MULTIMODAL USER RECOGNITION ROBUST TO ENVIRONMENT VARIATION

摘要

The present invention provides a method to enhance user recognition performance with a combination or selection of modality depending on variation of the external environment. The present invention relates to a multi-modal user recognition method, which provides a multi-modal user recognition method which includes: a step where facial similarity regarding facial feature vector of the an input image is calculated using face model templates; a step where vocal similarity regarding vocal feature vector of input audio is calculated using speech model templates; a step where a first value is created by applying to the facial similarity a first weight obtained from using the size of lighting associated with the input image; a step where a second value is created by applying to the vocal similarity a second weight obtained from using the size of noise associated with the input audio; and a step where the user is recognized using the final score obtained from combining the first and the second value.;COPYRIGHT KIPO 2015