Study On Automatic Registration And Mosaic For Remotely Sensed Imagery

Multi-source remote sensing images from different angles of incidence with different imaging mechanism provide different and complementary information in different space resolution, different electromagnetic wave bands and different time, so they are often employed comprehensively for many applications. The registration and mosaic of multi-source remote sensing images make the results more suitable to further applications. Therefore, the technique of multi-source remote sensing images registration had become a hotspot of research fields recently.In this paper, the automatic registration and mosaic methods between multi-source remote sensing images were expatiated. To meet the requirement of multi-source spectral image application, an automatic imagery registration algorithm based on image matching was proposed and and implemented firstly in this paper. There are five main steps in the workflow: 1) coarse registration, 2) extracting feature points by modified Harris detector, 3) getting registration control points (RCPs) based on image matching with pyramid image and feature points guiding strategy, 4) expurgating big error pairs by using absolute range detector and polynomial error checker, and 5) image rectifying by using TIN affine transformation. It could be clearly indicated by the results of many experimentations that the algorithm and workflow designed here are precise, prompt, and practical for geometry registration.Based on the studies above, an automatic registration method between SAR and optical images was discussed, and a new registration method based on region features matching was presented in this paper. First of all, the speckle noises in SAR image was suppressed by some modified filter, and then image segmentation based on region growing limited by the edge was used to extract close regions as invariable features. And meanwhile the close contours in optical image were extracted by edge extraction. Next, the precise feature matching is made by cost function and cross correlation, and then the center points of the matched region were used as RCPs. The images from EnviSat and TM were employed to testify accuracy of the proposed algorithm, and the results showed that the method in this paper could extract RCPs with high accuracy and accomplish the automatic registration for multi-sensor images.At last, a mosaic method was discussed. Mosaic implementation program was automatically established firstly after image registration and the methods including square error, histogram matching and weighted average were employed to reduce the radiation difference between the images, and desired results were shown by the experimentation.The paper presented several aspects in the research of remote sensing registration as following.1. A new auto-adapted feature points extraction method based on Harris operator was proposed in this paper.2. Towards the faults of slow speed, low precision and high manpower in RCPs manual selection, a mixed auto-selection RCPs algorithm was proposed, which was based on image intensity and feature.3. Towards the local high-frequency distortions, a local deformation model was employed to rectify the input image accurately.4. Some common filters were modified by taking into account the advantage of median filter.5. The contours were extracted automatically based on region growing limited by edges, and the automatic registration between SAR and TM images was achieved.6. Based on the studies mentioned above, the multi-source spectral image automatism registration and mosaic practical system were developed on the platform C++ Builder 6.