Search

This paper presents a novel iterative deep learning framework and apply it for document enhancement and binarization. Unlike the traditional methods which predict the binary label of each pixel on the input image, we train the neural network to learn the degradations in document images and produce the uniform images of the degraded input images, which allows the network to refine the output iteratively. Two different iterative methods have been studied in this paper: recurrent refinement (RR) which uses the same trained neural network in each iteration for document enhancement and stacked refinement (SR) which uses a stack of different neural networks for iterative output refinement. Given the learned uniform and enhanced image, the binarization map can be easy to obtain by a global or local threshold. The experimental results on several public benchmark data sets show that our proposed methods provide a new clean version of the degraded image which is suitable for visualization and promising results of binarization using the global Otsu's threshold based on the enhanced images learned iteratively by the neural network., Comment: Accepted by Pattern Recognition

This article gives an overview of design considerations for a handwriting search engine based on pattern recognition and high-performance computing, “Monk”. In order to satisfy multiple and often conflicting technological requirements, an architecture is used which heavily relies on high-performance computing, interactivity, and a Posix file-access model for the scientific programmers. The resulting system is able to handle billions of image files, in the order of petabytes of storage capacity, with a single mount point. Monk is operational since the year 2009.

We introduce a novel type of actor-critic approach for deep reinforcement learning which is based on learning vector quantization. We replace the softmax operator of the policy with a more general and more flexible operator that is similar to the robust soft learning vector quantization algorithm. We compare our approach to the default A3C architecture on three Atari 2600 games and a simplistic game called Catch. We show that the proposed algorithm outperforms the softmax architecture on Catch. On the Atari games, we observe a nonunanimous pattern in terms of the best performing model.

Feature engineering takes a very important role in writer identification which has been widely studied in the literature. Previous works have shown that the joint feature distribution of two properties can improve the performance. The joint feature distribution makes feature relationships explicit instead of roping that a trained classifier picks up a non-linear relation present in the data. In this paper, we propose two novel and curvature-free features: run-lengths of local binary pattern (LBPruns) and cloud of line distribution (COLD) features for writer identification. The LBPruns is the joint distribution of the traditional run-length and local binary pattern (LBP) methods, which computes the run-lengths of local binary patterns on both binarized and gray scale images. The COLD feature is the joint distribution of the relation between orientation and length of line segments obtained from writing contours in handwritten documents. Our proposed LBPruns and COLD are textural-based curvature-free features and capture the line information of handwritten texts instead of the curvature information. The combination of the LBPruns and COLD features provides a significant improvement on the CERUG data set, handwritten documents on which contain a large number of irregular-curvature strokes. The results of proposed features evaluated on other two widely used data sets (Firemaker and IAM) demonstrate promising results.

To understand the historical context of an ancient manuscript, scholars rely on the prior knowledge of writer and date of that document. In this paper, we study the Dead Sea Scrolls, a collection of ancient manuscripts with immense historical, religious, and linguistic significance, which was discovered in the mid-20th century near the Dead Sea. Most of the manuscripts of this collection have become digitally available only recently and techniques from the pattern recognition field can be applied to revise existing hypotheses on the writers and dates of these scrolls. This paper presents our ongoing work which aims to introduce digital palaeography to the field and generate fresh empirical data by means of pattern recognition and artificial intelligence. Challenges in analyzing the Dead Sea Scrolls are highlighted by a pilot experiment identifying the writers using several dedicated features. Finally, we discuss whether to use specifically-designed shape features for writer identification or to use the Deep Learning methods on a relatively limited ancient manuscript collection which is degraded over the course of time and is not labeled, as in the case of the Dead Sea Scrolls. ispartof: pages:693-702 ispartof: Proceedings of the 6th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM vol:2017-January pages:693-702 ispartof: International Conference on Pattern Recognition Applications and Mathods location:Porto, Portugal date:24 Feb - 26 Feb 2017 status: published

