The technology is a self-designing signal processing system utilizing a multi-criteria, multi-transform neural network for pattern classification. Said data discrimination is carried out using an adaptive means for recognizing a pattern from the received signals, thereby enhancing signal processing. Finally, the system selects optimum criteria for use in classification of the plural images/signals from noisy signals.
Pattern recognition is utilized in a variety of fields such as oil exploration, biomedical imaging, voice recognition, automated data entry and finger print recognition. It can also be seen as a classification process whereby its prime goal is to optimally extract patterns in data based on certain conditions, and to separate said data into classes or categories. Current pattern recognition technologies classify data based on specific features extracted from said signals in a pre-specified manner. In other words assigning any multidimensional signal (e.g. an image) a specific class, and matching that signal with only a specific subset of other signals found within a database. Key concerns in these methods are poor discrimination capabilities and poor computational efficiencies. For example, images of the same person taken at different times may have many differences due to factors such as facial expression, lighting variation and pose. In addition, there may be shifts from one image to another and imaging noise may be present. These computational efficiencies are critical parameters, especially in real time recognition systems where large size data sets may be present.
UCF engineering professors have designed a pattern recognition/classification technique that is capable of successive and/or parallel extraction of information from the available signal data, thus leading to minimal processing requirements. Unlike the current methods, this new design employs an algorithm that continuously enhances itself using all the up-front information available, thereby reducing processing complexity and computation time, as well as increasing the fidelity of these pattern recognition systems. More importantly, the system decides what features need to be utilized for selection and/or recognition, and is able to recognize an enormously large number of patterns with a high level of accuracy.
- Reduction in computation time and complexity with an increased computation efficiency
- Algorithm can be utilized in databases which contains a large number of patterns
- Efficient signal noise filtering
- Computer software and hardware manufacturing
- Image classification
- Detection of anomalies
- Fingerprint analysis
- Speech recognition
- Compression of signals using multi transform vector quantization
- Detection of signals in noise and the various pattern recognition applications found in the biomedical field