A Study Of Image Mosaic Of Unmanned Aerial Vehicle

Low altitude remote sensing images of Unmanned Aerial Vehicle have many advantages, such as high resolution, good flexibility, high efficiency and low cost. UAV images have been applied widely in survey and evaluations of natural disaster region, battle field scouts, environmental monitoring, etc. In order to expand field of vision and obtain higher quality and more precise position information in remote sensing image applications, two or many remote sensing images need be spliced one image. However, UAV image processing is still backward in relation to the rapid development of UAV technology, and UAV image mosaic has become the bottleneck of the effective application on UAV. Thus, it’s significant to study automatic splicing of UAV images.The thesis summarizes the effort of image mosaic that the researchers have achieved. The thesis introduces image mosaic in detail, depending on the steps of image mosaic:image pre-processing, image registration and image composition. To solve much work and low accuracy resulted from large amount of data, irregularity overlap, gray difference in UAV image mosaic, it improved image mosaic algorithm based on and proposed a novel method which used regional mutual information as similarity measure and salient regions to register was proposed.In image pre-processing, a simple algorithm based on noise-detected is presented to remove salt-and-pepper noise from gray-scale image. The signal pixels are kept untouched in the filtering process to preserve the detail of the image, while the possible noise pixels are removed by using an improved Adaptive Median Filter, which fixes the noise pixels by utilizing the dot product of the matrix of gray value and a binary matrix, named flag, that is used to record the noise detection results based on improved adaptive Extremum and Median value. A signal pixel is for the value of1while a noise is0in flag. Simulation results show that the algorithm can suppress noise effectively and preserve the image details.It’s known that the accuracy and efficiency of registration decides quality and speed of image mosaic, so registration is the cores of this algorithm. In order to increase accuracy, there has been many researchers from different angles to improve mutual information algorithm. The regional mutual information was used in this thesis, employing pixel neighborhood information to calculate information entropy. It includes comprehensive information about direction, adjacent pixels and frequency changes. The registration results using it as similarity measure have stronger robustness and higher precision. Otherwise, it is a critical problem for UAV image mosaic that how to increase registration speed with high precision. The efficiency of algorithms based on mutual information is low, and decreasing the number of pixels is an effective method to promote efficiency on image registration. In computer image processing, although salient region accounts for only a tiny portion of the whole image, it contains a wealth of information generally. First, the images were decomposed into layers with wavelet transformation and the approximate overlapping area was calculated with UAV flight parameters. Then, salient regions which were extracted by using Gabor energy and non-classical receptive field inhibition were registered layer by layer. And it used regional mutual information as similarity measure. We can get the registration parameters for the whole images according to the principle of Multi-Resolution. Experiments show that with the algorithm the stitching speed rises considerably, there is high robustness for images with noise, gray difference and low spatial resolution. And registration accuracy can reach sub-pixel level. So, the paper will be helpful to future research about this area.At last, the floating image had been rotated and translated by using registration parameters. The overlaps and flanking used a gradually weight smoothing algorithm. This method had eliminated the image mosaic seams and improved image visual effect. And finally, we can obtain the panoramic image which seems quite natural.According to the above description, this paper presents the experiment results in detail. In the end, the thesis makes a summary of the problems still existing in image mosaic and gives prospect.