Recognizing Planar Object From Line Drawing Based On Principle Of Homograph

Line drawing is an important media for humans to communicate with computers. The recognition of planar objects from a line drawing always results in multiple solutions because line drawing lost depth information while projecting. In order to obtain much information about surfaces, some vision clues should be added, such as texture and shade. But humans can recognition its three-dimensional structure even though no these vision clues. So an important research area in computer vision is developing algorithms that can make computer easily recognize object from a single two-dimensional line drawing of an object as humans do. Although line drawings only consist of line segments and vertices on the same plane, they convey much information about the three-dimensional object structures, such as the constraint relations of points , lines and planes; It is not purely geometrical consequences but mixtures of geometry and human conventions; it has both an analogical aspect and a symbolic aspect. All of these make people easily recognize three-dimensional objects structure. But it is very difficult for computer to do it.In this thesis, six topological characteristics and geometry characteristics are proposed that are invariant under affine transformations and axonometric projection. The definition of homograph is given based on these characteristics, and then the recognizing method of the homograph is proposed according to the definition of homograph at the same time.We introduce five topological characteristics and geometry characteristics that are invariant under projective transformation specially perspective projection. The definition of homograph is given by means of these characteristics, and then the recognizing method of the homograph is proposed according to the definition ofhomograph.A new algorithm is presented for recognizing planar objects from a line drawing under axonometric and perspective projections based on principle of homograph. Surfaces of a planar object are chosen as matching features, and a surface is chosen as initial matching one, which has more edges and fewer concave vertexes. The obtained homographs can be employed to distinguish between surfaces with different characteristics, and to classify the surfaces of a planar object. Regions in a line drawing and surfaces of an object are matched based on homographs. Once a region of the line drawing and a surface of the planar object are homographs, the adjacent surfaces are following to be analyzed. Repeat this processing until all surfaces of the planar object have been examined. If the corresponding surfaces pairs are all homographs, the line drawing is corresponding to the planar object.The algorithm has a great reduction in searching space and computation time. The topological structure is introduced to analyze homograph qualitatively, the algorithm is robust and insensitive to noises of images. The geometrical structure is used to analyze homograph quantitatively, the fine discrimination between planar objects can be shown. Our algorithm can not only recognize objects with different shape, but also distinguish between different size and scale objects with same shape. Our algorithm is implemented in computer simulations. Comparing with other algorithms, a better recognized result is obtained by using of our algorithm.This thesis is the project supported by Shaanxi National Science Foundation (No.2002E224) , Jiangxi National Science Foundation (No.0311018) and Testing Technology, and Control Engineering in Nanchang Institute of Aeronautical Technology (No.2003. 013 )...