Improving machine vision
Machine vision originated from the greater movement of artificial intelligence and the 'cognitive revolution' which sought to mimic human physiological functions, such as vision and hearing, with a computer. Early efforts were simple attempts at electronic sorting in the food industry. The first true machine vision systems were created in the 1930s and were capable of converting a picture into an array of numbers corresponding to brightness, producing a grey scale image. Both the science of machine vision and the technology supporting it have progressed greatly since that time. Today high level mathematics and cognitive psychology theory form the backbone of current machine vision systems. Following on research funded by the ESPRIT 3 Programme, scientists from universities and research centres across Europe have developed an innovative approach to machine vision. The approach takes advantage of neural network technology and employs non-linear methods, in the form of non-linear filter families (e.g. polynomial), to process digital images. An important advance is the use of algorithms to derive image features from sensor data without any intermediate pre-processing step. In addition, it is also possible to determine invariant image features, even if the particular object or objects of interest are rotated or translated with respect to the rest of the image. The results of this work can be applied to many different fields, including automated quality control (e.g. in the food industry), medical imaging, robotics, etc. The group is looking to provide training on the new techniques to ensure transfer of the technology to the European business and scientific communities.
