APHELION™ VERSION 3.X
Aphelion's Fuzzy Logic Toolkit
In image analysis applications, it is often desired to separate or classify parts of the image, so that measurement data can be obtained only for those objects of interest. This process can be implemented by considering several features of the objects, such as shape, color, position, texture, etc.
To support this process, Aphelion provides a Fuzzy Logic Toolkit that provides users with methods for using binary and fuzzy logic to create “rules” about how the information extracted from images should be interpreted. This toolkit allows users to convey their domain knowledge about the classification process, thereby facilitating analysis through the use of computer automation.
Multiple rules can be combined to make classification decisions which more closely mimic the human decision making process. Since these functions can be more than simple yes/no decisions, they can be flexible and forgiving in the presence of noise, ambiguity in the data, and inaccuracies in the training process. If needed, the functions can be made binary to mimic if-then-else logic. The important distinction here is that this logic is external to any code, and therefore is easily viewed, understood, and changed.
When used with Aphelion's extensive segmentation and feature extraction capabilities, users can create analysis procedures that mimic domain experts. The Classification Toolkit allows non-programmers to quickly enter 'knowledge' about how the data should be interpreted, thereby instructing the computer in how to duplicate this decision making process.
Another advantage of this approach is that classification schemes can be created based on small training sets, as statistics are not involved. Interpretation of data is based on plausibility, not probability - classification is based on supporting information present in the data and is not dependent on the frequency of occurrence. Conversely, statistical and neural network based approaches need a hundred or more training samples at a minimum. Often, such systems are sensitive to the order and frequency of occurrence of the classes being represented in the training data, a problem which does not affect fuzzy logic based systems.