Research Of Feature-based Image Mosaic

Image mosaicing is an active research field in digital image processing and it hasabundant applications in medical image analysis, virtual reality and remote sensingimage processing. Image mosaicing procedure usually comprises image acquisition,geometric correction, image registration and blending. Among this procedure, imageregistration is a key and core technique.In this thesis, normalized moment of inertia (NMI) and Zernike moment are usedto describe the image features. Depending on their characteristics, two imagemosaicing methods are carried out and these two methods have their owncharacteristics and complementary advantages.NMI based Image mosaicing is a fast method and it can stitch translated images(such as microscopic images) efficiently. We can match the features with NMI and itcould be calculated easily. After feature matching, calculate translation of matchingpoints in x and y directions each, then the translation can be gained by a statistical chartof quantities of matching points at different translation. This method can be implementedwith great efficiency and speed.Zernike moment based Image mosaicing can match these affined images. First,match features with Zernike moment and then eliminate false matches by trustfunction; at last affine transformation can be obtained with least mean square. Thismethod can achieve panoramic image mosaicing with cylindrical manifold.The results of experiment on zebra fish and microscopic images have verifiedNMI based image mosaicing is available when there are not many feature points andfaster than Zernike moment/phase correlation methods. The results of experiment onphotographs taken by pinhole camera shows Zernike based method can eliminate falsematches effectively and proves its robustness.