Recognition of Postal Codes from Fingerspelling Video Sequence

In this paper, we propose an appearance based model, specifically 2D PCA model for recognising fingerspelling pin codes in a video. The video is handled as a sequence of frames and the frames are labelled by the digits 0-9 using user specific 2D PCA models to result in a sequence of labels. A fingerspelt pin code is inferred by analysing the resulting sequence of labels using a knowledge base of pin codes. Two measures, existence score and sequence score are introduced to achieve better recognition. Experimental results on relatively large database of fingerspelt video are presented.