Research Of Normalized Cuts Algorithm Based On SLIC Superpixels

In the field of image segmentation,the normalized cuts image segmentation algorithm based on graph theory is an unsupervised algorithm which considers comprehensive of the global and local image information,obtaining the global optimal.Normalized cuts gets all the attention because of its good versatility and segmentation results.But as the segmentation image gets bigger,the execution time of the algorithm increases sharply,it is hard for real-time applications.Simple linear iterative clustering method has better comprehensive performance,which can generate compact,uniform superpixels.It remains in line with expectations in running time and the object profile.But the algorithm doesn't fully consider the similarity between the isolated point and the nearest neighbor superpixels in the process of clustering iterations,which results the superpixels have no better image boundary.According to the above defects,the paper designed a kind of normalized Cuts algorithm based on improved SLIC superpixels,not only reduced the running time,but also obtained the good segmentation effect.In this paper,the main work and contributions are as follows:(1)Research and analysis of the theory and progress of the normalized cuts segmentation method based on graph theory and the superpixels preprocessing segmentation algorithm,the feasibility of the combination of the normalized cuts and SLIC superpixels algorithm was explored.(2)Based on the normalized cuts method long-running time,we put forward a kind of pre-segmentation method based on the SLIC superpixels,to decrease the number of pixels points by the image preprocessing,and then using the normalized cuts to reprocess,which greatly reduced the running time of the normalized cuts segmentation algorithm.(3)In view of the SLIC pre-segmentation will produce tiny area,make the boundary segmentation inaccurately,the paper introduced the bilateral smoothing filter to cover edge image,and make smoothing pretreatment.Proposed a new merge method of small areas,Solved the deficiency of the SLIC presegmentation algorithm effectively.(4)when using normalized cuts segmentation algorithm,the paper improved the weight of the original function,which reduced the running time and didn't influence the segmentation circumstances.At last,Using the part of the images in The Berkeley Segmentation Dataset,making the quantitative evaluation and comparative analysis between the algorithm proposed in this paper with the original algorithm.The experimental results demonstrate that the algorithm presented is efficiently and effectively.