Sangheeta Roy

Bayesian classifier for multi-oriented video text recognition system

A new system for recognizing text in video through binarization.Exploring wavelet and gradient sub-bands for enhancing text information.Introducing Bayesian classifier in novel way for binarization.Use of connected component analysis for restoring missing information.Experimental results on both video and scene text shows the method is superior. Developing an automatic system for recognizing video texts such as signboards, street names, room numbers, building names and hotels names is challenging due to low resolution, complex background, font or font size variations, and multiple orientations of texts. In this paper, we develop a new system to recognize video texts through binarization by introducing a Bayesian classifier. We explore wavelet decomposition and gradient sub-bands to enhance text information in video. The enhanced information is used in different ways to calculate the requirement of Bayesian classifier, such as a priori probability and conditional probabilities of text pixels to estimate the posterior probability automatically, which results in text components. Connected component analysis is then applied to restore missing text information before sending it to an OCR engine if any disconnection exists in the text components. Experimental results on video data, the benchmark ICDAR scene character data (camera images) and arbitrary orientation data (camera images) show that the proposed method outperforms existing baseline methods in terms of recognition rates at both character and pixel levels.

A Novel Approach of Bangla Handwritten Text Recognition Using HMM

This paper presents a novel approach for offline Bangla (Bengali) handwritten word recognition by Hidden Markov Model (HMM). Due to the presence of complex features such as headline, vowels, modifiers, etc., character segmentation in Bangla script is not easy. Also, the position of vowels and compound characters make the segmentation task of words into characters very complex. To take care of these problems we propose a novel method considering a zone-wise break up of words and next perform HMM based recognition. In particular, the word image is segmented into 3 zones, upper, middle and lower, respectively. The components in middle zone are modeled using HMM. By this zone segmentation approach we reduce the number of distinct component classes compared to total number of classes in Bangla character set. Once the middle zone portion is recognized, HMM based forced alignment is applied in this zone to mark the boundaries of individual components. The segmentation paths are extended later to other zones. Next, the residue components, if any, in upper and lower zones in their respective boundary are combined to achieve the final word level recognition. We have performed a preliminary experiment on a dataset of 10,120 Bangla handwritten words and found that the proposed approach outperforms the custom way of HMM based recognition.

HMM-Based Multi Oriented Text Recognition in Natural Scene Image

Recognition of curved text in natural scene image is a challenging task. Due to complex background and unpredictable characteristics of scene text and noise, text characters in strings are often touching that affects the performance of segmentation and recognition. This paper presents a novel approach for curved text recognition using Hidden Markov Models (HMM). From curved text, a path of sliding window is estimated and features extracted from the sliding window are fed to the HMM system for recognition. We evaluate two frame-wise feature extraction algorithms namely Marti-Bunk and local gradient histogram. The proposed approach has been tested on different natural scene benchmark as well as video databases, e.g. ICDAR-2003competition scene images, MSRA-TD500 and NUS. We have achieved word recognition accuracy of about 63.28%, 58.41% and 53.62%y for horizontal text, non-horizontal text and curved text, respectively.

Fractional poisson enhancement model for text detection and recognition in video frames

Performing Laplacian operation on video images is a common technique to improve image contrast to achieve good text detection and recognition accuracies. However, it is a fact that when Laplacian operation enhances contrast, at the same time it introduces too many noises. To alleviate this, the existing methods propose different enhancement methods and filters. In this paper, we propose a generalized enhancement model based on fractional calculus to increase the quality of images obtained by Laplacian operation. The proposed method considers edges and their neighbor information to derive a mathematical model for enhancing low contrast information in video as well as in scene images. Experimental results of text detection and recognition methods on different databases show that the proposed enhancement model improves their accuracies significantly. The enhancement model is compared with standard enhancement models to show that the proposed model outperforms the existing models in terms of quality measures. The usefulness of the proposed model is validated through text detection and recognition experiments. A new Fractional Poisson enhancement model for noise removal in video.The model can be used for reducing distortion effects in text detection and recognition.The model enhances the results of existing text detection and recognition methods.

