Research, Implementation And Application Of Image Mosaic

As a research hot spot in computer graphics image processing, computer vision field and so on in recent years, image mosaic is widely used in many application areas, such as:video monitoring, TV conference, virtual reality, medical image processing, remote sensing technology and military affairs. It includes two different research branches, real-time video mosaic technique and static image mosaic technique, which are used in different situations. There are still many problems needed to be solved in these two application scenarios, this paper focus on the key restricting the speed and quality of image matching. Main contents are shown as follows:1. Due to the objective camera parallax and imperfect mosaic method, traditional multi-camera refraction real-time video mosaic system cannot achieve well mosaic effect and leads to "ghost" phenomenon in close range. In order to solve this problem, this paper proposes an adaptive real-time video image mosaic system. Firstly, in order to reduce the influence of the camera parallax, we used small pinhole camera type in the design of image collection device, and corrected the image distortion using the calibration method of Zhang; Secondly, we determine the initial geometric image registration relationships by using phase correlation method and Levenberg-Marquardt algorithm, based on all of these, we put forward a template matching algorithm to adjust and refine the image registration relation in real-time, which can forward eliminate the influence of the parallax. The experiment shows that our adaptive real-time image mosaic system can obtain good mosaic effect in different depth of field, with lower computational complexity, better real-time and easier to design, and can solve the "ghost" phenomenon happens a lot in traditional mosaics system in close range place.2. Mirror catadioptric method is a hot topic in recent years. To solve the problems of mirror pyramid cameras, such as too high precision required for designing and video image with inevitable stitching line, we proposed an improved mirror catadioptric video image mosaic system. Firstly, by the establishment of geometric imaging mathematics model, we derived the relative contribution parallax to image mosaic, and analyzed the reason of inconsistent viewpoint and the inevitability and regularity influence on image mosaic, based on this, we put forward a multi-mirror device with low requirement of designing and machining, with a screw nut, we can fine-adjust the cameras’height to make their optic centers located at a single point, register the images with the improvement ratio template matching algorithm, by the method of nonlinear weighting fusion, we can eliminate the stitching line effectively at the last step. Our system can be used in Trouble of Moving Freight Car Detection System (TFDS) to capture the whole image of van’s bottom, what the traditional TFDS image acquisition device cannot do before. Experiments indicate that our system has the characteristics of good real-time, high resolution, without stitching line and not affected by the depth of a field. By mosaic the images of van’s bottom from three cameras in TFDS, we can get a non stitching line and integrated van bottom image. At the same time, based on our design idea, we can expand it to design a mirror catadioptric panoramic camera with good mosaic effect.3. In dynamic scene images mosaic, the current image mosaic software and algorithm cannot get well mosaic result, or even worse cannot complete the mosaic process, to solve this problem, based on SIFT algorithm and dynamic programming method, we propose an improved optimal stitch line search criteria. First of all, we register the images with SIFT and RANSAC. Secondly, to resolve the problem of dividing a integrate object to two parts, that happens a lot when using traditional stitch line search criteria, we put forward an improved optimal stitch line criteria. Thirdly, we search the optimal stitch line criteria by using dynamic programming method, in this step, our method is lower in computation complexity and easier to be understood and used. At last, we bring a hierarchy multi-resolution mosaic idea, to realize the smooth transition processing of stitch line and exposure differences. Comparing with professional mosaic software (AutoStitch, Microsoft ICE, and Panorama Maker), comparative experimental results show that, our method can get well results both in general scenario and in dynamic scenario with huge exposure differences and wide baseline, and also in the static scenario with wide baseline.4. This paper uses different image fusion methods for different applications. Chapter3and chapter4are real-time video image applications, so we take a simple and fast weighted fusion algorithm, it works well on mosaic imaging in real-time; but in chapter5, for no requirement of speed, we take multiple resolution image fusion algorithm to ensure the mosaic quality, with higher algorithm complexity and well fusion result.