Research On Focal Plane Terahertz Image Stitching Technology

Terahertz focal plane imaging technology is an important branch of current terahertz imaging,enabling high-frame rate terahertz real-time imaging.Image quality is affected by many factors,such as the power stability,uniformity of the terahertz source,the shape and size of the spot,the noise of the back-end circuitry and so on.In addition to overcoming hardware barriers such as the resolution,accuracy and speed limitations of terahertz imaging systems,terahertz image processing techniques are also important.Image stitching is the process of combining multiple images with overlapping fields to obtain a high resolution image.It is widely used in remote sensing,medical imaging,computer vision and other fields.There are many existing image registration algorithms,which can be divided into region-based registration algorithms and feature-based registration algorithms.Compared to region-based registration algorithms,feature-based registration algorithms have the advantage of being more stable and faster for scene movement.Due to the characteristics of terahertz radiation itself,interference fringe noise and diffraction noise often appear in image acquisition,and the image contrast is low.If the existing image registration is directly applied to the terahertz image field,the effect is very low.In view of this situation,this paper studies the THz image stitching technology.The main results are as follows.1.In this paper,a focal plane terahertz real-time imaging system is established.In order to ensure the proportion of overlapping areas of the acquired images,a twodimensional translation stage is combined on the basis of the self-built imaging system.And for it to write a test system that can realize real-time imaging and save image data with the movement of two-dimensional translation stage,the whole system can realize scanning imaging and data acquisition of objects with the movement of two-dimensional translation stage.2.Aiming at the characteristics of laboratory THz images,an image preprocessing method is proposed.The image is processed by weight distribution method,Butterworth band rejection filter and high frequency nonlinear enhancement method,which can realize the targets of THz image denoising,highlighting the details and enhancing the image contrast.3.Using the single SURF algorithm for image feature point detection and matching processing,there are defects in feature point matching precision,high computational complexity and long time-consuming.Based on this conclusion,RANSAC algorithm is proposed.The feature point is further filtered to eliminate a large number of incorrect matching point pairs,which improves the matching accuracy and reduces the duration.4.Based on the image preprocessing,feature point detection and matching,the stitching of THz images is realized.The result of image stitching by weighted average method is better than that of the maximum method for image stitching.The overlapping part is smooth,the brightness is changed and the seam is not obvious,which meets the requirements of image stitching.