Research On Image Segmentation In Terahertz Digital Holography Reconstruction Images

Terahertz inline digital holography technology is now attracting more and more attention.It has wide application prospects in high resolution imaging and biological target imaging.Images of continuous-wave terahertz inline digital holography reconstruction are always of small size,large noise,which make the target segmentation become difficult.Therefore,it is necessary to search on the image segmentation for it.The second chapter of this thesis improves the algorithm of curve fitting threshold method and the ostu method by adding image denoising pre-processing steps,to smooth the noise of gear Thz image and gasket Thz image.In order to verify the applicability of the algorithm to those similar images,simulation experiments are also conducted for the biological leaf image and the match image with the similar target size.These two images have the finest structure of one single pixel.Similar levels of noise is added artificially to perform segmentation studies.Since we know the threshold segmentation method can not consider the pixel position information,then it is difficult to avoid the loss of the single-pixel details of the gear image.So the position and gray level information of the image is converted into data,and K-means clustering and mean shift algorithm are used.The terahertz image is segmented with the combination of the clustering algorithm,ostu method and morphological method,to explore the best parameters,and the leaf image and the match image are used for simulation research.To deal with the target structure fracture in the process of gear image segmentation,in the fourth chapter the level set algorithm with boundary contour length constraint is introduced.Morphology methods are combined into C-V model and local statistics level-set method,to explore the best parameters and improve the quality of segmentation,and the two simulation image are used for verification.