Matrox Imaging is pleased to unveil Matrox Imaging Library (MIL) X, the latest name for its celebrated vision software, featuring two major updates. This field-proven software development kit (SDK) includes an extensive collection of tools for developing machine vision applications.
The latest MIL X service pack delivers a range of new features and functionality, including the training of deep neural networks for image-oriented classification; coarse segmentation using image-oriented classification based on deep learning; a revamped plus augmented 3D display, processing, and analysis offer; and support for High-Dynamic-Range (HDR) imaging. A companion update also introduces many enhancements to the MIL CoPilot interactive environment including, most notably, support for training a deep neural network.
Hands-on training and training-as-a-service options
The newest MIL X service pack expands the capabilities of its classification tools, which make use of deep learning technology—specifically convolutional neural networks (CNNs)—to analyze images of highly textured, naturally varying, and acceptably deformed goods. To perform inference, the CNN must first undergo training. MIL X provides the necessary infrastructure to build the required training dataset—including the labeling of images and augmenting the dataset with synthesized images—as well as monitoring and analyzing the training process. It supports different types of training, such as transfer learning and fine-tuning, all starting from one of the supplied pre-defined CNN architectures.
With the aim of catering to different user needs and constraints, this MIL X service pack now affords customers two options for CNN training. Users can opt to train a CNN on their own, or they can continue to engage Matrox Imaging’s team of vision experts to perform the training on the users’ behalf through Matrox Professional Services. For users with limited deep learning experience, Matrox Imaging-led training provides a way to jump-start the process of using a CNN for particular automated visual inspection applications, with the confidence of being assisted by a team of skilled practitioners.
Coarse segmentation from classification using a CNN
MIL X’s image-oriented classification makes use of deep learning technology in two distinct approaches: a global approach that assigns images to classes and a coarse segmentation approach—introduced in the new service pack—that maps image neighborhoods according to categories. The latter ultimately identifies and roughly locates the presence of specific features or defects.
HDR imaging within registration toolset
With the latest service pack, MIL X similarly expands the registration toolset, adding a new feature for performing HDR imaging. The HDR technique produces an image with a greater dynamic range of luminosity than what is possible in a conventional image. The resulting single image thus brings out detail in both the dark and bright areas that are not otherwise seen together.
Faster prototyping and development with MIL CoPilot
MIL X incorporates MIL CoPilot, the interactive environment for experimenting, prototyping, and generating code. A companion update to the newest service pack adds training and inference support for image-oriented classification using deep learning.
Serving as the interface for user-directed CNN training, MIL CoPilot lets users label and augment the required dataset, visually monitor the training process, and finally, view results in clear, concise tables.
“This latest release of MIL delivers a broad range of new functionality,” said Pierantonio Boriero, director of product management, Matrox Imaging. “With the expansion in classification capabilities, inclusion of HDR imaging, a makeover to its 3D functionality, and much more, MIL X continues to forge ahead as a premier SDK for machine vision application development.”
Availability
MIL X Service Pack 4 and the companion update to MIL CoPilot are available now in early access form through the software’s update service. Their official releases are slated for Q2 2020.
