Systematic studies on shape analysis of binary images

1.直线特征点提取方法的比较五种角点、拐点、直线与曲线连接点的提取方法。实验是在手写有模板的流程图符号上进行的。结果描述了每种方法的优点和缺点。二值图像形状分析的新算法基于数字曲线的局部对称性，提出了一种新的角点检测方法。实验表明，该方法得到了直观合理的结果，而不受所用参数值的影响。对称轴可以通过评估检测到的角点之间的对称程度来确定。该方法甚至适用于变形或部分遮挡的图形，将线条图案分解成角点处的线段并进行匹配，定义了两条线条图案之间的相似性。提出了两种二值图像的快速细化算法：一种是基于距离变换的顺序细化算法，另一种是两个子周期的并行处理算法。提出了一种顺序并行相结合的灰度图像细化算法，对灰度图像的结构分析具有一定的参考价值。正在开发一种将二值图像分解成有意义的部分的方法。流程图的结构理解研究手写流程图图像中符号、流程线和字符的识别。对图形图像中特征点的不细化检测方法进行了实验研究。提出了一种在局部子结构识别和全局一致性检查之间转换的控制机制。彩色图像中的形状分析开始于应用形状分析方法的项目，因此开发了理解彩色图像并在立体视觉中匹配两个彩色图像的项目。

1. Comparison of methods for extracting feature points from line patternsFive methods for extracting corner points, inflection points, and points connecting a straight line with a curve. Wxperiments are made on flowchart symbols handwritten with templates. The results depict merits and demerits of each method.2. New algorithms for shape analysis of binary imagesA new method for detecting corner points on digital curves is developed based on local symmetry of the shape. Experiments reveal that the method yields intuitively reasonable results independent of the applied parameter values. The axes of symmetry can be determined by evaluating the extent of symmetry among the detected corner points. This method can be applied to even deformed or partially occluded figures.Decomposing line patterns into line segments at corner points and matching those segments leads to a definition of similarity between two line patterns.3. Fast thinning and structural analysis algorithmsTwo fast thinning algorithms for binary images are developed: one is sequential type using distance transformation and the other is 2-subcycle parallel processing type. A combined sequential-parallel thinning algorithm for gray-level images are also proposed, which is useful for structural analysis of gray-level images. A method for decomposing a binary image into meaningful parts is being developed.4. Structure understanding of a diagramDiscrimination of symbols, flow lines, and characters in a handwritten flowchart image is investigated. Methods for detecting feature points in diagram images without thinning are studied experimentally. A control mechanism for transferring between local substructure recognition and global consistency check is proposed.5. Shape analysis in color imagesA project started for applying shape analysis method thus developed to understanding color images and matching two color images in stereo vision.