Home      Technology      Applications      Products      Solutions      About Us      Contact Us

Focused on One Common Challenge to Pattern Recognition Developers:
*  High speed identification of similar patterns in raw data that is typically variable and noisy.

Focused on Providing Easy And Powerful Applications Development Tools:
Providing a library of 4 simple functions written in C++:
 *  PMTSetup - setup and initialization of all parameters
 *  RecognizePattern - initiates the recognition of a pattern
 *  TrainNewPattern - trains a new pattern if not recognized
 *  ReEnforceLearning - reinforces a pattern if recognized (expands its range of similarity)

Focused on Embedded Applications that Require Speed:
Given enough time PMT tasks can be done using brute force search methods. Therefore application focus is on real time applications (e.g. video) or very large databases (e.g. Internet) where processing speed is essential.

The most complex computation internal to PMT is integer addition which simply drives the sequence of pattern sampling. Its overall simplicity makes it ideal for developers to use in embedded applications.

Focused on Two Application Types:
Most applications can be divided into two types both of which are in the product demo:

Unsupervised Learning: These are Data Mining types of applications. The objective is to find meaningful data (usually clusters) in a very large database or in real time digitized signals.

Supervised learning: In this case examples of patterns are already known and the objective is to identify similar ones. For example, finding similar images on the internet or learning to recognizing hand printed characters using a training set of known characters.

Focused on Patterns Represented by a List of Numbers:
This version of PMT is best applied to patterns consisting of a list of numbers that represent magnitudes rather than categories (e.g. ASCII characters). For example in signal processing a digitized magnitude of 8 is similar to a magnitude of 7 but the ASCII code for character B represents no similarity relative to the code's nearest magnitude.

Most patterns can be represented by a list of magnitudes including:
*  Digitized signals from video, audio, images and from a wide range of sensors including:
    biometric, seismic, electromagnetic, physiological, tactile, etc.
*  Histograms representing localized color distributions in images, word distributions in text
    documents, demographic data, frequency spectrums in sound, etc