Método para o reconhecimento eficaz de palavras em cenários ruídosos combinando índices de Mor-fraenkel com hashing perfeito mínimo

Abstract

Eye-based typing systems need to recognize the words typed by the users interpreting their eye movements on an onscreen virtual keyboard. This process can be modeled as the transmission of words through a noisy channel. Thus, recognizing a word consists on selecting from a dictionary the words which are most similar to a distorted word (the typed word) that was received using a noisy channel. To accomplish this, the system selects the set of words which can be transformed into the typed one using until k character edit operations. These operations are weighted according to the knowledge about noise sources and error distributions observed in the eye typing scenario. To get these estimates can be hardly viable for large dictionaries and very noisy scenarios. In this work, we address this problem by proposing efficient methods to estimate edit distance using Mor-Fraenkel indices combined with a minimum perfect hashing. These methods allow the early processing of promising candidates enabling faster and better word recognition.