METHOD FOR RECOGNIZING FACE USING VARIABLE ELLIPSOIDAL MASK AND MORPHOLOGICAL FEATURES FOR ROTATED FACE OR PRESENCE OF ADDITIONAL OBJECTS

摘要

PURPOSE: A method for recognizing a face using a variable ellipsoidal mask and morphological features for a rotated face or presence of additional objects is provided to correctly extract a face region regardless of size/number/rotation of the face on an input image by determining a candidate face region, extracting the features, and applying a learned neural network. CONSTITUTION: An outline is detected from the input image(S100). A weight is calculated by mapping the variable ellipsoidal mask to the outline while position/size of the mask is changed. If the weight is placed within a weight range by comparing the weight with a weight range of the set candidate face region, the region is set as the candidate face region(S200). A candidate eye region and a candidate mouse region are detected within the candidate face region by using an accumulative histogram(S500). Using each candidate region, a parameter representing the morphological/brightness features of the face is calculated(S600). The face region is detected by inputting the parameter to the neural network that learns diverse face pattern images(S700).