数字图像处理中二维经验模式分解关键问题研究
发布时间：2019-03-09 13:01
【摘要】：二维经验模式分解是黄变换经验模式分解方法的二维拓展,它具有良好的自适应性处理能力,为满足图像处理中非平稳信号特征分析和矿山企业的实际需要提供了新的技术措施。然而,实际应用发现,它存在插值优化、端部效应、停止条件以及图像融合时分解图像的不一致而较难融合等问题。为解决这些问题,本文开展了深入系统的研究,涉及到如下四个方面的工作。(1)提出了一种基于粒子群分形的插值算法。利用分形布朗函数理论对图像分析,得到图像特征量,再进行自适应插值计算。该算法不仅可以提高二维经验模式分解的插值精度和效率,实现图像快速分解,还为其它几个问题的解决奠定了良好的基础。(2)提出了一种基于自适应支持向量机延拓和镜像闭合技术相结合的端部效应处理算法。通过自适应支持向量机延拓图像,得到图像边缘极值点,对延拓图像进行镜像闭合处理,实施图像分解。相关图像的二维经验模式分解结果表明,这一算法的应用可以减弱甚至消除二维经验模式分解过程中的端部效应,从而较好地保证图像边缘信息及细节信息的有效提取。(3)建立了一种基于筛分曲面在零值平面上投影位置不重合极值点数目及其变化速度的停止条件。先对图像在二维经验模式分解过程中得到的包络曲面进行分析,再对分解过程中极值点的演化规律进行跟踪,最后通过不断筛分得到曲面空间位置信息判断筛分停止条件。它使图像进行二维经验模式分解得到的二维固有模态函数更趋于图像本身特性,能有效消除图像的过度分解和欠分解现象。(4)建立了基于二维经验模式分解-SIFT图像特征提取新算法与基于二维经验模式分解-极值点协调图像融合新算法。前者利用二维经验模式分解得到多个二维固有模态函数,提取其特征后累加合成。该算法可以避免不同模态信息间的相互干扰,有利于更快更准地获得图像各类特征信息。后者基于多幅图像二维经验模式分解得到极大值点集和极小值点集,进行多幅图像自适应基函数的协同操作,使其得到共同自适应基函数,再对图像分解得到的二维固有模态函数分别融合并合成累加得到融合图像,由此构造了一种具有自适应特性的协调操作图像融合算法。研究结果表明该算法在获得更多原始图像细节、边缘信息以及提高融合图像质量方面具有良好的效果。大量图像处理实例证实了所提出技术方法具有良好的应用效果。综上所述,本文的研究工作是对二维经验模式分解在图像处理中应用的发展与完善,也为图像处理中非平稳信号特征分析提供了有益的基础性技术措施。同时,它也是自适应数字图像处理方法在矿山安全监控的应用拓展。
[Abstract]:The two-dimensional empirical mode decomposition is a two-dimensional extension of the yellow transform empirical mode decomposition method, and it has good adaptive processing ability. New technical measures are provided to satisfy the characteristics analysis of non-stationary signals in image processing and the practical needs of mining enterprises. However, it is found that it has many problems, such as interpolation optimization, end effect, stop condition and the inconsistency of decomposing image in image fusion, which makes it difficult to fuse. In order to solve these problems, this paper has carried out in-depth and systematic research, involving the following four aspects of work. (1) an interpolation algorithm based on particle swarm fractal is proposed. The fractal Brownian function theory is used to analyze the image, get the feature quantity of the image, and then carry on the adaptive interpolation calculation. This algorithm can not only improve the interpolation accuracy and efficiency of 2-D empirical mode decomposition, but also realize the fast image decomposition. It also lays a good foundation for solving other problems. (2) an end effect processing algorithm based on the combination of adaptive support vector machine extension and mirror image closure technique is proposed. By using adaptive support vector machine (SVM) to extend the image, the extreme points of the edge of the image are obtained, and the image closure is processed and the image decomposition is carried out. The results of two-dimensional empirical mode decomposition of correlated images show that the application of this algorithm can weaken or even eliminate the end effect in the process of two-dimensional empirical mode decomposition. Thus, the edge information and detail information of the image can be extracted effectively. (3) A stopping condition based on the number of non-coincident extreme points in the projection position of the screen surface on the zero-valued plane and its changing speed is established. Firstly, the enveloping surface of the image in the process of two-dimensional empirical mode decomposition is analyzed, and then the evolution rule of the extreme point in the decomposition process is tracked. Finally, the space position information of the surface is continuously screened to judge the screening stop condition. It makes the two-dimensional intrinsic mode function obtained from the two-dimensional empirical mode decomposition of the image tend to the characteristics of the image itself. It can effectively eliminate the phenomenon of over-decomposition and under-decomposition. (4) A new image fusion algorithm based on two-dimensional empirical mode decomposition-SIFT image feature extraction and two-dimensional empirical mode decomposition-extreme point coordination image fusion is proposed. The former uses two-dimensional empirical mode decomposition to obtain multiple two-dimensional intrinsic modal functions, and then extracts their features and then accumulates them. This algorithm can avoid the mutual interference between different modal information, and it is helpful to get all kinds of image feature information more quickly and accurately. Based on the two-dimensional empirical mode decomposition of multiple images, the maximum point set and the minimum point set are obtained, and the cooperative operation of multiple image adaptive basis functions is carried out, so that the common adaptive basis functions can be obtained from the two-dimensional empirical mode decomposition of multiple images. Then the two-dimensional intrinsic modal functions obtained by image decomposition are fused and synthesized to get the fused image, and then an adaptive image fusion algorithm with coordinated operation is proposed. The results show that the algorithm has a good effect on obtaining more details of original image, edge information and improving the quality of fusion image. A large number of image processing examples show that the proposed technique has a good application effect. To sum up, the research work of this paper is to develop and perfect the application of two-dimensional empirical mode decomposition in image processing, and to provide useful basic technical measures for the analysis of non-stationary signal features in image processing. At the same time, it is also an extension of the application of adaptive digital image processing method in mine safety monitoring.
