Computers and automated systems have accelerated productivity and improved quality and reliability for nearly everything, and are destined to take on increasing roles as time moves on. One major limiting factor for automated systems is their inability to categorize and recognize objects, particularly under changing lighting or other conditions. Examples of how this could be useful include automatically detecting manufacturing defects, analyzing changes between two images (e.g. medical scans), noise filtering in radio frequency communications, and extracting weak signals or images from various sources.
Current approaches to making “smart” systems generally build custom solutions for every problem with very specific outcomes; for example, self-driving car systems and facial recognition software, which have very specific features and approaches built in that usually do not translate to other applications very well. Other examples such as automatic defect detection require tightly controlled lighting, and often require the object being inspected to be in the same position to be able to identify problems. Generally, these automated systems can, when conditions match the programmed expectations, identify that there is a problem, but have very limited ability when measurement conditions are dynamic or the situation changes in unanticipated ways.
The MorphoHawk analysis system enables automatic feature detection and classification across a host of applications in changing environmental conditions. In general, MorphoHawk can be trained to identify features of interest, and will then group features in a scene (e.g. image, signal, etc.) and categorize them according to the rules with which it was conditioned. After it has categorized an image or other multi-dimensional data set, it can compare the features it has identified with subsequent data sets, allowing it to detect changes (e.g. manufacturing quality control), or detect the introduction of new features (e.g. a tumor in medical scans or a person entering a scene monitored by a camera). It can discern between an object and its shadow, meaning it can handle differences in registration and light conditions in dynamic environments. This is possible because MorphoHawk algorithms characterize and compare morphological features, rather than conducting a binary analysis (e.g. light vs. dark).