基于改进的标记分水岭方法的棒材识别

Bar identification based on improved marking watershed method

针对图像中棒材识别问题,提出了一种新的标记分水岭的棒材分割识别方法,该方法结合距离变换与梯度重构对标记点进行限定,使用分水岭变换完成棒材识别。首先利用棒材图像低频部分进行阈值的自动提取,对区域对应的距离图中的局部极值点进行筛选。提取棒材图像中低频成份对应的局部极值区域,结合局部极值点得到前景标记。然后基于前景标记进行距离分水岭变换,将分水岭变换得到的脊线作为背景标记。最后对梯度重构以后的棒材图像基于标记进行梯度分水岭变换,得到棒材分割识别结果。本文方法一方面结合距离图进行标记点筛选及补充得到准确的棒材数目,有效解决了分水岭算法目标识别的过分割问题。另一方面保留了梯度图像分水岭变换后边缘定位准确的优点,得到完整的棒材轮廓。实验结果表明,该方法满足图像识别的实时性、鲁棒性、准确性要求,可将其应用于类圆目标识别及相关领域。

Aiming at the problem of bar identification in image,A new method of bar segmentation and identification for marking watershed is proposed.This method combines the distance transformation and gradient reconstruction to define the marking points,and uses watershed transformation to complete bar identification.Firstly,the low-frequency part of the bar image is used to automatically extract the threshold,and the local extreme points in the distance map corresponding to the area are screened.Extract the local extremum region corresponding to the low-frequency component in the bar image and combine the local extremum to obtain the foreground mark.Then the watershed is transformed based on the foreground marker,and the ridge line obtained by the watershed transformation is used as the background marker.Finally,after the gradient reconstruction,the bar image is the gradient-based watershed based on the marker to obtain the bar segmentation recognition result.In this paper,we combine the distance graph to select the marker point and add the exact number of bars,and solve the over-segmentation problem in the target recognition of watershed algorithm effectively.On the other hand,the advantages of accurate edge location after watershed transformation of gradient image are preserved,and the complete bar profile is obtained.Experimental results show that this method satisfies the real-time,robustness and accuracy requirements of image recognition,and can be applied to the recognition of quasi circle target and related fields.