Real-time on-line unconstrained handwriting recognition using statistical methods

HMM and NN for Gesture Recognition

“…In the opposite direction, this value would be negative. From this relationship the Wiimote orientation [16] can be computed. …”

“…Bell Laboratories [9] and IBM Research [16] have had the best results for handwriting recognition with probabilistic and statistic methods. Similar projects like WiiGee [20], GesApp [6], part of the Google Summer of Code 2008, and [18] have been developed to evaluate and extract features of trajectories and movements taken from accelerometers.…”

Language modeling using stochastic automata with variable length contexts

“…A popular and successful approach in speech and, more recently, handwriting recognition, is to combine linguistic knowledge with the knowledge of a specific feature domain (acoustic features in speech recognition, shape features in handwriting recognition) to form an integrated recognizer (decoder). A common technique is to embed feature models (often HMMs) into a language model by replacing each transition in the language model with the feature model of the corresponding symbol (Bahl & Jelinek, 1983;Hu, Brown & Turin, 1994;Makhoul, Starner, Schwartz & Chou, 1994;Nathan, Beigi, Subrahmonia, Cleary & Maruyama, 1995). There are also less integrated approaches where the language model network is multiplied by a segmentation network where each transition is weighted by a score generated by the corresponding symbol recognizer (Bengio, Cun & Henderson, 1994;Schenkel, Guyon & Henderson, 1995).…”

Parameter tying in writer-dependent recognition of on-line handwriting

“…In [5] the framework for a writer independent large vocabulary recognition system is discussed. Although, such systems offer the user the convenience of operation 'out ofthe box', i.e.…”