The use of machine learning and computer vision methods for recognizing different plants from images has attracted lots of attention from the community. This paper aims at comparing local feature descriptors and bags of visual words with different classifiers to deep convolutional neural networks (CNNs) on three plant datasets; AgrilPlant, LeafSnap, and Folio. To achieve this, we study the use of both scratch and fine-tuned versions of the GoogleNet and the AlexNet architectures and compare them to a local feature descriptor with k-nearest neighbors and the bag of visual words with the histogram of oriented gradients combined with either support vector machines and multi-layer perceptrons. The results shows that the deep CNN methods outperform the hand-crafted features. The CNN techniques can also learn well on a relatively small dataset, Folio.

Text localization is a main portal to character recognition in scene images. The detection of text regions in an image is a great challenge. However, many locating methods use a bottom-up scheme that consumes relatively high computation to identify the text regions. Therefore, this paper presents a novel text localization in scene images using a top-down scheme, with a faster computation. While benchmark datasets are usually based on segmented words, there are also script types where character strings are continuous, such as Thai. Models of general text-chunk and non-text have been produced using the SIFT algorithm. Furthermore, we present a codebook-based classifier, which significantly enhances the detection performance. The proposed method has been evaluated for the text-detection and classifying processes with the TSIB (continuous-word language). The results show that our proposed technique improves the text localization particularly for a continuous-word script (Thai) which is rich in ornaments. The proposed technique is also beneficial for locating the text of the ICDAR dataset even though there are many text lines in an image.

Estimating the date of undated medieval manuscripts by evaluating the script they contain, using document image analysis, is helpful for scholars of various disciplines studying the Middle Ages. However, there are, as yet, no systems to automatically and effectively infer the age of historical scripts using machine learning methods. To build a system to date medieval documents is a challenging problem in several aspects: 1) As yet, no suitable reference dataset of medieval handwriting exists; 2) relatively little is known about the evolution of writing styles in the Middle Ages, and especially in the later Middle Ages. Our Medieval Paleographic Scale (MPS) project aims at solving these problems. We have collected a corpus of charters from the Medieval Dutch language area, dating from the period 1300 to 1550. A global and local regression method is proposed for learning and estimating the year in which these documents were written, using several features which have been successfully used in writer identification. The proposed system can serve as a basic tool for the medievalist or paleographer. The experimental results of the proposed method demonstrate its effectiveness. Keywords—Medieval Paleographic Scale, historical document dating, age estimation, global and local regression

Hidden Markov models are frequently used in handwriting-recognitionapplications. While a large number of methodological variants have beendeveloped to accommodate different use cases, the core concepts have notbeen changed much. In this paper, we develop a number of datasets tobenchmark our own implementation as well as various other tool kits. Weintroduce a gradual scale of difficulty that allows comparison of datasetsin terms of separability of classes. Two experiments are performed toreview the basic HMM functions, especially aimed at evaluating the role ofthe transition probability matrix. We found that the transition matrix maybe far less important than the observation probabilities. Furthermore, thetraditional training methods are not always able to find the proper (true)topology of the transition matrix. These findings support the view thatthe quality of the features may require more attention than the aspect oftemporal modelling addressed by HMMs.

Although most of the previous studies in writer identification in music scores assumed successful prior staff-line removal, this assumption does not hold when the music scores suffer from a certain level of degradation or deformation. The impact of staff-line removal on the result of writer identification in such documents is rather vague. In this study, we propose a novel writer identification method that requires no staff-line removal and no segmentation. Staff-line removal is virtually achieved without image processing, by dimensionality reduction with an autoencoder in Contour-Hinge feature space. The experimental result with a wide range of music manuscripts shows the proposed method can achieve favourable results without prior staff-line removal.

