Research Of Image Mosaic Algorithm Based On Feature Points Detection

Automatically creating large viewing angle, high-resolution images are core study of remotesensing, computer graphics&image processing, video automation monitoring, etc. Moreover, imagemosaic is related to several key aspects such as detection and matching of image feature pointdetection, transformation model parameter estimation between the images, image mosaic effectoptimization after transformation and so on.After presenting the research status, research significance, application area of this field, the thesismainly made a detailed study and experimental verification of image mosaic based on feature pointsand spherical model. Especially, it comparatively analyzed the two common feature point detectionalgorithm in image mosaic----SIFT and SURF.In addition, this thesis studied and experimentally verified the model estimation methods such asfeature points matching based on k-d tree and random sampling statistic.Based on the above research, a spherical mosaic method based on feature points matching wasdesigned and implemented. It basically estimated the transformation parameters among the imagesequences with the help of feature point detection and matching, and mapped the image sequences tothe spherical model on the basis of the transformation parameters. The experiments showed that thespherical model can get more flexibility and greater perspective than cylindrical splicing.Finally, the thesis studied the post-optimization method of image mosaic, includingcompensation gaining and multi-channel integration technology. Based on this research, it designedand implemented two algorithms adapting to spherical stitching, that is, the simplified compensationgaining algorithm and multi-resolution images blending algorithm on multi-image overlapping region.The experiments showed that this method can better meet the “seamless” requirement of sphericalmosaic under the circumstance of multi-image and uneven brightness.