图像配准和图像分割在印刷品的缺陷检测中占据着举足轻重的地位，成为机器视觉软件设计开发成败的关键要素。 本文研究的图像配准技术是以图像的纹理和几何特征为基础，采用VE4000视觉系统中图像配准算法理论进行实验，设计单一模板，进而根据待配准图像的特征，裁剪单一模板拼接生成基准参考图像。在VE4000配准算法的基础上，重新设计模板，将七个模板无缝拼接形成标准图像，再通过模板匹配，将待配准图像与标准图像中进行模板匹配，进而在标准图像上裁剪出与待配准图像的大小一致的基准参考图像。 本文研究的图像分割技术是以印刷品的区域特征为基础，采用C-V模型的水平集算法和BLOB分析技术进行实验。对图像分别用投影和种子预处理算法对图像进行去噪。进而采用水平集算法和BLOB分析分别对特定区域进行分割。水平集算法是根据曲线内外的能量推动初始曲线的变化，迭代，最终达到图像分割的目的。基于BLOB分析，首先对图像进行二值化处理，采用连通性BLOB分析，取得最大连通区域的特征，并对最大连通区域进行图像校正和二线性插值操作，最终，比对水平集和BLOB分析的分割结果。 比对图像配准和图像分割算法的实验结果。对VE4000图像配准算法和改进的图形配准算法分别进行操作，两种算法都能获取配准图像，然后对配准图像进行差影操作，根据BLOB分析的效果，得出改进算法的配准精度优于VE4000配准算法。采用C-V水平集算法对图像进行分割，并没有达到预期的分割效果。进而分析图像特点，采用BLOB图像分析算法对图像进行处理，该算法达到要求的分割目的，而且分割精确，速度较快，鲁棒性较强，完全满足了生产的需求。

Image registration and image segmentation play decisive role in the defect inspection of printed material, and become a key factor in the success of the machine vision software design and development. Image registration technique: aiming at the printed texture feature and geometrical characteristics, this paper uses image registration algorithm of VE4000 vision system to realize, using template matching technique to design single template, and then according to the characteristics of images, cutting single template to splice registration image, and doing registration operation. Because the VE4000 forms registration image according to each image, it need to run a long time. On the basis of this, design a new template again, seamlessly splice seven templates to standard images, and through template matching, cut out registration image of the same size of the processing image in the standard image, finally do the subtraction operation for registration images of the two algorithms. Image segmentation technique: aiming at region features in prints, this paper respectively uses level set algorithm of C-V model and BLOB analysis technology for image segmentation. Due to image shooting angles, use pre processing algorithm, then make level set operation for images after preprocessing. Because of particularity of image features and algorithms, it failed to achieve the expected effect of image segmentation. Therefore, make further analysis of characteristics of image areas, use the BLOB analysis to do two value processing for image. BLOB analysis uses projection correction and bilinear interpolation algorithm, which can eventually get better segmentation effect for image. Based on the above image registration and image segmentation theory, make experiments for the corresponding prints. For the image registration algorithm in VE4000 and the improved operation, the two methods both can obtain registration image, then make subtraction operation for the two registration images to get the difference image. According to difference image, it can prove that the improved algorithm is better than the VE4000 registration algorithm for registration. Image segmentation is aiming at the nameplate printing, firstly use the C-V model in level set to make image segmentation, this algorithm did not meet the requirements of the segmentation effect; then analyze characteristics of image, use BLOB image analysis algorithm to do segmentation, this algorithm not only achieves image segmentation, but also has higher segmentation level, faster speed, stronger robustness, and fully meets the needs of practice.