19-09-2017, 10:31 AM
A very robust face detection algorithm against illumination, focus and scale variations on input images has been developed based on edge-based image representation. The multi-face detection algorithm developed in our previous work has been used in conjunction with a new decision criterion called "density rule", where only the high density groups of detected face candidates remain as faces.
The task of extracting features consists mainly in making sense of the characteristics / patterns of facial information and extracting them. However, most real-world face images are almost always intertwined with the imaging mode problems of which illumination is a strong factor. The compensation of the illumination factor using various lighting compensation techniques has been of interest in literatures with little emphasis on the adverse effect of the techniques to the task of extracting the real discriminatory characteristics of a sample image for its recognition. In this work we present comparative analyzes of light compensation techniques for the extraction of significant characteristics for the recognition using a single characteristic extraction method. In addition, red, green, blue (rgbGE) gamma coding is improved in the logarithmic domain to address the separability problem within a class of person that is applied in most techniques. From experiments with plastic surgery sample faces, it is evident that the effect of light compensation techniques on face images after pretreatment is highly significant for recognition accuracy.
The task of extracting features consists mainly in making sense of the characteristics / patterns of facial information and extracting them. However, most real-world face images are almost always intertwined with the imaging mode problems of which illumination is a strong factor. The compensation of the illumination factor using various lighting compensation techniques has been of interest in literatures with little emphasis on the adverse effect of the techniques to the task of extracting the real discriminatory characteristics of a sample image for its recognition. In this work we present comparative analyzes of light compensation techniques for the extraction of significant characteristics for the recognition using a single characteristic extraction method. In addition, red, green, blue (rgbGE) gamma coding is improved in the logarithmic domain to address the separability problem within a class of person that is applied in most techniques. From experiments with plastic surgery sample faces, it is evident that the effect of light compensation techniques on face images after pretreatment is highly significant for recognition accuracy.