Research Of The Target Detection Algorithm Based On Digital Image

Target detection is an important research topic in image processing, which decides thefinal result and quality of image analysis and image understanding. Due to its importance,many scholars at home or abroad all begin for its research, and a large number of sorts ofdetection algorithm is proposed. The existing algorithms are mostly aimed at the specificresearch object, and the general theory there has yet to put forward, so people are stillconstantly exploring the new detection algorithm and theory.First of all, the static image segmentation algorithm is studied. According to thecharacteristics of spectral clustering algorithm needs large memory space and computation, arandom sampling probability factors spectral clustering algorithm is put forward. In thealgorithm, a similarity function is constructed according to the grey value and space distanceof image, therefore got a similarity matrix based on the grey value, and then converted theproblem of image segmentation into the minimum cut problem in graph theory. Finallysegmentation is finished by spectral clustering algorithm. In the concrete implementation, inorder to reduce the influence of background noise to the segmentation result, the pretreatmentof image filtering can be used first. The results show that the algorithm has good stability, andhas an improved segmentation results relative to the existing algorithm.Secondly, in view of the moving targets detection problem under complicatedbackground from static single camera, Gaussian mixture model, the inter-frame differencemethod, the background difference method, the statistical average method background modeland mixed Gaussian background model is studied. Aimed at the advantages and disadvantagesof these models, a median model is put forward based on those theory, and which given aestimation of the background value of each location. And then, moving object in each frame isextracted with background difference method. Due to the calculated background value is forthe estimation of the real background, and the background in each frame is the sample of thereal, there exist part of the noise in the background of the binary image, which obtained by thebackground difference method. Therefore, some morphology processing method such asextracting connected components is used here for denoising. Moving target area in each frameis obtained, and then for the further moving targets detection. The experimental results showthe algorithm with a small computational complexity and good detection effect.