1Alexander Golovan, 1Seong-Whan Lee, 2Lubov Podladchikova,
1Center for Artificial Vision Research, Korea University, Seoul, Korea.
2A.B.Kogan Research Institute for Neurocybernetics, Rostov State University, Russia.
At present, in computer vision, the approach based on modeling the biological vision mechanisms is developed in several directions. The main direction is considered to be the creation of various foveal active vision systems as the most prospective for solution of computational problems of real world image processing. Foveal systems imitate changing of the vision acuity from the center of the human eye retina (the fovea) to the periphery, and attention mechanisms for gaze control. Evidently, known computational advantages of foveal vision can be significantly improved by means of advanced biologically plausible algorithms for detection of primary features and their conjunctions, imitating composite hypercolumn sensory tuning and multiple representations of the same sensory surface in the visual cortex, and choosing perceptually important image fragments for processing with high resolution.
Face recognition on natural scene is one of complicated areas of real world image processing. It demands a fast and reliable solution of many particular tasks, such as face identification on a scene, tracking, recognition, etc. Up to now, face recognition has no effective solution in frameworks of both biologically inspired and conventional approaches. Various algorithms based on pre-attentive feature maps and attention mechanisms for detection of most informative facial regions (MIFR) have been elaborated. However, most of them detect a great number of features outside MIFR, which limits their application for processing of real world images.
In the present work, basing on the results of quantitative comparison of primary visual features inside different facial regions several biologically plausible algorithms for fast and reliable detection of MIFR have been developed and tested at processing ORL database.