Future watch: Pattern recognition
Pattern recognition enables computers to identify and distinguish objects, such as green from red apples, based on patterns. The technology is the foundation of future developments such as handwriting recognition and medical-records analysis. Expected to be viable in the next four to six years, pattern recognition will give computers 'vision', enabling them to identify customers by their faceMost people recognise The Mona Lisa with just a glimpse of the figure's mysterious smile, their senses interacting instantly with their memory to identify the image. A computer can achieve similar results by translating the same image into a huge amount of digital information, filtering out useless data, and then comparing the remaining data to a reference model. This technology, called pattern recognition, is one of the foundations for creating computer systems that are friendlier and easier to use.
Commercial applications of this technology are available today.
Handwriting recognition applications enable a Palm device to translate simple stylus strokes into text and help the US Postal Service automatically sort mail. Data mining tools, such as IBM DB2 Intelligent Miner for Data, hide complex data pattern recognition capabilities behind an easy-to-use interface.
An increasing number of medical applications take advantage of pattern recognition technology. A medical procedure called Muscle Pattern Recognition, from Myo Diagnostics, helps a physician recognise muscle impairment in patients who suffered back injuries.
Many interesting research projects are under way, focused on improving the reliability and the accuracy of pattern recognition. The technology will be used to identify a person by his or her face and, in general, to recognise, classify, and retrieve images according to their content. This application might enable an organisation to identify individual customers on its Web site via a Web cam, rather than depending on unreliable cookies.
Another interesting future application comes from research at Massachusetts Institute of Technology on recognising emotional patterns. Think of a future around-the-clock automated customer service application that can gauge a customer's mood from the dialogue, be it satisfaction or frustration, and react accordingly. The potential advantages of improved customer loyalty and the effect on the bottom line are apparent.
At the core of pattern recognition technology are highly sophisticated math formulas that convert data into manageable chunks that are easier to classify and to reconcile to a reference model. The problem is that the process produces results with a noticeable error rate. Businesses will find applications of this technology more appealing when the error rate closes the gap with the 2 per cent to 3 per cent error rate of the average human being.
For the future, expect an evolution that will lead to even more sophisticated and accurate algorithms and a continued overlap with other disciplines, such as image processing, speech recognition and robotics. It is fair to say that researchers are limited by the processing power of their computers. They have tried to overcome this by devising parallel processing schemes, called neural networks, to crunch huge amounts of data faster and more reliably. Increased processing power, together with refinements of application-specific algorithms, will be the key to advancements in pattern recognition.
We should see many improved applications of pattern recognition technology in four to six years, although not necessarily marketed as pattern recognition. When you can identify visitors to your Web site by their image, remember it was made possible through advanced pattern recognition technology.