Cognition-based parameter setting of non-linear filters using a face recognition system
Mitsuharu Matsumoto · IET Image Processing · 2012
This study introduces cognition-based parameter setting of non-linear filters using a face recognition system. The parameter setting problem is generally solved by the maximisation or minimisation of some objective evaluation functions such as correlation and statistical independence. However, the tuned filter output is not always adequate for face recognition systems because filter and system are separately tuned using different criterion. When a single image with noise is considered, it is difficult to employ such objective functions because only a single image exists, and the original image cannot be known. When the authors consider some problems such as contrast adjustment and non-photo-realistic rendering, it is also difficult to set an objective criterion for parameter setting because there are no correct solutions other than humans’ subjectivity. To handle such cases, the authors look to some subjective information such as a face in an image, and directly employ a face recognition system as the evaluation function for parameter setting. Experimental results show that cognition-based evaluation has the potential to handle these types of problems.