基于改进FCM和径向基函数插值的图像修复

图像破损区域的检测提取是图像修复过程中的关键预处理步骤,模糊C均值聚类算法(FCM)在聚类过程中易受到初始聚类中心影响并陷入局部最优.提出一种基于差分演化的改进模糊C均值聚类算法(DEFCM),该方法通过建立图像的灰度-梯度直方图获取聚类数目,作为差分演化算法(DE)问题的维数,结合改进的FCM自适应提取图像破损区域,在此基础上,利用径向基函数插值方法(RBF)对图像进行修复.经实验验证,该方法能解决FCM算法陷入局部最优的问题,能正确、稳定的提取灰度图像的多种破损区域,RBF通过对破损区域的插值得到缺失信息,实现图像的修复.

Digital image inpainting by DEFCM and RBF interpolation

Detecting damaged areas of an image to extract is an essential step of pretreatment in the image restoration process. Fuzzy C-Means clustering algorithm （FCM） in the process of cluster is susceptible to initial cluster centers into a local optimum. A method for digital image inpainting based on Different Evolution Fuzzy C-Means （DEFCM） and Radial Basis Functions （RBF） interpolation is proposed. Through the establishment of gray - gradient histogram, it obtains the number of clustering as dimension of Differential Evolution algorithm （DE）, combining with FCM adaption to extract damaged areas, and inpaint them by radial basis function interpolation. The experimental results show that the proposed algotithm can solve the problem of FCM algorithm falling into the local optimum, and can extract grayscale image from various damaged areas correctly and stably; the missing information can be restored correctly in the damaged areas by RBF interpolation algorithm to get the image inpainted.