类圆形目标识别方法研究及系统实现

【作者】 肖飞；

【导师】 罗三定；

【作者基本信息】 中南大学 ， 计算机应用技术， 2004， 硕士

【摘要】 类圆形的物体是指形状接近圆形但不规则的复杂物体，该物体形状在现实生活中普遍存在，如农业生产中的果实、工业生产中的棒材、林业中的圆木端面等等，故该类目标的识别一直是图像模式识别中研究的重点。某些文章对类圆形团块目标识别问题的解决提出了一些方法，例如模板匹配、Hough变换的方法等等，这些已有的算法很多不能适应实际需要，部分算法的计算速度无法适应现实系统需求，部分算法存在识别误差，而某些针对特定目标识别效果很好的算法，适用领域较为狭窄。 针对已有的类圆形识别方法所存在的缺陷，本文在分析和研究了类圆形目标的特征之后，提出了类圆形粒子团块目标模式识别新策略。该方法充分利用最佳边缘检测过程伴生的方向信息，设计了区域统计、中心集聚算法，将团块目标轮廓信息会聚到目标中心以实现中心信息增强，同时有效地解决了目标轮廓分割及归属问题；采用重心聚合和团块特征判定算法，实现中心位置准确识别定位。 论文以该算法作为技术核心，将其应用于钢铁厂成捆棒材产品图像的目标识别中，构建了成捆棒材离线点支系统。该系统使用VB设计前台友好操作界面，采用VC++编写的dll动态库实现快速的计算机图像识别和处理算法供前台调用，达到了自动选定待处理区域、自动识别棒材半径、快速定位棒材中心、对成捆棒材点支计数的目的。 实验结果验证了本文一系列算法的有效性，证明该算法实现了计算机对类圆形目标图像的自动识别与检测。另外从其他类圆形图像的识别结果可以看出，该算法也具有较强的普遍适应性。更多还原

【Abstract】 The recognition of circlular object is always an emphasis in image pattern identification. The particle object means a complex and anomalous quasi-circle shape, it commonly exists in realistic life, such as fruit in agriculture , sticks in industry product and round log in forestry and so on. Many articles have presented various methods to solve the recogniton of the circle object, such as template matching,Hough transform. There are many disadvantages about the existing method of identifying the round type images , such as low speed, low veracity and lack of wide appliance and so on.Aim at the existing method of recognizing the quasi-circle, a new policy was proposed after researching on its features. It makes full use of the direction information which goes along with the process of the optimum edge detection, and two algorithms: region statistics and center collection are designed to realize the enhancement of the central information by converging the outline of gobbet object to the center of it,and at the same time it effectively resolves the problem of cutting up the outline of the object and attribution of it; the algorithm of barycenter convergence and gobbet feature decision are used to realize the accurate location of the central position.The above methods are successfully applied in the recognition of stick bundle images in steel factory. The way of constructing the stick bundles offline counting system would be introduced below in detail.. The interface used by the system is designed by Visual Basic and the fast kernel arithmetics are turned into dll designed by Visual C++, thus the system can automatically first select the expectant processing area ; then recognize the radius of stick; and locate the center of stick in a quick way and in the end count the sticks in bundles.The validity of algorithms mentioned above are testified by the result of the experiment, and the goal of automatic identification and detection of quasi-circle object images with computer, are achieved by it too. In addition ,the result of the recognition of other quasi-circle object shows the better universality of the new method.更多还原