Showing papers by "Michael Blumenstein published in 2012"

A New Method for Arbitrarily-Oriented Text Detection in Video

Abstract: Text detection in video frames plays a vital role in enhancing the performance of information extraction systems because the text in video frames helps in indexing and retrieving video efficiently and accurately. This paper presents a new method for arbitrarily-oriented text detection in video, based on dominant text pixel selection, text representatives and region growing. The method uses gradient pixel direction and magnitude corresponding to Sobel edge pixels of the input frame to obtain dominant text pixels. Edge components in the Sobel edge map corresponding to dominant text pixels are then extracted and we call them text representatives. We eliminate broken segments of each text representatives to get candidate text representatives. Then the perimeter of candidate text representatives grows along the text direction in the Sobel edge map to group the neighboring text components which we call word patches. The word patches are used for finding the direction of text lines and then the word patches are expanded in the same direction in the Sobel edge map to group the neighboring word patches and to restore missing text information. This results in extraction of arbitrarily-oriented text from the video frame. To evaluate the method, we considered arbitrarily-oriented data, non-horizontal data, horizontal data, Hua's data and ICDAR-2003 competition data (Camera images). The experimental results show that the proposed method outperforms the existing method in terms of recall and f-measure.

Recent Advances in Video Based Document Processing: A Review

Abstract: Extraction and recognition of text present in video has become a very popular research area in the last decade. Generally, text present in video frames is of different size, orientation, style, etc. with complex backgrounds, noise, low resolution and contrast. These factors make the automatic text extraction and recognition in video frames a challenging task. A large number of techniques have been proposed by various researchers in the recent past to address the problem. This paper presents a review of various state-of-the-art techniques proposed towards different stages (e.g. detection, localization, extraction, etc.) of text information processing in video frames. Looking at the growing popularity and the recent developments in the processing of text in video frames, this review imparts details of current trends and potential directions for further research activities to assist researchers.

Multi-script off-line signature identification

Abstract: In this paper, we present an empirical contribution towards the understanding of multi-script signature identification. In the proposed signature identification system, the signatures of Bengali (Bangla), Hindi (Devanagari) and English are considered for the identification process. This system will identify whether a claimed signature belongs to the group of Bengali, Hindi or English signatures. Zernike Moment and histogram of gradient are employed as two different feature extraction techniques. In the proposed system, Support Vector Machines (SVMs) are considered as classifiers for signature identification. A database of 2100 Bangla signatures, 2100 Hindi signatures and 2100 English signatures are used for experimentation. Two different results based on two different feature sets are calculated and analysed. The highest accuracy of 92.14% is obtained based on the gradient features using 4200 (1400 Bangla +1400 Hindi + 1400 English) samples for training and 2100 (700 Bangla +700 Hindi +700 English) samples for testing.

Off-line signature verification using G-SURF

Abstract: In the field of biometric authentication, automatic signature identification and verification has been a strong research area because of the social and legal acceptance and extensive use of the written signature as an easy method for authentication. Signature verification is a process in which the questioned signature is examined in detail in order to determine whether it belongs to the claimed person or not. Signatures provide a secure means for confirmation and authorization in legal documents. So nowadays, signature identification and verification becomes an essential component in automating the rapid processing of documents containing embedded signatures. Sometimes, part-based signature verification can be useful when a questioned signature has lost its original shape due to inferior scanning quality. In order to address the above-mentioned adverse scenario, we propose a new feature encoding technique. This feature encoding is based on the amalgamation of Gabor filter-based features with SURF features (G-SURF). Features generated from a signature are applied to a Support Vector Machine (SVM) classifier. For experimentation, 1500 (50×30) forgeries and 1200 (50×24) genuine signatures from the GPDS signature database were used. A verification accuracy of 97.05% was obtained from the experiments.

Off-line English and Chinese signature identification using foreground and background features

Abstract: In the field of information security, the usage of biometrics is growing for user authentication. Automatic signature recognition and verification is one of the biometric techniques, which is only one of several used to verify the identity of individuals. In this paper, a foreground and background based technique is proposed for identification of scripts from bi-lingual (English/Roman and Chinese) off-line signatures. This system will identify whether a claimed signature belongs to the group of English signatures or Chinese signatures. The identification of signatures based on its script is a major contribution for multi-script signature verification. Two background information extraction techniques are used to produce the background components of the signature images. Gradient-based method was used to extract the features of the foreground as well as background components. Zernike Moment feature was also employed on signature samples. Support Vector Machine (SVM) is used as the classifier for signature identification in the proposed system. A database of 1120 (640 English+480 Chinese) signature samples were used for training and 560 (320 English+240 Chinese) signature samples were used for testing the proposed system. An encouraging identification accuracy of 97.70% was obtained using gradient feature from the experiment.

