A real-time fuzzy logic-based neural facial feature extraction technique
Edna Johnson, Chwan-Hwa John Wu · 1994
An innovative fuzzy logic-based neural facial feature extraction technique has been developed which is invariant to translation, rotation, and scale transformations. Because human reasoning is nonnumeric and imprecise, fuzzy logic is implemented to extract facial features in a human-like manner by first locating the "corners" of a profile silhouette, which include the nose, lips, and chin. The five critical points which define these corners include the eye, the tip of the nose, the point under the nose, the lips, and the chin. Translation, rotation, and scale invariance is achieved by performing recognition based on feature vectors composed of relative distances between critical points and the angles between these distance. This feature extraction technique is not only applicable to object and character recognition but also for an image with N pixels, and it outperforms conventional image convolution approaches. The fuzzy logic-based facial feature extraction algorithm correctly extracts features for 98% of 111 database images.>