Recent research in the cognitive process of reading hypothesized that we do not read words by sequentially recognizing letters, but rather by identifing open-bigrams, i.e. couple of letters that are not necessarily next to each other. In this paper, we evaluate an handwritten word recognition method based on original open-bigrams representation. We trained Long Short-Term Memory Recurrent Neural Networks (LSTM-RNNs) to predict open-bigrams rather than characters, and we show that such models are able to learn the long-range, complicated and intertwined dependencies in the input signal, necessary to the prediction. For decoding, we decomposed each word of a large vocabulary into the set of constituent bigrams, and apply a simple cosine similarity measure between this representation and the bagged RNN prediction to retrieve the vocabulary word. We compare this method to standard word recognition techniques based on sequential character recognition. Experiments are carried out on two public databases of handwritten words (Rimes and IAM). The bigram decoder results with our bigram decoder are comparable to more conventional decoding methods based on sequences of letters.

We present in this paper a model that we have developed for indexing and querying web pages based on their visual rendering. In this model pages are split up into a set of visual blocks. The indexing of a block takes into account its content, its visual importance and, by permeability, the indexing of neighbors blocks. A page is modeled as a directed acyclic graph. Each node is associated with a block and labeled by the coefficient of importance of this block. Each edge is labeled by the coefficient of permeability of the target node content to the source node content. Importance and permeability coefficients cannot be manually quantified. the second part of this paper, we present an experiment consisting in learning optimal permeability coefficients by gradient descent for indexing images of a web page from the text blocks of this page. The dataset is drawn from real web pages of the train and test set of the ImagEval task2 corpus. Results demonstrate an improvement of the indexing using non uniform block permeabilities.