Convolutional Neural Networks (CNNs) are readily used today with many image processing tasks in various applications, including medical imaging, manufacturing, and remote sensing. In recent years, CNNs have shown promising results in registration, or alignment, of two images, for the purpose of combining information together from different imaging modalities. Much of this work has been performed in the area of medical imaging; however, there remains further advances to be made in other applications, such as manufacturing. Our research focuses on the development of two different image registration frameworks: (1) Sequential Registration CNN (SeqRegCNN) a model that takes two images and processes them as a single two channel input; (2) Parallel Registration CNN (ParRegCNN) a model where each image is processed individually in its own CNN. We applied both models to register a set of Polarized Light Microscopy (PLM) images to support an ongoing research effort that investigates non-destructive methodologies for detecting Microtexture Regions (MTR) in titanium metals to perform quality control in the manufacturing process of airplane parts. Our research has shown a valid proof of concept of using these models. The combination of both sequential and parallel models can assist in image registration techniques across a wide spectrum of applications, while reducing the effort in preprocessing the images.