The Research On Multiview Based X-Ray Image Analysis Methods

Multiview X-ray image analysis is a very important technique in the field of security inspection. Compared to single view, multi views have an advantage on shape analysis and material detection, so it is necessary to do research on them.For general images, feature matching is a common method to locate one object in two views for shape analysis. X-ray images are the X-ray attenuation values when X-rays pass through the object, which indicate the attributes of materials including the density and the effective atomic number. The multiview security inspection system in public place can detect the contrabands by analyzing X-ray images.In order to analyze the shape of object, we apply feature matching methods on X-ray images and choose the SIFT of high robustness as theoretical basis. The SIFT matching is realized by matching the SIFT descriptors and using the global mapping model to eliminate the wrong matching points. For higher matching ratio, the SIFT point feature is replaced by other region features. Two region detection methods are employed, which are MSER based method and median filtering based method. The experimental results show that the matching ratios are both increased and the latter shows the highest matching rate. Then the matching experiments are carried out on X-ray images between different angle of views and the applicability of feature matching methods is discussed.In order to detect the attribute of object, we study the dual view security inspection system which can detect the overlapping objects. Firstly, the existing CT reconstruction method for density detection is discussed. Considering the special construction of the system, an improved dual view image analysis method is proposed. First of all, the interesting regions in each single view are segmented by region growing method. Then the regions from two views are assumed to be matched and the object contour of the third view is constructed in the three-dimensional rectangular coordinate system under each assumption. Finally, the back projection method is used to reconstruct the fault plane for the true object contour in order to find the most similar assumed contour, so that we get the true region matching. According to the experimental result, the improved method is an efficient dual view image analysis method which can be applied to detect the overlapping object through effective atomic number.