Research On Superpixel Algorithm Based On Density Clustering

At present,most of the image processing is based on pixels,and the two dimensional matrix is used to represent an image without considering the spatial organization relationship between pixels,which makes the processing efficiency of the algorithm too low.Therefore,the concept of superpixel is proposed.The so called superpixel refers to the pixel block composed of adjacent pixels with similar texture,color,brightness and other characteristics.It uses the similarity degree of features between pixels to group pixels,which can obtain redundant information of images.Most of the traditional image processing methods take pixel as the basic operation unit,and represent the image by matrix,so the processing efficiency of the algorithm is low.However,superpixel compress hundreds of thousands of pixels into hundreds of superpixels based on region,which greatly reduces the complexity of subsequent image processing tasks.In view of the shortcomings of DBSCAN superpixel algorithm and the deficiency of traditional superpixel in seed selection,this paper improves them respectively.The main contents are as follows:(1)DBSCAN superpixel algorithm is better for boundary processing,but the shape of the superpixel is irregular,and the number of the superpixel can not be directly controlled.In view of its shortcomings,a spatial constrained DBSCAN clustering algorithm for superpixel segmentation is proposed.Firstly,seed points are evenly seeded on the image,and the seed points are taken as the clustering center,and then the spatial constrained density clustering is used to expand outward gradually until the whole image is covered,and then the seed points are updated and the above steps are iterated.The experimental results using BSDS500 dataset show that the superpixel segmentation algorithm based on spatial constraint density clustering can improve the compactness obviously after passing the evenly distributed seed points and spatial constraints.(2)In the traditional superpixel segmentation method,the selection of seed points is easily affected by noise.In view of its shortcomings,a new superpixel segmentation algorithm is proposed.By using the noisy multimedia database clustering algorithm DENCLUE as the seed point selection method,it can effectively prevent the interference of noise on the selection of seed points.To reduce the computational cost,we divide the image into K grids,use the DENCLUE algorithm to find the local density attractor x~*in each grid,use the density attractor x~*as seed points for clustering,and use the mixed gradient formula to cluster to improve the algorithm's fit to the original boundary of the image.The experimental results using BSDS500 dataset show that the proposed algorithm can produce superpixels with higher boundary recall and accuracy than the traditional ones.