Research On Superpixel Segmentation Of Multispectral Images

With the development of the technology,the resolution of imaging equipment increase,which makes many pixel-based traditional segmentation algorithms become more and more inefficient,so the concept of superpixel segmentation came into being.However,it is difficult for the current superpixel partitioning algorithm to meet the basic requirements of fast speed,high edge fit,high uniform superpixel size,and strong anti-noise performance.In view of this,in this paper,the super-pixel segmentation is the main research direction,we propose a hierarchical superpixel segmentation model based on histogram 1-dimensional differential distance and improved algorithms based on multispectral images.The innovation of the superpixel segmentation algorithm is embodied in the following.First,a new histogram similarity criterion based on structure is proposed.This criterion reflects the structure information of the histogram,reflects the weak gap between histograms,and has the advantages of simplicity and high accuracy.Moreover,a hierarchical superpixel segmentation model based on pyramid is presented.Regional block segmentation accuracy is priority guaranteed due to the introduction of the pyramid theory.Then,the weakened details are restored when the image is restored to the original resolution.This idea improves the anti-noise capability of the algorithm.And the tightness constraint term is introduced,it encourages the superpixel with similar size.The experimental results show that,the perfect combination of accuracy,efficiency and tightness can be achieved through this algorithm,which has better anti-noise performance.As a more complete reflection of the target information can be reflected in the multispectral images,the classical SLIC(simple linear iterative cluster)algorithm and the above algorithm are respectively extended to the improved superpixel segmentation algorithm for multi-spectral images.Proved by experiment,the use of these two algorithms results in superior segmentation results,allowing more efficient use of multispectral informationIn summary,the above algorithm can effectively take into account the time complexity,accuracy,practicality and anti-noise performance,with high practical value.