Application of Data Mining and Knowledge Discovery Techniques to Enhance Binary Target Detection and Decision-Making for Compromised Visual Images

摘要

In an effort to improve decision-making on the identity of unknown objects appearing in visual images when the surrounding environment may be noisy and cluttered, a highly sensitive target detection scheme is developed employing nonlinear dynamical equations. It is first shown that the signal to noise ratio of this particular operation on rudimentary signals can be amplified by a factor of over one million. This means (for elementary signals) that it is possible to effectively magnify the 'quality of information' in an input signal. This procedure affords exciting opportunities in target detection. The input signal may be a sum of sine waves, it could be an auditory signal, or possibly a visual rendering of a scene. Since image processing is an area in which the original data are stationary in some sense (auditory signals suffer from nonstationary effects), the algorithm is applied to a visual rendering scene in a noisy environment. A description of the mathematical details of the algorithm used for the image enhancement is described in the appendix for completeness. The algorithm is based on a concept from nonlinear dynamics, termed 'stochastic resonance.' Such a procedure has a biological basis, and may be termed 'biomimicry' or 'biologically inspired.'.