Off-Line Bangla Signature Verification

Abstract: In the field of information security, biometric systems play an important role. Within biometrics, automatic signature identification and verification has been a strong research area because of the social and legal acceptance and extensive use of the written signature as an individual authentication. Signature verification is a process in which the questioned signature is examined in detail in order to determine whether it belongs to the claimed person or not. Despite substantial research in the field of signature verification involving Western signatures, very few works have been dedicated to non-Western signatures such as Chinese, Japanese, Arabic, or Persian etc. In this paper, the performance of an off-line signature verification system involving Bangla signatures, whose style is distinct from Western scripts, was investigated. The Gaussian Grid feature extraction technique was employed for feature extraction and Support Vector Machines (SVMs) were considered for classification. The Bangla signature database employed in the experiments consisted of 3000 forgeries and 2400 genuine signatures. An encouraging accuracy of 90.4% was obtained from the experiments.

A New Method for Word Segmentation from Arbitrarily-Oriented Video Text Lines

Abstract: Word segmentation has become a research topic to improve OCR accuracy for video text recognition, because a video text line suffers from arbitrary orientation, complex background and low resolution. Therefore, for word segmentation from arbitrarily-oriented video text lines, in this paper, we extract four new gradient directional features for each Canny edge pixel of the input text line image to produce four respective pixel candidate images. The union of four pixel candidate images is performed to obtain a text candidate image. The sequence of the components in the text candidate image according to the text line is determined using nearest neighbor criteria. Then we propose a two-stage method for segmenting words. In the first stage, for the distances between the components, we apply K-means clustering with K=2 to get probable word and non-word spacing clusters. The words are segmented based on probable word spacing and all other components are passed to the second stage for segmenting correct words. For each segmented and un-segmented words passed to the second stage, the method repeats all the steps until the K-means clustering step to find probable word and non-word spacing clusters. Then the method considers cluster nature, height and width of the components to identify the correct word spacing. The method is tested extensively on video curved text lines, non-horizontal straight lines, horizontal straight lines and text lines from the ICDAR-2003 competition data. Experimental results and a comparative study shows the results are encouraging and promising.

Hindi Off-Line Signature Verification

Abstract: Handwritten Signatures are one of the widely used biometrics for document authentication as well as human authorization. The purpose of this paper is to present an off-line signature verification system involving Hindi signatures. Signature verification is a process by which the questioned signature is examined in detail in order to determine whether it belongs to the claimed person or not. Despite of substantial research in the field of signature verification involving Western signatures, very little attention has been dedicated to non-Western signatures such as Chinese, Japanese, Arabic, Persian etc. In this paper, the performance of an off-line signature verification system involving Hindi signatures, whose style is distinct from Western scripts, has been investigated. The gradient and Zernike moment features were employed and Support Vector Machines (SVMs) were considered for verification. To the best of the authors' knowledge, Hindi signatures have never been used for the task of signature verification and this is the first report of using Hindi signatures in this area. The Hindi signature database employed for experimentation consisted of 840 (35x24) genuine signatures and 1050 (35x30) forgeries. An encouraging accuracy of 7.42% FRR and 4.28% FAR were obtained following experimentation when the gradient features were employed.

Concept Learning for EL++ by Refinement and Reinforcement

Abstract: Ontology construction in OWL is an important and yet time-consuming task even for knowledge engineers and thus a (semi-) automatic approach will greatly assist in constructing ontologies. In this paper, we propose a novel approach to learning concept definitions in EL++ from a collection of assertions. Our approach is based on both refinement operator in inductive logic programming and reinforcement learning algorithm. The use of reinforcement learning significantly reduces the search space of candidate concepts. Besides, we present an experimental evaluation of constructing a family ontology. The results show that our approach is competitive with an existing learning system for EL.

Concept learning for EL ++ by refinement and reinforcement

Abstract: Ontology construction in OWL is an important and yet time-consuming task even for knowledge engineers and thus a (semi-) automatic approach will greatly assist in constructing ontologies. In this paper, we propose a novel approach to learning concept definitions in $\ensuremath{\ensuremath{\cal E}\ensuremath{\cal L}^{++}} $ from a collection of assertions. Our approach is based on both refinement operator in inductive logic programming and reinforcement learning algorithm. The use of reinforcement learning significantly reduces the search space of candidate concepts. Besides, we present an experimental evaluation of constructing a family ontology. The results show that our approach is competitive with an existing learning system for $\ensuremath{\cal E}\ensuremath{\cal L}$.

Refinement of Backward Prediction Method for Reliable Artificial Intelligence-Based Bridge Deterioration Modelling

Abstract: A deterioration model is the most critical component of a Bridge Management System (BMS). Artificial Intelligence (AI)-based bridge deterioration model has recently been developed to minimise uncertainties in predicting long-term performance of bridge structural elements. This model contains two components: (1) using Neural Network-based Backward Prediction Model (BPM) to generate unavailable historical condition ratings; and (2) using Time Delay Neural Network (TDNN) to perform long-term performance prediction of bridge structural elements. However new problems have emerged in the process of TDNN prediction. In this study, the BPM-generated condition ratings are used together with the actual overall condition ratings. The incompatibility between the two sets of data produces unreliable prediction outcomes during the TDNN process. This research therefore aims to introduce a new data processing procedure for BPM outcomes, by removing meaningless condition ratings that cause poor training outcomes for long-...

