【作者】 杨光宇；

【导师】 郑永果；

【作者基本信息】 山东科技大学 ， 计算机应用技术， 2019， 硕士

【摘要】 图像分割作为一项基础的数字图像处理技术,广泛应用于生产生活的各个领域。针对传统活动轮廓模型中初始轮廓线选择,噪声污染,复杂的纹理边界和图像灰度不均匀的问题进行了研究,充分利用多种统计信息,将混合活动模型和多相水平集结合起来,提出了改进的拉普拉斯自适应多相分割模型和基于局部和全局拟合的自适应多相分割模型。在改进的拉普拉斯自适应多相分割模型中,利用梯度倒数加权平滑和中值滤波结合对图像预处理,引入结构张量自动定义初始轮廓线。在全局项中,加入了改进的零交叉拉普拉斯拟合能量,提高对弱边界的定位能力,提出一种基于图像区域信息的自适应分割项,提升了对灰度不均匀图像区域的分割效果。在局部拟合项中,运用以局部灰度均值和方差为变量的高斯核函数进行分割。引入一个加权系数方程平衡局部项和全局项的权重进行计算。实验表明,改进的拉普拉斯自适应多相分割模型抗噪能力强,适用于分割灰度不均匀的图像。在基于局部和全局的自适应多相分割模型中,提出了多尺度信息增强和各向异性张量扩散滤波结合的方法去噪和保留纹理细节,并利用结合二维最大熵和改进遗传算法获得初始轮廓线。在全局项中,用多种局部特征变量重构拉普拉斯拟合能量函数,提出了一个分割深度可控的区域分割模型和深度系数,更好地分割拓扑结构复杂的目标。在局部项中,采用LGDF模型处理灰度不均匀区域。模型使用指数函数作为停止速度函数,并加入鲁棒性的曲线演化停止条件,提升了分割效率。实验表明基于局部和全局的鲁棒性多相分割模型分割速度快,对内部结构和纹理特征较为复杂的图像有较好的分割效果。更多还原

【Abstract】 As a basic digital image processing technology,image segmentation is widely used in various fields of production and life.For the problems existing in the traditional active contour model,such as initial contour selection,noise pollution,complex texture boundaries and image gray unevenness,the paper makes full use of various statistical information to combination mixed activity models and multi-phase levels.This paper proposes the improved Laplace adaptive multiphase segmentation model and the robust multiphase segmentation model based on local and global fitting.In the improved Laplace adaptive multiphase segmentation model,gradient pre-weighted smoothing and median filtering are combined to preprocess the image,and the structural tensor is introduced to automatically define the initial contour.In the global term,the improved zero-crossing Laplacian fitting energy is added to improve the positioning ability of the weak boundary.An adaptive segmentation fitting term is added to the global term to improve the segmentation effect of intensity inhomogeneous.In the local fitting term,the Gaussian kernel function with local gray mean and variance is used for segmentation.A weighting coeffcient equation is introduced to balance the weights of local and global terms.Experiments show that the LGLAF model has strong anti-noise ability and has a good effect on inhomogeneous images.In the robust multiphase segmentation model based on local and global fitting,a combination of multi-scale information enhancement and anisotropic tensor diffusion filtering is proposed to de-noise and preserve texture details,and the initial contour is obtained by combining two-dimensional maximum entropy and improved genetic algorithm.In the global term,the Laplace fitting energy function is reconstructed by using various local feature variables.A segmentation model with controlled depth and depth coefficient is proposed to better segment the complex topology.In the local term,the LGDF model is used to process the gray uneven region.The model uses the exponential function as the speed stopped function,and adds the robust curve evolution stopping condition to improve the segmentation efficiency.Experiments show that the HMDLG model has a better segmentation effect on gray-scale inhomogeneous images with complex internal structures and texture features.更多还原