Image Segmentation And Three-dimensional Geometric Information Extracted

Computer vision has taken an important role in AI research. This paper concentrates on digital image segmentation and 3D geometric information extraction, which are parts of the computer vision research field.Firstly, a review of image segmentation is given and some classical segmentation schemes are discussed in detail. An improved Otsu automatic threshold selection method is tested and applied on real image. A novel low-level image processing method named SUSAN is also tested and the experiment results demonstrate the effectiveness.Secondly, a practical method for 2D object description is presented and implemented. A novel line-extracting method is proposed and its efficiency and accuracy are demonstrated by corresponding experiments. Further more, two methods to extract feature points are tested by experiments and the advantage of each method is also discussed.Then, the techniques of image matching are discussed and a new image-matching method that can resist image rotation is given. The points matching techniques are analyzed by implementing a kind of relaxing methods.Finally, the techniques of camera calibration and 3D reconstruction are discussed. A camera calibration method is tested by simulation.