Feature extraction techniques can be important in character recognition, because they can enhance the efficacy of recognition in comparison to featureless or pixel-based approaches. This study aims to investigate the novel feature extraction technique called the hotspot technique in order to use it for representing handwritten characters and digits. In the hotspot technique, the distance values between the closest black pixels and the hotspots in each direction are used as representation for a character. The hotspot technique is applied to three data sets including Thai handwritten characters (65 classes), Bangla numeric (10 classes), and MNIST (10 classes). The hotspot technique consists of two parameters including the number of hotspots and the number of chain code directions. The data sets are then classified by the k-Nearest Neighbors algorithm using the Euclidean distance as function for computing distances between data points. In this study, the classification rates obtained from the hotspot, mark direction, and direction of chain code techniques are compared. The results revealed that the hotspot technique provides the largest average classification rates.

A method of time-series analysis and a time-beating experiment were used to test the structural and perceptual validity of notated metre. Autocorrelation applied to the flow of melodic intervals between notes from thirty fragments of compositions for solo instruments by J S Bach strongly supported the validity of bar length specifications. Time-beating data, obtained with four stimuli from the same set, played in an expressionless mode, and presented under categorically distinct tempos to different subgroups of musically trained subjects, were rather inconsistent with respect to tapped bar lengths. However, taps were most frequently given to the events in the stimuli that corresponded with the first beats according to the score notations. No significant effects of tempo on tapping patterns were observed. The findings are discussed in comparison with other examinations of metre inference from musical compositions.

It is proposed that the speed of aiming movements is the optimized outcome of a stochastic, oscillatory recruitment signal to the muscles and filtering properties of the effector limb. The filtering characteristic of the limb is seen to be modulated through a stiffness parameter, to be set by the subject, and directly related to the perceived accuracy demand. This filtering is assumed to be a necessary condition to delimit the noisy character of the recruitment signal in accordance with the accuracy demands of the task. Because an increase of limb stiffness, at the same time, decreases the gain of the transfer function of the limb system, Fitts' law may be explained as a decreasing net velocity of the limb with increased accuracy demands that is due to increased filtering in the effector limb. In an experimental aiming task, subjects drew line trajectories with a ballpoint pen on a digitizer tablet to goals of varying width and distance. To estimate the effective noise that was characteristic for each level of accuracy, Power Spectral Density Analysis was applied to the deviation scores of the acceleration profile for each single movement, as compared to the average Power Spectral Density Function (PSDF) for that movement condition. Temporal resolution of the apparatus allowed for analysis of periodic fluctuations of the acceleration signal in spectral bands up to 49 Hz. Increase of movement time with movement difficulty was significantly related to an inhibition of the lower frequencies of the PSDF together with a relative increase of energy in the higher bands of the spectrum. This outcome is analogous to the results of a mathematical simulation of single-joint movements of the human arm in which the effect of increased limb stiffness on the PSDFs was calculated by the algorithm. The results support the theory that limb stiffness is a more probable candidate to modulate Fitts' speed-accuracy trade-off than control processes based on the processing of feedback.

A important problem in many areas of complex information processing is to integrate heterogeneous information from different sources. For handwriting recognition we accomplished such an integration through an agent architecture of three cooperating agents. One agent represents the bottom-up knowledge as generated by a shape-based classifier of unistrokes. The second agent contains top-down (syntactic) knowledge about the language to be interpreted. The third agent is the user who produces written symbol shapes and who can be queried by the other agents. The chosen application paradigm is the real-time input and execution of scheme code. Although this is not a practical domain, it contains all the essential elements of a full text-entry application. Notable improvements can be gained over a pure bottom-up approach to pen-based computing.

Communication between humans uses many modalities. We communicate not only via verbal language, but also through our use of intonation, gaze, hand gestures, body gestures, and facial expressions. Using these modalities, we can add, modify, and substitute information in spoken conversations. Complementary use of several modalities in human-to-human communication ensures high accuracy, and only few communication problems occur. When communication problems do occur, conversation partners can easily recover, using the redundancy and complementarity of modalities. The goal of research on multimodal systems is to investigate how human—computer interaction can benefit from multiple modalities in similar ways.