Abstract
In this paper we present a new approach in Tifinagh character recognition using a combination of, k-nearest neighbor algorithm and the bigram language model. After the preprocessing of the text image, and the word segmentation, for each image character, the k-NN algorithm proposes candidates weighted of their membership degree. Then we use the bigram language model to choose the most appropriate sequence of characters. Results show that our method increases the recognition rate.
Highlights
The task of handwritten text recognition is challenging and has been a focus of research for several decades [1][3][7]
Modern recognizers use a statistical approach in which the returned characters sequence is the one that maximizes the probability with respect to the input image according to a model trained on handwritten text as well as the language according to a model trained on a language corpus
In addition to the character images database, and the dictionary size, Results obtained depends on many other parameters: the features extraction method, the size of the grid used in the features extraction
Summary
The task of handwritten text recognition is challenging and has been a focus of research for several decades [1][3][7]. A human always use his language knowledge when reading a text. The input image is not enough to estimate the most likely characters (or words) sequence, and the target language [2]. Modern recognizers use a statistical approach in which the returned characters (words) sequence is the one that maximizes the probability with respect to the input image according to a model trained on handwritten text as well as the language according to a model trained on a language corpus.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have