改进型相位相关法在药液图像配准中的应用

摘要

For bottled liquid micro foreign particle visual inspection system, due to the platform and camera vibration, resulting in the bottle in image sequences with varying degrees of random offset, we propose a liquid medicine image registration method based on modified phase correlation. First, we introduce the projected image into origin phase correlation registration algorithm to divide it into two one-dimensional registration algorithms, which can greatly improve the algorithm’s operation speed. Second, we introduce Hanning window function filtering to overcome the harm to the registration accuracy due to the “discontinuity” of the image edges. Third, we introduce the peak estimation technique to improve the registration accuracy to sub-pixel level. Experimental results show that the proposed method can achieve 1/100 pixel registration accuracy, and has certain immunity to luminance change and noise. The method can meet the requirement of the actual production of bottled liquid micro foreign particle visual inspection system.%针对瓶装药液微小异物视觉检测系统中，由于平台和相机振动所导致的序列图像中瓶体出现不同程度的随机偏移问题，提出了一种基于改进型相位相关法的药液图像配准方法。首先，引入图像投影将原始相位相关配准算法分解为两个一维配准算法，显著提高算法实时性。其次，引入Hanning窗函数滤波，克服图像边缘“不连续性”对配准精度的影响。最后，针对原始相位相关配准算法只能达到像素级配准精度的缺点，引入峰值估计技术，将配准精度提高到亚像素级。实验结果表明，本文方法配准精度可达1/100像素级，并对于亮度变化、噪声干扰具有一定免疫性，能满足瓶装药液微小异物视觉检测系统的实际生产要求。