Research On Fast Image Registration And Automatic Mosaic

Image mosaic is brought for improving the visual filed of common camera which can not shoot an ultra-wide panoramic image once. Image mosaic is the art of registering and stitching some overlapped pictures into a bigger one, which is a high-resolution image with large visual angle, the whole information of multiple images and good visual effect. Image mosaic is an important research topic, and has been widely used in photogrammetry, virtual reality, video compression, video index, and so on. Therefore, in this dissertation, the author has systematically studied the basic theories and key technologies of image mosaic, and accomplished the following research work:1. Pretreatment method of image mosaicPretreatment of mosaic is mainly to prepare for image registration in order to improve its accuracy and speed, which includes image enhancement, geometrical correction, sorting of the original images. This paper has three aspects to pretreatment, they are: pretreatment based on statistical characteristics, selection for image sequence and the automatic fast sorting algorithm.2. Image matching based on local gray value codingAs the big calculation of traditional algorithms (normalized cross correlation, sum of squared difference), an improved image matching method (ILGVC) based on local gray value coding is presented. The complexity of template matching algorithm can be greatly reduced by coding using the grey distribution of different spatial area. Experiments demonstrate that ILGVC is robust to the linear transformation of grey value and is stable with noise, and it also has the stability of small angle rotation. The ILGVC is faster than traditional algorithms by two orders of magnitude, and the speed has improved twice compared with existing sub-block coding method.3. Image mosaic algorithm based on feature points This paper has systematically studied image mosaic algorithm based on corners, focused on analyzing the lack of Harris operator and traditional matching strategies. An optimized automatic mosaic algorithm is presented based on Harris. Experiment results show that the matching speed by proposed algorithm has improved by 40% as compared with those of traditional methods, and it has a high success rate. The MAD(Mean Absolute Error)and RMSE(Root Mean Square Error)of corresponding points is within 1 pixel precision, and the correlative coefficient between overlapping part also is high. The mosaic image has a desired visual effect.4. Image mosaic algorithm based on Fourier-Melllin transformationInspired by the human visual model, this paper used the characteristics of the human eyes model, which can quickly adapt rotating/scaling images, brings forward an image matching method: Fourier-Melllin. According to the problems of traditional Fourier-Melllin transformation, the improved measures are proposed to improve the reliability and accuracy. The results show that, the Fourier-Melllin transformation can be well used in image registration and mosaic with RST transformation images.5. Processing method for image mosaicAnalyzing the causes of stitching seamless, two measures are proposed: the brightness adjustment based on the mean and variance, the improvement on the linear weighted function. The mosaic image has a consistent brightness and good visual effect.