Research On Image Mosaic Based On Feature Points

Image mosaic technology can compound several images of the same target which taken with different equipment, at different time and from different angle to a high-resolution and wild-vision image. It eliminates the redundant information in the image sequences on the one hand,also solves the contradiction between vision and resolution on the other hand, thus having strong practicality, and has been widely used in medical, military, education and other fields. In recent years, with the rapid development and the wide applications of intelligent monitoring system, the demand for image mosaic technology in intelligent monitoring environment stands out increasingly. In the process of using the camera monitoring, there are also problems that information is redundant and global vision conflicts local resolution. And because of the persistence of monitoring and the mobility of the monitoring object, these problems occur more frequently. Therefore, image mosaic technology has important application requirement and research significance in areas such as intelligent monitoring. Image mosaic includes three key technologies-- image matching, image transformation and image fusion. This thesis focuses on image matching technology. The main research contents and achievements are as follows:① The image mosaic algorithm based on SIFT(Scale Invariant Feature Transform) is verified using a lot of real images which have different number of feature points. The verification process demonstrates that, no matter how many feature points can be extracted, the main factor affecting the time consuming of mosaic algorithm is the feature description process in image matching. Therefore, in order to reduce the time consuming of SIFT feature description, this thesis researches a series of improved SIFT description algorithms at home and aboard from three directions including neighborhood shape, description object and dimension reduction, discovers that the algorithms which improve description object not only perform best on reducing description time, but also do well on matching performance and descriptor dimension. Thus improving description object can be regards as a feasible research point to improve performance.② This thesis chooses to improve description object. Aiming at the defect of LBP(Local Binary Patterns) and CS-LBP(Center Symmetric Local Binary Patterns), first, considering that the samples’ neighbors have the same contribution when keeping one sample out of two, the neighbor contribution weighting strategy is proposed. Then considering of the stability of the feature regions under different resolutions, the multi-resolution fusion strategy is added. According to the two strategies, the CWCS-LBP(Contribution Weighting Center Symmetric Local Binary Patterns) descriptor and the MFCS-LBP(Multi-resolution Fusion Center Symmetric Local Binary Patterns)descriptor are constructed respectively, thus improves the readability and robustness of the descriptor under transformations such as viewpoint and scale, finally creates an image mosaic algorithm based on MFCS-LBP.③ The descriptor experiment, mosaic algorithm experiment and practical application experiment are respectively carried out to test the time-consuming and performance of SIFT, CS-LBP and MFCS-LBP. Experimental results demonstrate that, compared with the image mosaic algorithm based on SIFT, the algorithm base on MFCS-LBP proposed by this thesis reduces the whole time-consuming of mosaic and improves the stability under the scale and viewpoint transforms which are common in the process of mosaic.