环状物新检测方法及在敏感图像识别中的应用

A novel circular object detection method and its application in pornographic image detection

为了克服传统的Hough变换类环状物体检测的局限,提出了1种结合物体形状与外观特征的环状物体识别检测算法.识别算法使用cascade结构,分别使用灰度、纹理以及外观综合特征,按Bagging的方法训练产生一组弱分类器.这些弱分类器串接而成,并结合局部物体分割,逐个处理当前扫描窗口.相比于传统的Hough算法,新方法具有更快的检测速度.选择敏感图像作为实验对象,采集数据进行训练和检测,实验结果表明,新方法具有明显的性能优势.使用更全面的物体外观信息,按Bagging产生弱分类器的组合,能够在提高环状物体的检测性能的同时,获得理想的处理速度.

The paper proposes a new method to detect circular objects in images which performs better than Hough-like approaches. The method makes use of the shape and appearance information of objects. The detector is composed of a cascade of weak classifiers constructed by the Bagging algorithm and a local segmentation module. Three groups of local features involving gray value, texture and appearance were used in these classifiers in turn. An image＇ s window is reported as a circular object when it passes all the weak classifiers. Compared with the Hough algorithm, the detector has a faster detection speed and a better performance on the pornographic test data. Utilizing the information of object＇ s appearance and integrating it into a cascade of weak classifiers can improve the performance of circular object detection and lower the computational cost.