Research On Image Enhancement And Object Segmentation Algorithms For Terahertz Passive Imaging

The current mainstream human body security inspection equipment is mostly metal detection doors,but they can only detect metal objects carried by the human body,and have insufficient detection capabilities for other non-metallic objects.The terahertz passive imaging technology adopts a passive working mode,passively receives the electromagnetic waves in the terahertz frequency band radiated by the human body and the hidden target itself for imaging,and has no radiation hazard to the human body.In addition,electromagnetic waves in the terahertz frequency band have high permeability to clothing,etc.,and can detect various dangerous goods hidden in human clothing.This technology has important application prospects and research value in human security inspection equipment in public places.The research of this thesis is based on actual scientific research projects.Because the original image collected by the terahertz passive imaging system is affected by noise and the system's imaging resolution is limited,the edges of the image of the concealed target imaging become blurred.Therefore,before the hidden target segmentation,it is necessary to perform image enhancement to remove the noise in the image and enhance the edge details of the image;and then use the enhanced image to accurately segment the hidden target.Based on the above requirements,the main work of this thesis is as follows:(1)The composition structure and working principle of the terahertz passive imaging system are introduced,and the characteristics of the original image,such as noise and edges,are analyzed to provide a theoretical basis for the subsequent research on image enhancement algorithms.(2)Analyze the basic principles of anisotropic diffusion and total variation image enhancement methods.Aiming at the problem of insufficient protection of edge details while removing noise,an anisotropy based on energy minimization is proposed.Diffusion improvement algorithm: The energy minimization principle of the total variation method is applied to the anisotropic diffusion method,and a linear penalty function is added to convert the original problem into an unconstrained optimization problem to achieve image enhancement.The simulation shows that the improved image enhancement algorithm can well protect the edge of the target while smoothing the image.(3)Analyzed the basic principle of the maximum between-class variance segmentation method.Aiming at the problem that this method cannot effectively segment the hidden target when processing the unimodal terahertz image,an improved target segmentation algorithm based on image block and gradient amplitude is proposed: The image is divided into blocks,and the histogram is improved by using the gradient magnitude of each sub-image after the block.Finally,the improved histogram is used to segment the maximum between-class variance method.The simulation shows that the improved threshold segmentation algorithm effectively solves the under-segmentation problem of the maximum between-class variance method.(4)Analyze the basic principles of the region growing segmentation method,and improve the three key factors of the method.In view of the low degree of automation in the selection of initial seed points,the central point of the target area segmented by the fuzzy clustering method is proposed as the initial seed point;in view of the problem that the growth threshold is not adaptable,it is proposed to use the gradient of the pixels to be grown The amplitude dynamically adjusts the growth threshold;for the improvement of the growth termination condition,the target edge after Canny edge detection is used as the growth termination condition.Simulation shows that the improved region growing algorithm can effectively improve the accuracy of segmentation.