163 research outputs found
Document image classification combining textual and visual features.
No full textThis research contributes to the problem of classifying document images. The main addition of this thesis is the exploitation of textual and visual features through an approach that uses Convolutional Neural Networks.
The study uses a combination of Optical Character Recognition and Natural Language Processing algorithms to extract and manipulate relevant text concepts from document images.
Such content information are embedded within document images, with the aim of adding elements which help to improve the classification results of a Convolutional Neural Network.
The experimental phase proves that the overall document classification accuracy of a Convolutional Neural Network trained using these text-augmented document images, is considerably higher than the one achieved by a similar model trained solely on classic document images, especially when different classes of documents share similar visual characteristics. The comparison between our method and state-of-the-art approaches demonstrates the effectiveness of combining visual and textual features.
Although this thesis is about document image classification, the idea of using textual and visual features is not restricted to this context and comes from the observation that textual and visual information are complementary and synergetic in many aspects
Augmented text character proposals and convolutional neural networks for text spotting from scene images
No full textIn this work we propose a novel method for text spotting from scene images based on augmented Multi-resolution Maximally Stable Extremal Regions and Convolutional Neural Networks. The goal of this work is augmenting text character proposals to maximize their coverage rate over text elements in scene images, to obtain satisfying text detection rates without the need of using very deep architectures nor large amount of training data. Using simple and fast geometric transformations on multi-resolution proposals our system achieves good results for several challenging datasets while also being computationally efficient to train and test on a desktop computer
Robust Angle Invariant GAS Meter Reading
No full textIn this work we propose a novel method for automatic gas meter reading from real world images. In a wide range of countries all over the world, the existing automatic technology is not adopted, usually the reading is manually done on site, and a picture is taken through a mobile device as a proof of reading. In order to confirm the reading, a tedious work of checking the proof images is commonly done offline by an operator. With this contribution we aim to supply an effective system, able to provide a real support to the validation process reducing the human effort and the time consumed. We exploit both region-based and Maximally Stable Extremal Regions techniques, during the phase involving the localization of the meter area and to detect the meter counter digits in the detection step respectively. The evaluation has been carried out on every step of our approach, as well as on the overall assessment; although the problem is complex, the proposed method leads to good results even when applied to degraded images, it represents an effective solution to the gas meter reading problem and it can be utilized in real applications
Content extraction from marketing flyers
No full textThe rise of online shopping has hurt physical retailers, which struggle to persuade customers to buy products in physical stores rather than online. Marketing flyers are a great mean to increase the visibility of physical retailers, but the unstructured offers appearing in those documents cannot be easily compared with similar online deals, making it hard for a customer to understand whether it is more convenient to order a product online or to buy it from the physical shop. In this work we tackle this problem, introducing a content extraction algorithm that automatically extracts structured data from flyers. Unlike competing approaches that mainly focus on textual content or simply analyze font type, color and text positioning, we propose novel and more advanced visual features that capture the properties of graphic elements typically used in marketing materials to attract the attention of readers towards specific deals, obtaining excellent results and a high language and genre independence
Embedded Textual Content for Document Image Classification with Convolutional Neural Networks
No full textText Localization based on Fast Feature Pyramids and Multi-resolution Maximally Stable Extremal Regions
No full textUsing convolutional neural networks for content extraction from online flyers
No full textThe rise of online shopping has hurt physical retailers, which struggle to persuade customers to buy products in physical stores rather than online. Marketing flyers are a great mean to increase the visibility of physical retailers, but the unstructured offers appearing in those documents cannot be easily compared with similar online deals, making it hard for a customer to understand whether it is more convenient to order a product online or to buy it from the physical shop. In this work we tackle this problem, introducing a content extraction algorithm that automatically extracts structured data from flyers. Unlike competing approaches that mainly focus on textual content or simply analyze font type, color and text positioning, we propose a new approach that uses Convolutional Neural Networks to classify words extracted from flyers typically used in marketing materials to attract the attention of readers towards specific deals. We obtained good results and a high language and genre independence
Combining Textual and Visual Features to Identify Anomalous User-generated Content
No full textAnomaly detection has extensive use in a wide variety of applications, such techniques aim to find patterns in data that do not conform to expected behavior. In this work we apply anomaly detection to the task of discovering anomalies from user-generated content of commercial product descriptions. While most of the other works in literature rely exclusively on textual features, we combine those textual descriptors with visual information extracted from the media resources associated with each product description. Given a large corpus of documents, the proposed system infers the key features describing the behavioral traits of expert users, and automatically reports whenever a newly generated description contains suspicious or low quality textual/visual elements. We prove that the joint use of textual and visual features helps in obtaining a robust detection model that can be employed in an enterprise environment to automatically mark suspicious descriptions for further manual inspection
A query and product suggestion method for price comparison search engines
No full textIn this paper we propose a query suggestion method for price comparison search engines. Query suggestion techniques are used for generating alternative queries to facilitate web users in information seeking; in this specific domain, suggestions provided to web users need to be properly generated taking into account that the suggested products must be still available for sale. We propose a novel approach based on a slightly variant of classical query-URL graphs: the query-product click-through bipartite graph. Information extracted both from search engine logs and specific domain features are exploited to build the graph, and one of the advantages of this model is that such a graph can be used to suggest not only related queries but also related products. Concepts used in the proposed method are not restricted to our context but are used in many other major e-commerce and search engine websites, we tested the model on several challenging datasets, and also compared with a recent query suggestion approach specifically designed for price comparison engines. Our solution outperforms the competing approach, achieving higher results in terms of relevance of the provided suggestions and coverage rates on top-8 suggestions
- …