【学位授予单位】：北京科技大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41
本文编号：2437477
[Abstract]:The two-dimensional empirical mode decomposition is a two-dimensional extension of the yellow transform empirical mode decomposition method, and it has good adaptive processing ability. New technical measures are provided to satisfy the characteristics analysis of non-stationary signals in image processing and the practical needs of mining enterprises. However, it is found that it has many problems, such as interpolation optimization, end effect, stop condition and the inconsistency of decomposing image in image fusion, which makes it difficult to fuse. In order to solve these problems, this paper has carried out in-depth and systematic research, involving the following four aspects of work. (1) an interpolation algorithm based on particle swarm fractal is proposed. The fractal Brownian function theory is used to analyze the image, get the feature quantity of the image, and then carry on the adaptive interpolation calculation. This algorithm can not only improve the interpolation accuracy and efficiency of 2-D empirical mode decomposition, but also realize the fast image decomposition. It also lays a good foundation for solving other problems. (2) an end effect processing algorithm based on the combination of adaptive support vector machine extension and mirror image closure technique is proposed. By using adaptive support vector machine (SVM) to extend the image, the extreme points of the edge of the image are obtained, and the image closure is processed and the image decomposition is carried out. The results of two-dimensional empirical mode decomposition of correlated images show that the application of this algorithm can weaken or even eliminate the end effect in the process of two-dimensional empirical mode decomposition. Thus, the edge information and detail information of the image can be extracted effectively. (3) A stopping condition based on the number of non-coincident extreme points in the projection position of the screen surface on the zero-valued plane and its changing speed is established. Firstly, the enveloping surface of the image in the process of two-dimensional empirical mode decomposition is analyzed, and then the evolution rule of the extreme point in the decomposition process is tracked. Finally, the space position information of the surface is continuously screened to judge the screening stop condition. It makes the two-dimensional intrinsic mode function obtained from the two-dimensional empirical mode decomposition of the image tend to the characteristics of the image itself. It can effectively eliminate the phenomenon of over-decomposition and under-decomposition. (4) A new image fusion algorithm based on two-dimensional empirical mode decomposition-SIFT image feature extraction and two-dimensional empirical mode decomposition-extreme point coordination image fusion is proposed. The former uses two-dimensional empirical mode decomposition to obtain multiple two-dimensional intrinsic modal functions, and then extracts their features and then accumulates them. This algorithm can avoid the mutual interference between different modal information, and it is helpful to get all kinds of image feature information more quickly and accurately. Based on the two-dimensional empirical mode decomposition of multiple images, the maximum point set and the minimum point set are obtained, and the cooperative operation of multiple image adaptive basis functions is carried out, so that the common adaptive basis functions can be obtained from the two-dimensional empirical mode decomposition of multiple images. Then the two-dimensional intrinsic modal functions obtained by image decomposition are fused and synthesized to get the fused image, and then an adaptive image fusion algorithm with coordinated operation is proposed. The results show that the algorithm has a good effect on obtaining more details of original image, edge information and improving the quality of fusion image. A large number of image processing examples show that the proposed technique has a good application effect. To sum up, the research work of this paper is to develop and perfect the application of two-dimensional empirical mode decomposition in image processing, and to provide useful basic technical measures for the analysis of non-stationary signal features in image processing. At the same time, it is also an extension of the application of adaptive digital image processing method in mine safety monitoring.
【学位授予单位】：北京科技大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41
【参考文献】
相关期刊论文 前2条
1 汤雅妃;张云勇;张尼;;基于指纹识别的云安全认证技术[J];电信科学;2015年08期
2 邓蕾;胡小林;李锋;汤宝平;;基于支持向量机的BS-EMD端点效应消除方法[J];振动.测试与诊断;2011年03期
，本文编号：2437477
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2437477.html
