Abstract: Vision and Language are broadly regarded as cornerstones of intelligence. Even though language and vision have different aims – language having the purpose of communication, transmission of information and vision having the purpose of constructing mental representations around us to navigate and interact with objects – they cooperate and depend on one another in many tasks we perform effortlessly. This reliance is actively being studied in various Computer Vision tasks, e.g. image captioning, visual question answering, image-sentence retrieval, phrase grounding, just to name a few. All of these tasks share the inherent difficulty of the aligning the two modalities, while being robust to language
priors and various biases existing in the datasets. One of the ultimate goal for vision and language research is to be able to inject world knowledge while getting rid of the biases that come with the datasets. In this thesis, we mainly focus on two vision and language tasks, namely Image Captioning and Scene-Text Visual Question Answering (STVQA).
In both domains, we start by defining a new task that requires the utilization of world knowledge and in both tasks, we find that the models commonly employed are prone to biases that exist in the data. Concretely, we introduce new tasks and discover several problems that impede performance at each level and provide remedies or possible solutions in each chapter: i) We define a new task to move beyond Image Captioning to Image Interpretation that can utilize Named Entities in the form of world knowledge. ii) We study the object hallucination problem in classic Image Captioning systems and develop an architecture-agnostic solution. iii) We define a sub-task of Visual Question Answering that requires reading the text in the image (STVQA), where we highlight the limitations of current models. iv) We propose an architecture for the STVQA task that can point to the answer in the image and show how to combine it with classic VQA models. v) We show how far language can get us in STVQA and discover yet another bias which causes the models to disregard the image while doing Visual Question Answering.

Abstract: In this paper we present a crowdsourcing web-based application for extracting information from demographic handwritten document images. The proposed application integrates two points of view: the semantic information for demographic research, and the ground-truthing for document analysis research. Concretely, the application has the contents view, where the information is recorded into forms, and the labeling view, with the word labels for evaluating document analysis techniques. The crowdsourcing architecture allows to accelerate the information extraction (many users can work simultaneously), validate the information, and easily provide feedback to the users. We finally show how the proposed application can be extended to other kind of demographic historical manuscripts.

Abstract: Colorectal cancer (CRC) is the third cause of cancer death worldwide. Currently, the standard approach to reduce CRC-related mortality is to perform regular screening in search for polyps and colonoscopy is the screening tool of choice. The main limitations of this screening procedure are polyp miss-rate and inability to perform visual assessment of polyp malignancy. These drawbacks can be reduced by designing Decision Support Systems (DSS) aiming to help clinicians in the different stages of the procedure by providing endoluminal scene segmentation. Thus, in this paper, we introduce an extended benchmark of colonoscopy image, with the hope of establishing a new strong benchmark for colonoscopy image analysis research. We provide new baselines on this dataset by training standard fully convolutional networks (FCN) for semantic segmentation and significantly outperforming, without any further post-processing, prior results in endoluminal scene segmentation.

Abstract: In this paper we propose an error tolerant subgraph matching algorithm based on bag-of-paths for solving the problem of symbol spotting in line drawings. Bag-of-paths is a factorized representation of graphs where the factorization is done by considering all the acyclic paths between each pair of connected nodes. Similar paths within the whole collection of documents are clustered and organized in a lookup table for efficient indexing. The lookup table contains the index key of each cluster and the corresponding list of locations as a single entry. The mean path of each of the clusters serves as the index key for each table entry. The spotting method is then formulated by a spatial voting scheme to the list of locations of the paths that are decided in terms of search of similar paths that compose the query symbol. Efficient indexing of common substructures helps to reduce the computational burden of usual graph based methods. The proposed method can also be seen as a way to serialize graphs which allows to reduce the complexity of the subgraph isomorphism. We have encoded the paths in terms of both attributed strings and turning functions, and presented a comparative results between them within the symbol spotting framework. Experimentations for matching different shape silhouettes are also reported and the method has been proved to work in noisy environment also.

Abstract: Determining the authorship of a document, namely writer identification, can be an important source of information for document categorization. Contrary to text documents, the identification of the writer of graphical documents is still a challenge. In this paper we present a robust approach for writer identification in a particular kind of graphical documents, old music scores. This approach adapts the bag of visual terms method for coping with graphic documents. The identification is performed only using the graphical music notation. For this purpose, we generate a graphic vocabulary without recognizing any music symbols, and consequently, avoiding the difficulties in the recognition of hand-drawn symbols in old and degraded documents. The proposed method has been tested on a database of old music scores from the 17th to 19th centuries, achieving very high identification rates.

Abstract: The most relevant modules of a pedestrian detector are the candidate generation and the candidate classification. The former aims at presenting image windows to the latter so that they are classified as containing a pedestrian or not. Much attention has being paid to the classification module, while candidate generation has mainly relied on (multiscale) sliding window pyramid. However, candidate generation is critical for achieving real-time. In this paper we assume a context of autonomous driving based on stereo vision. Accordingly, we evaluate the effect of taking into account the 3D information (derived from the stereo) in order to prune the hundred of thousands windows per image generated by classical pyramidal sliding window. For our study we use a multimodal (RGB, disparity) and multi-descriptor (HOG, LBP, HOG+LBP) holistic ensemble based on linear SVM. Evaluation on data from the challenging KITTI benchmark suite shows the effectiveness of using 3D information to dramatically reduce the number of candidate windows, even improving the overall pedestrian detection accuracy.