Research On Key Technologies Of Image Automatic Mosaic On Image Space

Based on color images acquired from large view angle CCD cameras and unmanned planes, this paper analyzes the model of uneven illumination, which is solved by MASK dodging and wavelet transform. It analyzes the spread model of geometric error. Several simulative images are processed by the method of free block adjustment. Finally we get big frame mosaic images through the arithmetic which is based on triangulated irregular network and improved SZELISKI. The major contents are presented as flows:1. According to the uneven illumination problems of large view angle camera, this paper deduces the calculating formula of image plane illumination, applies template dodging and contrast correlation simultaneously to solve this problem and accelerate the speed of data processing. Through adjusting the wavelet approximates and detail coefficients, illumination of the whole image plane is corrected, and at the same time, image details are maintained.2. Several geometric distortion models are discussed and compared. Error spread model of image connection is then analyzed. Finally, influences of both system and accidental error to coordinates of images are explaining.3. The principle and process of free network adjustment are discussed, and the error model of which are deduced. Then the normal equation is calculated. Depending on a group of simulated data, the advance value of weight is determined as a proof of true data. Finally the influence to mosaic result of angle and line element is analyzed.4. A few basic resampling methods are simply compared, and confirming overlapping image areas are determined through computer graphics. TIN method is applied to image registering and mosaic. SZELISKI mosaic algorithm is improved in this paper to control the scope of mosaic area, which is more favorable to human vision.Through researches on the above techniques, original images that acquired from large view angle CCD cameras and unmanned planes can be processed sequentially by the method of image rectification, image registration, and image mosaic to get large frame images rapidly.