Study On Multi-dimensional Micro-imaging System And Its Data Processing Methods

With the in-depth study of microscopic objects, traditional optical microscopes are no longer sufficed for scientific researches in some occasions. This dissertation proposes a multi-dimensional micro imaging system and provides a series of methods on image data processing. By incorporating micro-vision technology,3D reconstruction technology, hyperspectral image technology and image stitching technology, the system will expand the hardware of traditional microscopes and realize the function of multimodal image acquisition and multi-dimensional data processing.On system design, this dissertation put forwards a set of programs to expand the hardware of optical microscopes and develop supporting software for the system. With the system, the exposure time of image acquisition can be limited to 0.1ms-65s, regulation accuracy up to 0.1ms; movement accuracy of samples in 3D space up to 0.1um; regulation of the wavelength of transmission light to 550-1000mm and spectrum resolution up to 1nm. Both hardware and software of the system are stable, which allows the system to implement high-quality multimodal image acquisition for over one hour. On data processing, the dissertation discusses image processing methods from multiple perspectives including 2D defocus image pre-processing,3D surface reconstruction based on 2D defocus images, image stitching between adjacent field of views and 4D information reconstruction of 3D spatial data and spectral data. In realizing of above four functions, the dissertation does a comparative analysis among existing methods and proposes improved methods to some steps:(1) In 2D image pre-processing stage, the dissertation presents a mask generation method based on blank images and a pixel value replacement method based on maximum between-class variance, which can effectively remove background information and highlight noise. (2) In full focus image generating on 3D surface stage, for the clear regions extracted from multi-layer defocus image, the dissertation proposes a weighted smoothing algorithm based on intensity compensation and a method to complete the missing regions, which achieve smoothness of the overlapping regions in full focus images and the completion of missing regions. (3) This dissertation combines 3D reconstruction and image stitching algorithm together, realizing the splice of full focus images in successive field of views and 3D models. (4) The dissertation achieves 4D image data acquisition of 3D spatial data and 1D spectral data, defines the format of the 4D data, combines the 3D reconstruction method and hyperspectral imaging technology together, which realizes 4D information reconstruction of 3D model and its corresponding spectral data.The dissertation verifies all the data processing functions designed in the system by experimenting, which is quite innovative and deserves further research. The dissertation also summarizes the limitations remained in current system and proposes advices for future improvement.