体液细胞多信息多尺度融合分割技术研究

发布时间：2018-05-07 12:45

本文选题：图像分割 + 小波变换　；参考：《湘潭大学》2017年硕士论文

【摘要】：医学图像分割技术是医学图像处理的关键技术,也是进一步进行图像分析识别和各种医学图像应用的基础。在临床诊断、辅助治疗等方面图像分割技术显示出了越来越重要的临床价值。从医学图像中自动分割出感兴趣的目标是件困难而艰巨的任务,除了医学图像本身的复杂多样性之外,其医学图像在成像过程中还会存在一定的噪声,此外分割算法的结果也会受到部分场偏移效应、局部体效应、灰度不均匀性、伪影等因素的影响。传统的分割方法显然很难满足医学图像分割的需要,所以对医学图像分割方法进行深入的研究是非常必要的。小波变换在时域和频域上都具有良好的局部检测能力和多分辨率分析的特点,这是我们将小波变换应用于细胞图像分割的理论依据。传统的边缘检测算子一般都是利用目标边缘的灰度不连续这一特性,通过计算出梯度局部极值像素点,连接这些像素点就是目标边缘,但是很容易受到噪声信息的干扰。为此提出采用自适应阈值的改进的B样条边缘检测算法,为了获得精确的边缘信息和除去噪声的干扰,以二次B样条函数为小波函数,利用多孔算法,计算出局部模极大值点,再根据边缘与噪声的特征自动提出自适应阈值,实现了噪声与边缘的分离,强边缘与弱边缘的分离,利用多尺度匹配融合策略,最终得到了综合各个尺度精确的细胞图像边缘,并通过实验分析验证了算法的有效性。医学图像处理提取细胞中使用分水岭方法时,容易产生过分割现象且对噪声的干扰极为敏感,为了解决此缺点,提出一种基于小波变换和形态学分水岭的细胞图像分割新方法。改算法利用小波变换多分辨率分析对图像进行分解,选取合适的小波基和改进去噪阈值函数对图像进行小波去噪,然后对去噪后小波重构的细胞图像应用数学形态学距离变换、灰度重建等技术产生的区域标记进行分水岭变换,最终得到分割结果。实验结果表明,该算法能稳定、准确地提取细胞和实现粘连细胞的自动分割,同时具有很好的鲁棒性和普适性。
[Abstract]:Medical image segmentation is the key technology of medical image processing, and it is also the basis of image analysis and recognition and various medical image applications. Image segmentation technology has shown more and more important clinical value in clinical diagnosis, adjuvant therapy and so on. Automatic segmentation of objects of interest from medical images is a difficult and difficult task. In addition to the complexity and diversity of medical images, there is still some noise in the imaging process. In addition, the results of the segmentation algorithm are also affected by some factors, such as partial field migration effect, local volume effect, gray inhomogeneity, artifact and so on. The traditional segmentation method is obviously difficult to meet the needs of medical image segmentation, so it is very necessary to deeply study the medical image segmentation method. Wavelet transform has good local detection ability and multi-resolution analysis in both time and frequency domain, which is the theoretical basis for the application of wavelet transform to cell image segmentation. The traditional edge detection operator usually utilizes the discontinuity of gray level of the edge of the target. By calculating the gradient local extremum pixels, connecting these pixels is the edge of the target, but it is easy to be disturbed by the noise information. An improved B-spline edge detection algorithm based on adaptive threshold is proposed. In order to obtain accurate edge information and remove noise, the quadratic B-spline function is used as wavelet function, and the local modulus maximum point is calculated by using porous algorithm. Then according to the feature of edge and noise, the adaptive threshold is put forward automatically, the separation of noise and edge, the separation of strong edge and weak edge, and the multi-scale matching fusion strategy are used to obtain the cell image edge which synthesizes all scales accurately. The validity of the algorithm is verified by experimental analysis. In order to solve this problem, a new method of cell image segmentation based on wavelet transform and morphological watershed is proposed. The modified algorithm decomposes the image by wavelet transform multi-resolution analysis, selects the suitable wavelet basis and the improved denoising threshold function to carry on the wavelet de-noising to the image, and then applies the mathematical morphological distance transform to the cell image reconstructed by the de-noising wavelet. Finally, the segmentation results are obtained by watershed transformation of the regional markers generated by gray level reconstruction. Experimental results show that the proposed algorithm can extract cells accurately and achieve automatic segmentation of adherent cells, and has good robustness and universality.
【学位授予单位】：湘潭大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TP391.41

【参考文献】