High-performance, massively parallel systems can be used to facilitate the following methodological steps.
Answers
Automated image processing can increase efficiency for a diverse range of applications from defect detection in manufacturing, to motion detection in video games or surveillance, to tumor detection in medical images. Advances in computer vision have made a significant impact to a variety of fields including neurobiology, artificial intelligence, pathology, security, automation, and entertainment.
As imaging systems become more efficient and produce higher resolution images, image processing workflows need to adapt to handle larger image data both in image size and number of images. The Pivotal Data Science team has recently turned their attention to image processing in the context of big image data, resulting in this series of blog posts on this topic.
Traditional approaches to image processing often start with loading images, as multidimensional arrays of intensity values, into an application’s memory for performing operations such as smoothing (filtering), segmentation, object detection, and classification. Here, we propose an alternative approach of processing images in-database in a distributed fashion, which offers 2 main advantages. First, in-memory storage is not required so images with high spatial resolution (i.e. medical images or video) can be processed regardless of the memory capacity of the underlying system.