Figure 8 - uploaded by Li Liu
Content may be subject to copyright.
12: Confusion matrices for three single stream viseme recognition. (a), (b) and (c) are recognitions based on 8 hand shapes, 8 lips visemes and 5 hand positions. The class "silence" (the first element) is included in the confusion matrix. The red rectangle contains the concerning visemes. The last column corresponds to the deletion error D, and the last row corresponds to the insertion error I. The brighter element corresponds to the higher occurrence in these confusion matrices.
Source publication
This PhD thesis deals with the automatic continuous Cued Speech (CS) recognition in French
based on the images of subjects without using any artificial landmark. In order to realize this
objective, we extract high-level features of three information flows (lips, hand positions and
shapes), and find an optimal approach to merge them for a robust CS...