Abstract

Conventional optical character recognition (OCR) systems operate on individual characters and words, and do not normally exploit document or collection context. We describe a Collection OCR which takes advantage of the fact that multiple examples of the same word (often in the same font) may occur in a document or collection. The idea here is that an OCR or a reCAPTCHA like process generates a partial set of recognized words. In the second stage, a nearest neighbor algorithm compares the remaining word-images to those already recognized and propagates labels from the nearest neighbors. It is shown that by using an approximate fast nearest neighbor algorithm based on Hierarchical K-Means (HKM), we can do this accurately and efficiently. It is also shown that profile based features perform much better than SIFT and Pyramid Histogram of Gradient (PHOG) features. We believe that this is because profile features are more robust to word degradations (common in our documents). This approach is applied to a collection of Telugu books - a language for which no commercial OCR exists. We show from a selection of 33 Telugu books that starting with OCR labels for only 30% of the collection we can recognize the remaining 70% of the words in the collection with 70% accuracy using this approach. Since the approach makes no language specific assumptions, it should be applicable to a large number of languages. In particular we are interested in its applicability to Indic languages and scripts.

Abstract

Most of the Indian scripts do not have any robust commercial OCRs. Many of the laboratory prototypes report reasonable results at recognition/classification stage. However, word level accuracies are still poor. It is well known that word accuracy decreases as the number of characters in a word increase. For Malayalam, the average number of characters in a word is almost twice that of English. Moreover, the number of words required to cover 80% of the Malayalam language is more than forty times that of other Indian languages such as Hindi. Hence a direct dictionary based post-processing scheme is not suitable for Malayalam. In this paper, we propose a post-processing scheme which uses statistical language models at the sub-character level to boost word level recognition results. We use a multi-stage graph representation and formulate the recognition task as an optimization problem. Edges of the graph encode the language information and nodes represent the visual similarities. An optimal path from source node to destination node represents the recognized text. We validate our method on more than 10,000 words from a Malayalam corpus.

Abstract

Kernel methods yield state-of-the-art performance in certain applications such as image classification and object detection. However, large scale problems require machine learning techniques of at most linear complexity and these are usually limited to linear kernels. This unfortunately rules out gold-standard kernels such as the generalized RBF kernels (e.g. exponential-χ 2 ). Recently, Maji and Berg [13] and Vedaldi and Zisserman [20] proposed explicit feature maps to approximate the additive kernels (intersection, χ 2 , etc.) by linear ones, thus enabling the use of fast machine learning technique in a non-linear context. An analogous technique was proposed by Rahimi and Recht [14] for the translation invariant RBF kernels. In this paper, we complete the construction and combine the two techniques to obtain explicit feature maps for the generalized RBF kernels. Furthermore, we investigate a learning method using l 1 regularization to encourage sparsity in the final vector representation, and thus reduce its dimension. We evaluate this technique on the VOC 2007 detection challenge, showing when it can improve on fast additive kernels, and the trade-offs in complexity and accuracy

Abstract

An important pre-requisite for many tasks like Visual Servoing and visual SLAM is the task of tracking the underlying features. The use of planar features for these purposes has gained importance recently. Complementing current planar tracking works in the robotics literature, which use multiple features, we formulate the tracking problem using multiple planes. Inspired by the maturity in understanding of geometric quantities like the homography in computer vision, we develop a system based on the Unscented Kalman Filter (UKF) that localizes the camera and estimates the plane parameters of a scene, using homographies as measurement. Homographies are estimated using tracked feature points. We show that this framework provides significant robustness and stability to the system under significant changes of illumination, occlusion etc. Finally, we also propose a Convex optimization based solution for the initialization of this system, which is capable of producing globally optimal estimates, and is a useful algorithm in its own right. Several synthetic and real results are presented to demonstrate the efficacy of our approach.

