A THEORETICAL MODEL FOR COMPUTER-AIDED IMAGE RECOGNITION SYSTEM BASED ON DIVERSE INPUTS

This paper proposes a multiple senses model to match with the object/pattern recognition accuracy, which the human being achieve, with the help of five sense organs, against the machines. The proposed model can help the computer-aided pattern recognition to reach as close as possible to natural visual perception. The initial investigation based on 16 X-ray images shows that even if the input from a single source (X-ray machine) is exploited in the best possible way, using spectral analysis (Fourier, wavelets, morphological etc.), geometry, fractals, statistics etc. The classification accuracy of an object/pattern can further be improved considerably if we effectively fuse inputs received from other diverse sources for the same object. An effective fusion of information, collected from diverse sources, can make computer-aided image recognition closer to natural visual perception. The initial investigation based on 16 X-ray images affected with Tuberculosis shows that the disease recognition accuracy got improved from 85% to 93 % when we added two additional features from a different source of input.