Abstract: Language models are used in automatic transcription system to resolve ambiguities. This is done by limiting the vocabulary of words that can be recognized as well as estimating the n-gram probability of the words in the given text. In the context of historical documents, a non-unified spelling and the limited amount of written text pose a substantial problem for the selection of the recognizable vocabulary as well as the computation of the word probabilities. In this paper we propose for the transcription of historical Spanish text to keep the corpus for the n-gram limited to a sample of the target text, but expand the vocabulary with words gathered from external resources. We analyze the performance of such a transcription system with different sizes of external vocabularies and demonstrate the applicability and the significant increase in recognition accuracy of using up to 300 thousand external words.

Abstract: End-to-end training from scratch of current deep architectures for new computer vision problems would require Imagenet-scale datasets, and this is not always possible. In this paper we present a method that is able to take advantage of freely available multi-modal content to train computer vision algorithms without human supervision. We put forward the idea of performing self-supervised learning of visual features by mining a large scale corpus of multi-modal (text and image) documents. We show that discriminative visual features can be learnt efficiently by training a CNN to predict the semantic context in which a particular image is more probable to appear as an illustration. For this we leverage the hidden semantic structures discovered in the text corpus with a well-known topic modeling technique. Our experiments demonstrate state of the art performance in image classification, object detection, and multi-modal retrieval compared to recent self-supervised or natural-supervised approaches.

Abstract: Zero-shot sketch-based image retrieval (SBIR) is an emerging task in computer vision, allowing to retrieve natural images relevant to sketch queries that might not been seen in the training phase. Existing works either require aligned sketch-image pairs or inefficient memory fusion layer for mapping the visual information to a semantic space. In this work, we propose a semantically aligned paired cycle-consistent generative (SEM-PCYC) model for zero-shot SBIR, where each branch maps the visual information to a common semantic space via an adversarial training. Each of these branches maintains a cycle consistency that only requires supervision at category levels, and avoids the need of highly-priced aligned sketch-image pairs. A classification criteria on the generators' outputs ensures the visual to semantic space mapping to be discriminating. Furthermore, we propose to combine textual and hierarchical side information via a feature selection auto-encoder that selects discriminating side information within a same end-to-end model. Our results demonstrate a significant boost in zero-shot SBIR performance over the state-of-the-art on the challenging Sketchy and TU-Berlin datasets.

Abstract: In this paper we propose an efficient queried-by-example retrieval system which is able to retrieve trademark images by similarity from patent and trademark offices' digital libraries. Logo images are described by both their semantic content, by means of the Vienna codes, and their visual contents, by using shape and color as visual cues. The trademark descriptors are then indexed by a locality-sensitive hashing data structure aiming to perform approximate k-NN search in high dimensional spaces in sub-linear time. The resulting ranked lists are combined by using the Condorcet method and a relevance feedback step helps to iteratively revise the query and refine the obtained results. The experiments demonstrate the effectiveness and efficiency of this system on a realistic and large dataset.