Research On Application Of Image Mosaic Based On TMS320DM642

Image mosaic is an important research area of modern digital image processing.The object of image mosaic is to compose a big sight image from two or more small sight images which have some overlap fields of view of one common scene. Therefore,image mosaic technology makes a solution to the contradiction between image resolution and field. So, it has been widely used in many fields, such as panoramic maps,medical imaging, remote sensing technology and so on. Image mosaic technique usually involves image capturing, pre-processing, image registration and image fusion etc, and image registration is the key step of it. So the registration accuracy affects the ultimate image directly. Therefore, image registration is the focus in this paper.In the first, this paper describes the significance of image mosaic, current research and applications generally. Then give out a brief introduction to the basic process and key technologies of image mosaic, and it contains:(1) camera imaging model and image transformation principle;(2) the methods and performance of image registration and image fusion.The image pre-processing gives two parts:(1) suppression of image noise, use a median filter which is get through comparing the performance of several common filter to decrease the effect of image noise;(2) correction of the image distortion, using the distortion parameter which is obtained from camera calibration to rectify the image.The main work of this paper is proposing an optimization of the feature points’ location of matching algorithm which based on SIFT(Scale-Invariant Feature Transform) algorithm. SIFT feature is a description of partial image, with scale, rotation,translation invariance. It is also robustness for illumination changes, affine transformation and three-dimensional projection transformation. But the algorithm requires high memory and time-consuming in the step of creating Gaussian pyramid and difference of Gaussian pyramid.The algorithm in this paper has four steps. First, sub-sample the images which will be registered and extract SIFT features. Second, match the two set of features and remove the error points. Third, find the location of the matched points in the Gaussian pyramid of original image and optimize the position. Finally, get the affine matrix between the two images based on the optimized features. The algorithm performancemeet the desired accuracy and time-consuming through experiment.In recent years, the development of DSP provides much help for the practical application of image processing. The binocular camera splicing fusion system shows the combination image processing algorithm and DSP chip. The system selected TMS320DM642 which is from TI and completed image capturing, display, stitching,infusion algorithm.