Face Recognition Method for Static Image
Chunxu Jiang, Xuejiao Zhang · 2016
The face recognition technology is mainly according to those characteristics that exist large differences among different individuals and are relatively stable for the same person.Which feature forms that will be adopted concretely vary with diverse recognition methods.The static face recognition methods of earlier studies basically included face recognition methods based on correlation matching and geometric features.At present, the face recognition method for static image has the following three major research directions: the first are identification methods based on the connecting mechanism, consisting of elastic graph matching method and general neural network method; the second are the identification methods according to statistics, including the hidden Markov model method and eigenface method; the third are some other comprehensive approaches combining the two. Main methods of face recognitionFace recognition method based on geometric features Common geometric features contain face facial features (such as the mouth, nose, eyes, etc.), face shape characteristics and the location features of the five sense organs on the face.When extracting these features, a priori knowledge of the human facial structure is usually applied.In the recognition method according to geometric features, the identification is actually the matching for feature vectors.The judgment method based on Euclidean distance is one of the most commonly used identification methods, and the most critical first step of positive recognition method is suitable normalization.It refers to not depend on human face rotation changes of the image and position scales.Geometry features used in the recognition mainly make geometrical relationships and shape of face organ as based feature vectors, whose components generally indicate the Euclidean distance, angle and curvature of specified two points on the face.For example, Brunelli and Piggio extracted 35D face feature vector method for pattern classification by using integral projection.Face recognition method based on geometric features has the following three advantages: ①conform to the mechanism of human face recognition, easy to understand; ②only needing to store a feature vector for each image with a smaller amount of storage; ③less sensitive to the change of the light.But this method also has the following three shortcomings: ①it is very difficult to extract stable feature from the image, especially when the feature is covered; ②it has poor robustness for strong attitude and the expression changes; ③ general geometric characteristics merely describe the basic shape and the structure relationship of the population, while it ignore the local fine features, thus resulting in loss of patial information, so this method is more suitable for rough classification.Face recognition method based on template matching Face recognition method based on template matching is a classic method of face recognition, which directly calculates the matching degree between two given images.Therefore, this method requires that two images have the same dimensions, light conditions and orientations, so we have to do the pre-processing work for intensity normalization and scale normalization.In order to overcome the shortcomings of general template matching, the elastic template matching method is proposed.Flexible template defines a set of adjustable parameters designed according to a priori knowledge of the characteristic shape.For purpose of obtaining the value of this set of parameters, a