Abstract: 2,292 JCR
The touching character segmentation problem becomes complex when touching strings are multi-oriented. Moreover in graphical documents sometimes characters in a single-touching string have different orientations. Segmentation of such complex touching is more challenging. In this paper, we present a scheme towards the segmentation of English multi-oriented touching strings into individual characters. When two or more characters touch, they generate a big cavity region in the background portion. Based on the convex hull information, at first, we use this background information to find some initial points for segmentation of a touching string into possible primitives (a primitive consists of a single character or part of a character). Next, the primitives are merged to get optimum segmentation. A dynamic programming algorithm is applied for this purpose using the total likelihood of characters as the objective function. A SVM classifier is used to find the likelihood of a character. To consider multi-oriented touching strings the features used in the SVM are invariant to character orientation. Experiments were performed in different databases of real and synthetic touching characters and the results show that the method is efficient in segmenting touching characters of arbitrary orientations and sizes.

Abstract: Convolutional Neural Networks are the de facto models for image recognition. However 3D CNNs, the straight forward extension of 2D CNNs for video recognition, have not achieved the same success on standard action recognition benchmarks. One of the main reasons for this reduced performance of 3D CNNs is the increased computational complexity requiring large scale annotated datasets to train them in scale. 3D kernel factorization approaches have been proposed to reduce the complexity of 3D CNNs. Existing kernel factorization approaches follow hand-designed and hard-wired techniques. In this paper we propose Gate-Shift-Fuse (GSF), a novel spatio-temporal feature extraction module which controls interactions in spatio-temporal decomposition and learns to adaptively route features through time and combine them in a data dependent manner. GSF leverages grouped spatial gating to decompose input tensor and channel weighting to fuse the decomposed tensors. GSF can be inserted into existing 2D CNNs to convert them into an efficient and high performing spatio-temporal feature extractor, with negligible parameter and compute overhead. We perform an extensive analysis of GSF using two popular 2D CNN families and achieve state-of-the-art or competitive performance on five standard action recognition benchmarks.

Abstract: Multi-label zero-shot learning strives to classify images into multiple unseen categories for which no data is available during training. The test samples can additionally contain seen categories in the generalized variant. Existing approaches rely on learning either shared or label-specific attention from the seen classes. Nevertheless, computing reliable attention maps for unseen classes during inference in a multi-label setting is still a challenge. In contrast, state-of-the-art single-label generative adversarial network (GAN) based approaches learn to directly synthesize the class-specific visual features from the corresponding class attribute embeddings. However, synthesizing multi-label features from GANs is still unexplored in the context of zero-shot setting. When multiple objects occur jointly in a single image, a critical question is how to effectively fuse multi-class information. In this work, we introduce different fusion approaches at the attribute-level, feature-level and cross-level (across attribute and feature-levels) for synthesizing multi-label features from their corresponding multi-label class embeddings. To the best of our knowledge, our work is the first to tackle the problem of multi-label feature synthesis in the (generalized) zero-shot setting. Our cross-level fusion-based generative approach outperforms the state-of-the-art on three zero-shot benchmarks: NUS-WIDE, Open Images and MS COCO. Furthermore, we show the generalization capabilities of our fusion approach in the zero-shot detection task on MS COCO, achieving favorable performance against existing methods.

Abstract: Writer Identification is an important task for the automatic processing of documents. However, the identification of the writer in graphical documents is still challenging. In this work, we adapt the Bag of Visual Words framework to the task of writer identification in handwritten musical scores. A vanilla implementation of this method already performs comparably to the state-of-the-art. Furthermore, we analyze the effect of two improvements of the representation: a Bhattacharyya embedding, which improves the results at virtually no extra cost, and a Fisher Vector representation that very significantly improves the results at the cost of a more complex and costly representation. Experimental evaluation shows results more than 20 points above the state-of-the-art in a new, challenging dataset.

Abstract: Two key factors affecting the performance of Error Correcting Output Codes (ECOC) in multiclass classification problems are the independence of binary classifiers and the problem-dependent coding design. In this paper, we propose an evolutionary algorithm-based approach to the design of an application-dependent codematrix in the ECOC framework. The central idea of this work is to design a three-dimensional codematrix, where the third dimension is the feature space of the problem domain. In order to do that, we consider the feature space in the design process of the codematrix with the aim of improving the independence and accuracy of binary classifiers. The proposed method takes advantage of some basic concepts of ensemble classification, such as diversity of classifiers, and also benefits from the evolutionary approach for optimizing the three-dimensional codematrix, taking into account the problem domain. We provide a set of experimental results using a set of benchmark datasets from the UCI Machine Learning Repository, as well as two real multiclass Computer Vision problems. Both sets of experiments are conducted using two different base learners: Neural Networks and Decision Trees. The results show that the proposed method increases the classification accuracy in comparison with the state-of-the-art ECOC coding techniques.

Abstract: In this paper we propose a symbol spotting technique in graphical documents. Graphs are used to represent the documents and a (sub)graph matching technique is used to detect the symbols in them. We propose a graph serialization to reduce the usual computational complexity of graph matching. Serialization of graphs is performed by computing acyclic graph paths between each pair of connected nodes. Graph paths are one-dimensional structures of graphs which are less expensive in terms of computation. At the same time they enable robust localization even in the presence of noise and distortion. Indexing in large graph databases involves a computational burden as well. We propose a graph factorization approach to tackle this problem. Factorization is intended to create a unified indexed structure over the database of graphical documents. Once graph paths are extracted, the entire database of graphical documents is indexed in hash tables by locality sensitive hashing (LSH) of shape descriptors of the paths. The hashing data structure aims to execute an approximate k-NN search in a sub-linear time. We have performed detailed experiments with various datasets of line drawings and compared our method with the state-of-the-art works. The results demonstrate the effectiveness and efficiency of our technique.