Long-Term Performance of Bridge Elements Using Integrated Deterioration Method Incorporating Elman Neural Network

Abstract: Currently, probabilistic deterioration modeling techniques have been employed in most state-of-the-art Bridge Management Systems (BMSs) to predict future bridge condition ratings. As confirmed by many researchers, the reliability of the probabilistic deterioration models rely heavily on the sufficient amount of condition data together with their well-distributed historical deterioration patterns over time. However, inspection records are usually insufficient in most bridge agencies. As a result, a typical standalone probabilistic model (e.g. state-based or time-based model) is not promising for forecasting a reliable bridge long-term performance. To minimise the shortcomings of lacking condition data, an integrated method using a combination of state- and time-based techniques has recently been developed and has demonstrated an improved performance as compared to the standalone techniques. However, certain shortcomings still remain in the integrated method which necessities further improvement. In this study, the core component of the state-based modeling is replaced by an Elman Neural Networks (ENN). The integrated method incorporated with ENN is more effective in predicting long-term bridge performance as compared to the typical deterioration modeling techniques. As part of comprehensive case studies, this paper presents the deterioration prediction of 52 bridge elements with material types of Steel (S), Timber (T) and Other (O). These elements are selected from 94 bridges (totaling 4,115 inspection records). The enhanced reliability of the proposed integrated method incorporating ENN is confirmed.

A Compact Size Feature Set for the Off-Line Signature Verification Problem

Abstract: With increasing computational power, researchers in the area of off-line signature verification have been able to investigate feature extraction techniques that produce large-dimensional feature vectors. However, a large feature vector is not necessarily associated with high performance. This paper investigates the performance of a small feature set consisting of 33 feature values. In the experiments using Support Vector Machines (SVMs), an average error rate (AER) of 16.80% was obtained together with a low false acceptance rate (FAR) for random forgeries of 0.19%. The significant reduction of the error rate was obtained when the proposed global features were employed, which demonstrates their astonishingly high discriminant power. These results suggest that the use of global features for the off-line signature verification problem is worth further investigation.

Performance Prediction of Concrete Elements in Bridge Substructures using Integrated Deterioration Method

Abstract: The typical probabilistic deterioration model cannot guarantee a reliable long-term prediction for various situations of available condition data. To minimise this limitation, this paper presents an advanced integrated method using state-/time-based model to build a reliable transition probability for prediction long-term performance of bridge elements. A selection process is developed in this method to automatically select a suitable prediction approach for a given situations of condition data. Furthermore, a Backward Prediction Model (BPM) is employed to effectively prediction the bridge performance when the inspection data are insufficient. In this study, a benchmark example-concrete element in bridge substructures is selected to demonstrate that the BPM in conjunction with time-based model can improve the reliability of long-term prediction.

Off-line restricted-set handwritten word recognition for student identification in a short answer question automated assessment system

Abstract: Handwriting recognition is one of the most intensive areas of study in the field of pattern recognition. Many applications are able to benefit from a robust off-line handwriting recognition technique. An automatic off-line assessment system and a writer identification system are two of those applications. Off-line automatic assessment systems can be an aid for teachers in the marking process; they can reduce the time consumed by the human marker. There has only been limited work undertaken in developing off-line automatic assessment systems using handwriting recognition, and none in developing student identification systems, even though such systems would clearly benefit the education sector. In order to develop a complete off-line automatic assessment system, student identification using full student names is proposed in this paper. The Gaussian Grid and Modified Direction Feature Extraction Techniques are investigated in order to develop the proposed system. The recognition rates achieved using both techniques are encouraging (up to 99.08% for the Modified Direction feature extraction technique, and up to 98.28% for the Gaussian Grid feature extraction technique.

Development of Methodology for Enhancing Visual Bridge Condition Assessment Using Image Processing Techniques

Abstract: Many bridge authorities have implemented Bridge Information Systems (BISs) or Bridge Management Systems (BMSs) to effectively manage their routine inspection information. The success of a BMS is highly dependent on the quality of bridge inspection outcomes and accurate estimation of future bridge condition ratings. To ensure such successful outcomes, a BMS must (1) contain reliable, consistent and accurate condition data from routine bridge inspections; and (2) encompass reliable deterioration modelling that overcomes the shortcomings of a lack of historical bridge inspection records. However published literature demonstrates that several limitations exist particularly in terms of inconsistency of inspection outcomes due to subjective judgment. To minimise such limitations, this paper presents a feasibility study for the enhancement of the current visual bridge inspection method using optical image processing techniques. The development work consists of image processing and knowledge-based approaches. It is anticipated that the proposed method is capable of minimising the shortcomings of subjective judgment on condition rating assessment and providing cost effective solutions to bridge agencies. Ultimately, the proposed bridge inspection methodology can provide consistent and accurate evaluation on the condition states of bridge elements. This in turn will lead to more reliable predictions of long-term bridge performance.