A Combined Matching Algorithm Based On Image Segmentation For Multi-Source And Multi-Temporal Satellite Imagery

As the continuous enrichment of data acquisition techniques, the number of satellites on orbit has increased dramatically both at home and abroad, and getting massive multi-source satellite data has become possible. And, due to the limitations of imaging environment, revisiting period and other factors, a single satellite cannot obtain valid data that covers large range in a short time, and the information it provides has been completely unable to meet the urgent need of wide application. By now, the modern photogrammetry has gradually developed into a new phase of multi-sensor, multi-spectral and multi-temporal. Therefore, as a core issue of photogrammetric processing, image matching results directly determine the quality of the final products. However, due to the current bound of techniques and limitations of traditional thinking, joint matching technology for multi-source, multi-temporal satellite imagery is not yet mature, and there are still many problems to be solved. Therefore, it has great significance to explore new methods of image matching and have a further research of joint matching technology for multi-source, multi-temporal satellite imagery.With the goal of matching different sensor and temporal satellite images, this thesis innovatively adopts image segmentation approaches in image matching algorithm in Photogrammetry field, and proposes a joint matching methods based on region segmentation. Also, the author designs a processed and practical matching programme. The main work is as follows:1?There still exist many problems in the present image segmentation technology, such as over segmentation, under segmentation and weak ability of processing massive satellite image. To solve these problems, this paper proposes a parallel partitioning algorithm on the basis of spectral characteristic, texture characteristic and edge characteristic. This algorithm runs with the condition of multi-computers and multi-cores. Firstly, the image is divided into a series of continuous, smooth segmentations for space constraints in subsequent matching. Secondly, image contour lines which will be useful for matching primitive is extracted in segmentation result by the method of vectorized method based on run-length coding. Additionally, according to the adjacent relation between the region segmentations, the crossing points between regions are used as feature points in the following image matching. Therefore, in this paper, the image segmentation and its results run through the algorithm, and play a role of guidance for matching images to ensure satisfactory matching results.2?In order to solve the problem of the large direct positioning error of orbit and attitude parameter acquired from Chinese satellites, this paper presents a methods based on contour correlation to correct the initial position of corresponding points which are predicted by positioning model. This methods can improve the prediction accuracy of initial corresponding points, and effectively reduce the searching region of matching, and enhance the reliability of the matching constraints, thereby improving the accuracy and stability of the matching results. Firstly, cloud rehabilitation based on SVM (Support Vector Machine) is used to detect and reject unreliable segment region and contour caused by cloud covered. Secondly, determine and describe the key point of contour to find the optimal local candidate curve segment of the contour. After that, a HOGC (Histogram of Oriented Gradients based on Contour) operator is presented to determine the corresponding contours in candidate curve segments. Finally, according to the correlation of key points on the corresponding contours, we can compensate prediction error of orbit and attitude parameters in the image space.3?Based on the analysis of problems in the existing matching methods, considering the characteristics of multi-source, multi-temporal satellite imagery, this paper presents a feature point matching method assisted by global SRTM (Shuttle Radar Topography Mission). By the following research:selection of the feature point, determination of matching constraints, optimization of matching strategy, detection of mismatch, we can access to reliable and high-precision corresponding points. Then, the corresponding points act as the seeds which can provide reliable prior knowledge for the subsequent matching propagation.4?In order to obtain a denser matches, we propose a matching propagation method based on segmented regions constraint. Using the labeled segmented regions as regional constraints, combined with the texture and geometric similarity, we constructed DANCC (An similarity measure integrates Distance, Angle, and Normalized Cross-Correlation), aiming at to enhance validity of the texture poor regions and repeat regions, and to improve the accuracy rate and robustness of matching propagation.In this paper, we present a region segmentation-based multi-source and multi- temporal satellite image matching algorithm, which provides a novel matching thought, takes account into the matching drawbacks in practical applications, and explores the matching feasibility in difficult matching regions. To estimate the applicable field of this method, using real data, the multi-view and large data satellite imagery matching experiment, control points automatic matching experiment with Google Earth images, and automatic DSM (Digital Surface Model) generation experiment are carried out. The experimental results proved that the algorithm proposed by this paper can be well used into control point acquisition base on automatic matching with the existed geographical data, and reliable corresponding points determination for bundle adjustment, and DEM (Digital Elevation Model) generation. Therefore, this algorithm lays a foundation to the combined photogrammetric processing with multi-source, multi-temporal satellite data in the future.It is worth mentioning that this method has been successfully applied to multiple model and engineering projects, such as Resource Series Satellites Processing System and High Resolution Series Satellites Processing System.