钢坯表面点印字符图像自适应阈值分割方法

Adaptive thresholding segmentation method for dot⁃printed character image on billet surface

因金属表面反光敏感和现场光照条件多变,钢坯表面点印字符图像难以有效分割和识别,文中提出一种基于改进的背景估计和最大熵法的钢坯表面点印字符图像自适应阈值分割方法。利用二维高斯函数与原图像卷积操作得到图像尺度空间与差分尺度空间,结合高斯函数特性判断出尺度空间中最佳背景估计层;采用改进的背景差法消除钢坯表面局部反光不均匀干扰,得到前景目标图像;为了进一步提升分割效果,采用一种重构一维直方图的鲁棒最大熵法进行全局阈值分割。利用生产现场采集的钢坯表面点印字符图像进行测试与对比,测试结果表明,所提方法分割所得二值图像噪声小、边缘轮廓清晰,具有较强的鲁棒性和较好的适应性。

It is difficult to effectively segment and recognize the dot⁃printed character image on a billet surface because of the reflective sensitivity of metal surface and the variable lighting conditions on the spot,so an adaptive threshold segmentation method based on improved background estimation and maximum entropy method(MEM)is proposed for dot⁃printed character images on billet surface.A two⁃dimensional Gaussian function and the original image convolution operation are used to obtain the image scale space and the difference scale space.The best background estimation layer in the scale space is determined in combination with the characteristics of Gaussian function.An improved background subtraction method is used to eliminate the interference from local uneven reflection on the billet surface,so as to obtain the foreground object image.A robust MEM for reconstructing one⁃dimensional histogram is used for global threshold segmentation to further improve the segmentation effect.Dot⁃printed character images collected on billet surfaces on the real production spot are adopted for the test and contrast.The results show that the binary images segmented by the proposed method has low noise and clear edge contour.In conclusion,it has been proved that the proposed method has strong robustness and good adaptability.