Method and apparatus for face recognition based on hierarchical Principal Component Analysis

摘要

PURPOSE: A face recognition method and system is provided to make the number of searched data proportional to a similarity of a corresponding eigen value when a learned image is searched for hierarchically from a lower level to a higher level so that it can enhance a calculation or a matching performance in recognizing a face image. CONSTITUTION: The method comprises steps of setting an n as "1" where 1- n - k, and the k is the number of the learned image(10), projecting learned images, converted in a PCA(Principal Component Analysis) space and stored at a database, and an input face image to be recognized, in an arbitrary nth eigen space for obtaining an egien vector and an eigen value(12), obtaining an Euclidean distance between the learned face image and the input face image by using the nth eigen value(14), extracting a set of learned vectors most similar to the input face image(16), determining if the number of the eigen values remained in the set of the learned vectors is one(18), in a case that the number is not one, increasing the n by one and repeating the step 12(20), and in a case that the number is one, determining the face image corresponding to the remaining eigen value as the most similar face image(22).