Word recognition in natural scene and video images using Hidden Markov Model

Text recognition from a natural scene and video is challenging compared to that in scanned document images. This is due to the problems of text on different sources of various styles, font variation, font size variations, background variations, etc. There are approaches for word segmentation from video and scene images to feed the word image into OCRs. Nevertheless, such methods often fail to yield satisfactory results in recognition. Therefore, in this paper, we propose to combine Hidden Markov Model (HMM) and Convolutional Neural Network (CNN) to achieve good recognition rate. Sequential gradient features with HMM help to find character alignment of a word. Later the character alignments are verified by Convolutional Neural network (CNN). The approach is tested on both video and scene data to show the effectiveness of the proposed approach. The results are found encouraging.

A New Multi-modal Technique for Bib Number/Text Detection in Natural Images

The detection and recognition of racing bib number/text, which is printed on paper, cardboard tag, or t-shirt in natural images in marathon, race and sports, is challenging due to person movement, non-rigid surface, distortion by non-illumination, severe occlusions, orientation variations etc. In this paper, we present a multi-modal technique that combines both biometric and textual features to achieve good results for bib number/text detection. We explore face and skin features in a new way for identifying text candidate regions from input natural images. For each text candidate region, we propose to use text detection and recognition methods for detecting and recognizing bib numbers/texts, respectively. To validate the usefulness of the proposed multi-modal technique, we conduct text detection and recognition experiments before text candidate region detection and after text candidate region detection in terms of recall, precision and f-measure. Experimental results show that the proposed multi-modal technique outperforms the existing bib number detection method.

Human localization at home using kinect

In this paper authors have presented a method to localize and detect human being from Kinect captured sequence of images. The proposed method takes a sequence of gray (G) scale image and the corresponding depth (D) image as input. The gray scale image and the depth information are captured using two different sensors within the same device, Kinect and the processing are executed in the processor attached with Kinect. The proposed method localizes the human by using their motion along x, y direction and then considers all pixels connected with those pixels and over a 3D plane to accomplish the segmentation with an accuracy of 77%. Experimental results demonstrate that our method is robust against existing method for human localization.

New texture-spatial features for keyword spotting in video images

Keyword spotting in video document images is challenging due to low resolution and complex background of video images. We propose the combination of Texture-Spatial-Features (TSF) for keyword spotting in video images without recognizing them. First, a segmentation method extracts words from text lines in each video image. Then we propose the set of texture features for identifying text candidates in the word image with the help of k-means clustering. The proposed method finds proximity between text candidates to study the spatial arrangement of pixels that result in feature vectors for spotting words in the input frame. The proposed method is evaluated on word images of different fonts, contrasts, backgrounds and font sizes, which are chosen from standard databases such as ICDAR 2013 video and our video data. Experimental results show that the proposed method outperforms the existing method in terms of recall, precision and f-measure.

View-Invariant Human Detection from RGB-D Data of Kinect Using Continuous Hidden Markov Model

In this paper authors have presented a method to detect human from a Kinect captured Gray-Depth (G-D) using Continuous Hidden Markov models (C-HMMs). In our proposed approach, we initially generate multiple gray scale images from a single gray scale image/ video frame based on their depth connectivity. Thus, we initially segment the G image using depth information and then relevant components were extracted. These components were further filtered out and features were extracted from the candidate components only. Here a robust feature named Local gradients histogram(LGH) is used to detect human from G-D video. We have evaluated our system against the data set published by LIRIS in ICPR 2012 and on our own data set captured in our lab. We have observed that our proposed method can detect human from this data-set with a 94.25% accuracy.

Multi-oriented Text Recognition in Graphical Documents Using HMM

The text lines in graphical documents (e.g., maps, engineering drawings), artistic documents etc., are often annotated in curve lines to illustrate different locations or symbols. For the optical character recognition of such documents, individual text lines from the documents need to be extracted and recognized. Due to presence of multi-oriented characters in such non-structured layout, word recognition is a challenging task. In this paper, we present an approach towards the recognition of scale and orientation invariant text words in graphical documents using Hidden Markov Models (HMM). First, a line extraction method is applied to segment text lines and the method is based on the foreground and background information of the text components. To effectively utilize the background information, a water reservoir concept is used here. For recognition of curved text lines, a path of sliding window is estimated and features extracted from the sliding window are fed to the HMM system for recognition. Local gradient histogram (LGH) based frame-wise feature is used in HMM. The experimental results are evaluated on a dataset of graphical words and we have obtained encouraging results.