Keyword spotting in unconstrained handwritten Chinese documents using contextual word model

a r t i c l e i n f o Keywords: Keyword spotting Chinese handwritten documents Word similarity Contextual word model This paper proposes a method for keyword spotting in off-line Chinese handwritten documents using a contextual word model, which measures the similarity between the query word and every candidate word in the document by combining a character classifier and the geometric context as well as linguistic context. The geometric context model characterizes the single-character likeliness and between-character relationship. The linguistic model utilizes the dependency of the word with the external adjacent characters. The combining weights are optimized on training documents. Experiments on a large handwriting database CASIA-HWDB demonstrate the effectiveness of the proposed method and justify the benefits of geometric and linguistic contexts. Compared to transcription-based text search, the proposed method can provide higher recall rate, and for spotting words of four characters, the proposed method provides both higher precision and recall rate. Due to the huge volume of existing documents and the ever increasing new documents in daily life, the need of efficient document retrieval techniques is prominent. On scanning documents into digital images, character recognition and retrieval techniques can help efficiently sort the documents, summarize, find documents and locate regions of interest. Among the techniques of document retrieval [1,2], keyword spotting finds relevant documents containing the queried words and locates the word instances for further investigation. For printed documents of degraded image or handwritten documents, keyword spotting is still an unsolved problem due to the difficulty of word segmentation and recognition. Traditional character and word recognition techniques do not give sufficiently high accuracy on these documents, such that text search based on transcription (text line recognition) does not perform satisfactorily. For example, a state-of-the-art handwriting recognizer reports word recognition rate of 79.7% on online data and 74.1% on off-line data [3]. Compared to transcription-based text search, keyword spotting has the advantage that the information related to the query word can be exploited adequately to improve the recall rate of retrieval. This is particularly important for the less frequently used words, which are usually recognized less accurately by traditional recognizers. As a two-class (binary) problem, keyword spotting can easily get variable points of precision–recall tradeoff by setting variable decision thresholds. Transcription-based search is less flexible in this respect, particularly, when the recognition accuracy is low. The recall rate of transcription-based search can be improved by giving multiple …