MULTIMODAL USER RECOGNITION ROBUST TO ENVIRONMENT VARIATION

摘要

the present invention is a user recognition through a combination or selection of modalities according to the environmental change. It provides a method to improve the performance. The present invention comprises the steps of calculating a face similarity on the facial feature vector of the input image by using a face model templates (face model templates) of the multimodal user identification method; Calculating a degree of similarity of the voice speech feature vector of the input audio using a speech model templates (speech model templates); Generating a first value by applying a first weight obtained using the amount of light associated with the input image to the face similarity; Generating a second value obtained by applying the second weight to the audio likelihood using a noise associated with the input audio; And using the first value and the final score obtained by combining the first value and provides a multimodal user identification comprises the step of recognizing the user. ;