Research On Underlying Topological Feature Extraction Algorithm Of Architecture Images

Feature extraction is a very important research topic in computer graphic, which extract invariant features via computer-aided processing and analysis, and then serve other technologies, such as3D reconstruction, image retrieval and so on. While3D reconstruction based on architecture images is a hot spot in computer vision, this paper proposes point, line and plane underlying topological feature extraction and matching algorithm in architecture images. Compared with the general image underlying features, features in this paper can better serve reconstruction because of containing rich geometric information. The main content of this paper consist of junction detection, line matching and plane extraction algorithm.1. Junction is important to image understanding and analysis as a significant feature. This paper made a research on junction detection in architecture images, and proposed a new junction detection algorithm. A junction is defined as the point where several lines intersect. Given the line segments in image, first potential junctions are located in a small neighborhood around the intersection of each pair of lines. Second, our detector searches the branches connecting to each potential junction in a circular neighborhood according to its scale. Finally, the actual junctions are selected according to their connecting branches.2. Feature matching, which aimed at finding corresponding features in different images of the same scene, is a basic work in computer vision. Several factors, such as the change of illumination and affine transformation, make feature matching becomes a challenging task. This paper makes a research on line matching in architecture image, propose a novel descriptor to describe the line segments. The descriptor encode the geometry relationship between a line segment and its neighboring SIFT features. By measuring the similarities between line segments, we could find accurate correspondences between multiple architecture images which embed more intuitive information for3D reconstruction.3. Image segmentation aim at extracting plane feature in image, this paper makes a research on plane extraction in image, proposes a new plane extracting algorithm. A undirected-graph is constructed by connecting junctions obtained in image, the weights of edge in graph is measured by the geometry relationship between each pair of junctions. The algorithm searches the simple loop in graph to obtain the corresponding plane feature. Not only the location, but also the geometric contour information is obtained.The main research direction of this paper is to construct new image underlying feature extracting and matching algorithms, a lot of experimental results have proved the effectiveness of these algorithm. The insufficient discussion of the algorithms will be concerned in the future work.