Research Of Microscopic Image Mosaics

Microscopic image analysis is widely utilized in many important fields such as biomedicine image analysis, geological image analysis,metallographic image analysis. Image mosaics is built to extend field of view without making resolution sacrifice, so image mosaics technique is of great values in research and applications of image processing. It is impossible to find a single algorithm that adapt to all images because of the universality and the diversity of Microscopic image. Simply, three kinds of typical Microscopic image are analyzed ,on the basis, corresponding strategy of image mosaics is presented.To the microscopic image of the large-scale objects, a simple and fast algorithm is presented. we use the refinement tactics progressively. Firstly, it carries on rough matching in a single direction, then accurate matching is employed in horizontal and vertical directions. During the course of rough matching, enhanced template projection matching algorithm (ETPMA) is proposed to realize the high-efficient, accurate localization. During the course of accurate matching, measurement of line ratio, SSDA and spiral style search are utilized to realize fast and accurate registration. The algorithm is proved to be high-efficient and robust to noise and general illumination unevenness according to the experiments.To the microscopic image of the edge-corner objects, a full-automatic and robust image mosaic algorithm is presented. In the algorithm, improving Harris corner detector is used to improve the precision, rough matching set is obtained by comparing based on round template, while decreasing the geometric distortion such as the rotating. The robust RANSAC algorithm is used to dispel outliers, the combination of closed form Least-square and LM nonlinear optimizing algorithm ensure the accurate estimation of the geometric parameters. The experiment's results show that the whole algorithm is good.To the microscopic image of the "round objects in shape similarly", an algorithm based on segmentation-line is presented while considering the whole analyzing system. First of all, segmentation-line is theoretically proved to be invariant on similarity transformation in this kind of image. By comparing the three Local neighborhoods of segmentation-line through intensity cross-correlation the accurate matching is realized and Least-square method is used to estimate the geometric parameters. The experiment's results are intending and good.At last, we discuss the global registration of mosaics. Globally optimal image mosaics based on feature points, bilinear interpolation and multi-resolution spine technique are emphasized.