Shearlet Transform,NSIST)重建的图像进行PCNN二次分割以提高最终的分割效果。
为了扩大算法应用范围,本文将算法应用到彩色图像的分割上。将RGB彩色图像转化到HSI彩色空间,对HSI各分量进行分割后,再合成最终的彩色分割结果。通过Matlab编程和仿真,以伯克利分割数据集(Berkeley Segmentation Dataset,BSDS)作为测试对象,将最大熵阈值法、OSTU分割算法、人工设定PCNN参数方法作为对比项,并设计科学的量化评估方案,从定性和定量两个方面证明,本算法具有较好的分割性能和较强的泛化能力。
本文最后提出了将算法移植到基于ARM的嵌入式平台上的方案,完善了算法的实际应用能力。
关键词:图像分割,脉冲耦合神经网络,粒子群算法,平移不变剪切变换,HSI彩色空间,ARM嵌入式平台
Abstract
A new image segmentation algorithm based on pulse coupled neural network was proposed in this paper. In order to improve the adaptive ability of the algorithm, PSO algorithm was applied to optimize parameters of the simplified PCNN model automatically, which avoid low efficiency of manual selection of parameters. At the same time, we designed a new scheme for image segmentation. Firstly, we separated the image into low and high sub-band frequencies using SIST. Secondly, the modified PCNN model mentioned above was used for the preliminary classification of the low coefficients. Eventually, we augmented the final segmentation result by classifying the rebuilt image through NSIST.
We applied this algorithm to segment RGB images aiming at expand its application scope. The RGB image was firstly transformed into HSI color space. Each component was processed and then mixed into final segmentation results. We chose Berkeley Segmentation Dataset as the test object and used Matlab for programming and simulation. The results were compared with those using maximum entropy threshold algorithm, OSTU algorithm and manual selection of PCNN parameters. Quantitative measures were used to evaluate the results. Our algorithm was proved better performed and generalization from both qualitative and quantitative aspects.
At the end of this paper, the algorithm was applied to the embedded platform based on ARM so that its practicality could be improved.
KEY WORDS: Image Segmentation, Pulse Coupled Neural Network, Particle Swarm Optimization, Shift Invariant Shearlet Transform, HSI Color Space, ARM Embedded Platform
目 录
摘 要 I
Abstract II
第一章 绪论 1
1.1引言 1
1.2国内外图像分割研究现状 1
1.2.1图像分割算法 1
1.2.2图像分割评价 3
1.2.3研究现状小结 4
1.3本文的研究目的和内容 4
第二章 预处理方案 6
2.1HSI彩色空间 6
2.2各项异性扩散 7
2.3平移不变剪切变换 10
2.4本章小结 12
第三章 粒子群算法 13
3.1粒子群算法背景 13
3.2粒子群算法概述 13
3.2.1算法描述 13
3.2.2参数分析 14
3.2.3算法流程 14
3.2.4算法改进 15
3.3测试与分析 15
3.4本章小结 16
第四章 PCNN模型及其应用 18
4.1PCNN模型概述 18
4.1.1基本模型 18
4.1.2简化模型 19
4.1.3工作原理 19
4.2PCNN模型应用 20
4.2.1基于PCNN模型的图像增强 20
4.2.2基于PCNN模型的图像分割 21
4.3测试与分析 21
4.4本章小结 23
第五章 基于PSO优化的PCNN分割算法 24
5.1PSO优化PCNN参数 24
5.1.1适应度函数设计 24
5.1.2参数优化具体步骤 24
5.2分阶段分割 25
5.2.1低频分量预分割 25
5.2.2重建图像二次分割 25
5.3测试与分析 25
5.4本章小结 29
结论 31
参考文献 33
致 谢 35
第一章 绪论
1.1引言
剩余内容已隐藏,请支付后下载全文,论文总字数:31680字
该课题毕业论文、开题报告、外文翻译、程序设计、图纸设计等资料可联系客服协助查找;