Research On The Basic Geometric Shapes Detection In Digital Images And The Application

In the field of image understanding (image analysis), shapes are very useful senior information. As the lines, circles and ellipses are common geometric shapes in nature and man made objects, extracting them from digital images has a very important role in image understanding. Other more complex shapes can be derived from those simplest shapes. The extracting processing is essential step in computer vision. Efficient recognition of lines, circles and ellipses have been widely used in many applications. For example, engineering drawing vectorization, robot vision, image segmentation, face detection, computer aided design, object locating, imaging based measuring system etc.In this thesis, we started a research on lines, circles and ellipses detection in digital images and made an automatic detecting and aided drawing system for microscope based on the research results. Detail research works and several new ideas can be described as following:(1) Presents some basic strategies of image understanding. Summarizes academic and experimental research accomplishments on lines, circles and ellipses detection. Introduces and analyzes some algorithms which are commonly used in basic geometric shapes detection.(2) Presents an edge detection guide line for lines, circles and ellipses detection. Chooses ADM operator as the best edge detection algorithm for basic geometric shapes detection by comparing the performance with three other operators: Canny, Sobel-Zernike and SUSAN.(3) Proposes an improved randomized ellipses detection based on least squares approach. Experiments with original algorithm show that the improvements reduces the dependence on parameters of detection algorithm, enhances the speed, stability and accuracy of ellipse detection, while preserving the anti-noise ability of original algorithm. Experiments with other popular algorithms show that the new method has much higher accuracy and can be used in much wider fields.(4) Gives a contour tracing (boundary tracing) and Curve Segmentation based lines, circles and ellipses detection. Analyzes the performance and points out some weak-nesses of the algorithm.(5) Makes an automatic lines, circles, ellipses detecting and aided drawing system for microscope by using ADM operator, Probabilistic Hough Transform Lines detection, Improved Randomized Circles, Ellipses detection based on Least Squares Approach and line, circle, ellipse fitting algorithms. Experiments show that the automatic detecting results of the system are very stable, error of the measurement is about 3%, the whole system can meet the measurement requirements.