Research On Image Mosaic And Fusion Technology

Three dimensional reconstruction based on image sequences has been a hot spot of research in computer vision, image sequences fusion splicing technology is the key technology of 3D reconstruction, especially the matching technology between images, directly affects subsequent image processing and the quality of the final target image. This paper carried out the research around the theme of fusion and splicing technology in 3D image reconstruction with 2D image sequences, the research content and the main work is as follows:1. Firstly, the paper introduces the related computer vision imaging theory, specifies the pinhole imaging principle, the three coordinates of Linear imaging model and their relationship and the multi view geometric principle. Then introduces the key technology in the fusion and splicing of image sequences, specifies the basic process of image mosaic, and the popular method of the key technology of image matching and image fusion.2. On the basis of the existing image feature extraction and matching algorithm, we propose an image matching algorithm based on improved Harris-SIFT and use it for image fast mosaic. The algorithm uses the scale invariant Harris operator to extracted the feature points, at the same time adding constraints to limit the number of the extracted feature points, thereby reducing the feature description and matching time, and then uses a 88 dimensional vector to build feature descriptor for each feature point, and then uses the Euclidean distance between feature points as the matching measure to match. After the matching stage, it uses the RANSAC algorithm to eliminate the error matching points and calculate the perspective transformation model parameters. Last, it uses the weighted average algorithm to merge the boundary brought on by Illumination difference.3. This paper focuses on the image fusion algorithm based on two-dimensional wavelet decomposition on the basis of the existing image fusion methods. Firstly, the basic theory of wavelet transform and image two-dimensional wavelet decomposition and reconstruction principle is introduced, and then the wavelet coefficients fusion rule is described in detail, an improved fusion rule for the low frequency and high frequency coefficients of different is put forward. Aiming at the low frequency coefficients the fusion operator combined neighborhood information entropy and local energy is used, and aiming at the high frequency coefficients the window and fractal combination is used, and the correctness and validity of the fusion rule is verified by experiment.