Abstract

In order to perform speech summmarization it is necessary to identify important segments in speech signal. The importance of a speech segment can be effectively determined by using infomation from lexical and prosodic features. Standard speech summarization systems depend on ASR transcripts or gold standard human reference summaries to train a supervised system which combines lexical and prosodic features to choose a segment into summary. We propose a method which uses prominence values of syllables in a speech segment to rank the segment for summarization. The proposed method does not depend on ASR transcripts or gold standard human summaries. Evaluation results showed that summaries generated by the proposed method are as good as the summaries generated using tf* idf scores and supervised system trained on gold standard summaries.

Abstract

The problem of extracting bilingual dictionaries from Wikipedia is well known and well researched. Given the structural and rich multilingual content of Wikipedia, a language independent approach is necessary for extracting dictionaries for various languages more so for under-resourced languages. In our attempt to mine dictionaries for under-resourced languages, we developed an iterative approach to construct parallel corpus for building a dictionary, for which we consider several kinds of Wikipedia article information like title, infobox information, category, article text and dictionaries already built at each phase. The average precision over various datasets is encouraging with maximum precision of 76.7%, performing better than existing systems. As no language-specific resources are used, our method is applicable to any pair of language with special focus on under-resourced languages and hence breaking the language barrier.

Abstract

The problem of measuring semantic relatedness between social tags remains largely open. Given the structure of social bookmarking systems, similarity measures need to be addressed from a social bookmarking systems perspective. We address the fundamental problem of weight model for tags over which every similarity measure is based. We propose a weight model for tagging systems that considers the user dimension unlike existing measures based on tag frequency. Visual analysis of tag clouds depicts that the proposed model provides intuitively better scores for weights than tag frequency. We also propose weighted similarity model that is conceptually different from the contemporary frequency based similarity measures. Based on the weighted similarity model, we present weighted variations of several existing measures like Dice and Cosine similarity measures. We evaluate the proposed similarity model using Spearman's correlation coefficient, with WordNet as the gold standard. Our method achieves 20% improvement over the traditional similarity measures like dice and cosine similarity and also over the most recent tag similarity measures like mutual information with distributional aggregation. Finally, we show the practical effectiveness of the proposed weighted similarity measures by performing search over tagged documents using Social SimRank over a large real world dataset.

Abstract

Analysis of human reference summaries of broadcast news showed that humans give preference to anchor speaker segments while constructing a summary. Therefore, we exploit the role of anchor speaker in a news show by tracking his/her speech to construct indicative/informative extractive audio summaries. Speaker tracking is done by Bayesian information criterion (BIC) technique. The proposed technique does not require Automatic Speech Recognition (ASR) transcripts or human reference summaries for training. The objective evaluation by ROUGE showed that summaries generated by the proposed technique are as good as summaries generated by a baseline text summarization system taking manual transcripts as input and summaries generated by a supervised speech summarization system trained using human summaries. The subjective evaluation of audio summaries by humans showed that they prefer summaries generated by proposed technique to summaries generated by supervised speech summarization system.

Abstract

The present invention provides a method for manufacture of supported noble metal based alloy catalysts with a high degree of alloying and a small crystallite size. The method is based on the use of polyol solvents as reaction medium and comprises of a two-step reduction process in the presence of a support material. In the first step, the first metal (M1= transition metal; eg Co, Cr, Ru) is activated by increasing the reaction temperature to 80 to 160 C. In the second step, the second metal (M2= noble metal; eg Pt, Pd, Au and mixtures thereof) is added and the slurry is heated to the boiling point of the polyol solvent in a range of 160 to 300 C. Due to this two-step method, an uniform reduction occurs, resulting in noble metal based catalysts with a high degree of alloying and a small crystallite size of less than 3 nm. Due to the high degree of alloying, the lattice constants are lowered. The catalysts manufactured …

Abstract

Gold nanostars provide high sensitivity for single nanoparticle label-free biosensing. The nanostars present multiple plasmon resonances of which the lower energy ones, corresponding to the nanostar tips and core−tip interactions, are the most sensitive to environmental changes. Streptavidin molecules are detected upon binding to individual, biotin-modified gold nanostars by spectral shifts in the plasmon resonances. Concentrations as low as 0.1 nM produce a shift of the tip related plasmon resonances of about 2.3 nm (5.3 meV).