局部二值拟合模型利用图像的局部邻域信息拟合能量函数,局部二值拟合模型相当于对活动轮廓内外进行局部均值滤波,因此该模型对高斯噪声有一定的抗噪性,但是该模型对椒盐噪声污染的图像不能取得令人满意的分割结果.为了提高活动轮廓模型对椒盐噪声的鲁棒性,结合局部二值拟合模型中的局部拟合项,提出一种能消除椒盐噪声影响的新的局部拟合项.提出的拟合项的能量函数极小值是局部区域的中值,新的拟合项相当于对活动轮廓内外进行局部中值滤波,中值滤波对椒盐噪声不敏感.原模型中边缘停止函数是基于图像梯度信息,难以区分图像不同区域间模糊的弱边缘和灰度渐进图像目标,并且容易受到各类噪声的影响,抗噪声能力弱.提出一种新的边缘检测算子,重新定义边缘停止函数,进一步提高模型的抗噪性,降低高斯噪声和椒盐噪声对分割结果的影响.为了保持活动轮廓在演化过程中的稳定性,在曲线演化迭代过程中必须周期地初始化以使水平集函数重新变成带符号的距离函数,但重新初始化的计算量大.引入一个惩罚能量,解决水平集函数在演化过程中的重新初始化难题.对不同噪声污染的图像进行试验的结果表明,提出的模型可以取得较好的图像分割结果,比CV模型、LBF模型和LIF模型更具有优势.%The local binary fitting(LBF) model fits energy function with local neighborhood information of the image.Since it is equivalent to the local average filtering inside and outside the active contour,LBF has some noise immunity for Gaussian noise.However,it cannot obtain satisfactory segmentation results for salt and pepper noise.In this paper,a novel model integrating a local fitting term into the fitting term of LBF was proposed to enhance the robustness for the salt and pepper noise.The new fitting term was equivalent to the local median filtering inside and outside the active contour,which is insensitive to the salt and pepper noise.As the edge stop function of the original model is based on image gradient information,which is difficult to distinguish the weak edge of the image and is sensitive to all kinds of noise,a new edge detection operator was proposed to redefine the edge stop function to further improve the noise immunity of the model.In order to keep the stability of the active contours,the original model must be periodically initialized to ensure the level set function be a signed distance function and the re-initialization performs a large amount of computation.In our model,a punish energy term was introduced to eliminate the re-initialization.Compared with CV model,LBF model and LIF mode,experiments on images corrupted by different noise demonstrated that the proposed model was effective and more superior for noise image segmentation.
展开▼
