Visual speech-recognition method by monitoring local deformations of a set of points of interest of the mouth of the speaker

摘要

the process includes steps of: (a) for each point of interest in each image, calculating a local descriptor and a descriptor of the local gradient of the movement; and (b) the formation of microstructures of n points of interest, defined each by an integral of order n 1; (c) convention for each nation, integral to a structured visual feature vector (d0. d3.based on local descriptors; and (d) for each page of this integral map vector by selecting a single codeword classification algorithm from a set of training codewords code book (cb); (e) generation of a time series codewords (a0 an... a3.for the successive images of the sequence of video; and (f), by means of a function of type string kernel, the similarity of the time series of codewords with another series of code words from a speaker.