The Research On The Background And False Alarm Suppression Through Graph Theory

Pixel-sized small targets are often small in size,so they often have similar characteristics of noise or background corners at the time of detection.As a result,they can not be completely differentiated at the time of small target routine detection.There are many false alarm points during the procedure of trajectory association,especially in the case of complex backgrounds.Reducing the false alarm rate,which can improve the efficiency and accuracy of small target detection and tracking,is very important.The topical subject of this paper is the background suppression and false alarm suppression in small target detection.In this paper,a background suppression method based on Fiedler Vector is proposed.At the same time,a whole procedure of false alarm suppression is proposed: a certain range of image blocks of all suspected target points are taken as samples to obtain the local characteristics and matching characteristics,the two characteristics combine as the two-dimensional feature vector,and then use the clustering method to automatically distinguish the target block and the background block,so as to achieve the purpose of false alarm point suppression.In this paper we use the metric of graph theory as the local feature measurement method,which classify the blocks on the basis of classification characteristic of the Fiedler Vector,Related experiment is done to discuss the influence of different graph structure.And then the background suppression algorithm based on the Fiedler Vector is proposed on the basis of previous content.The experiment proves that the algorithm can not only highlight the target point,but also suppress the background point;the matching characteristic measure between the frames is a method based on the cross correlation coefficient,and the matching judgment step is added before the matching,which is an improvement of the SUSAN corner detection algorithm,and the relevant experiments discuss about the most appropriate parameters and image block size.After composing the two-dimensional eigenvector and normalizing,we use the method of spectral clustering to automatically distinguish the type of image blocks.From the classification results,it is possible to suppress all the false alarm points,thus proving the feasibility and superiority of the whole procedure.The experiment also considers the risk of target omission,and select the appropriate weight coefficient to ensure that the target point would not be missed.