The Application Of Modified NLM And CS Algorithm In Image Processing

In recent years, some mathematical tools such as the partial differential equations(PDE)and optimization theory have become the basic tools in image processing. This paper mainly studys models and algorithms for their applications in image denoising and image segmentation. As for image denoising, we present a new denoising model based on NLM by introducing an index confidence. As for image segmentation, we combine the cuckoo search algorithm(CS) with K-means algorithm(K-means) and propose a new segmentation algorithm. The main ideas of the work are as follow:The nonlocal means(NLM) algorithm perform denoising by exploiting the self similarity of patterns inside an image and computing a weighted average of center pixels whose neighborhoods(patches) are similar. The method can reduce the noise level significantly and well preserve most of the image content, especially the smooth regions and textures.However, there exist over-smoothing effect in areas with low contrasts, and there exists residual noise in edges. The ROF model is another well-known denoising method which is effective in preserving edges. We propose a variational approach that can overcome the shortcomings of the NL-means by adaptively regularizing nonlocal methods with the total variation. The regularized NL-means algorithm combines these two methods and reduces their respective defects by minimizing an adaptive total variation with a nonlocaldata fidelity term. Experimental results show that our RNL model outforms NLM and ROF models in image denoising.The Cuckoo Search algorithm(CS) is a bionic algorithm. When applied to image segmentation, it suffers from some drawbacks such as the heavy computation burden and slow convergence rate. In addition, it is easy to stuck in a local minimum point. In order to solve these problems, we propose a multi-threshold image segmentation algorithm based on a modified cuckoo search algorithm. The proposed algorithm employs the Otsu method to construct the fitness function. We also combine the Cuckoo Search algorithm with K-means to increase the diversity of the population. The algorithm can determine the number and range of the thresholds adaptively. Experimental results show that the proposed algorithm outforms K-Means and Cuckoo Search(CS) in terms of segmentation thresholds and segmentation effect.