Showmik Bhowmik

Handwritten Bangla Word Recognition Using HOG Descriptor

The holistic approaches for handwritten word recognition treat the words as single, indivisible entity and attempt to recognize words from their overall shape. In the present work, a novel technique to recognize handwritten Bangla word is proposed. Histograms of Oriented Gradients (HOG) are used as the feature set to represent each word sample at the feature space and a neural network based classifier is applied to classify the word images. On the basis of the HOG feature set, the performance achieved by the technique on a small dataset is quite satisfactory.

Handwritten Bangla Word Recognition Using Elliptical Features

In the present work, a holistic word recognition technique is proposed for the recognition of the handwritten Bangla words. Holistic word recognition technique assumes a word as a single and indivisible entity and extracts features from the entire word to recognize it. In this work, a set of elliptical features is extracted from handwritten word images to represent them in the feature space. Then, a comparison among 5 well known classifiers is carried out in terms of their accuracies to select the suitable classifier for evaluating the present work. Based on that, finally, a neural network based classifier is chosen for the recognition task. Using the elliptical features, the proposed system provides a satisfactory result on a small dataset.

Memetic Algorithm Based Feature Selection for Handwritten City Name Recognition

Feature selection plays a key role to reduce the high-dimensionality of feature space in machine learning applications by discarding irrelevant and redundant features with the aim of obtaining a subset of features that accurately describe a given problem with a minimum or no degradation of performance. In this paper, a Memetic Algorithm (MA) based Wrapper-filter feature selection framework is proposed for the recognition of handwritten Bangla city names. For evaluating the MA framework, a recently published feature extraction technique, reported in [1], is used for the said pattern recognition problem. Experimentation is conducted on an in-house dataset of 6000 words written in Bangla script. Here, 40 most popular city names of West Bengal, a state in India, have been considered to prepare the dataset. Proposed technique not only reduces the feature dimension, but also enhances the performance of the word recognition technique significantly.

Bangla Handwritten City Name Recognition Using Gradient-Based Feature

In recent times, holistic word recognition has achieved enormous attention from the researchers due to its segmentation-free approach. In the present work, a holistic word recognition method is presented for the recognition of handwritten city names in Bangla script. At first, each word image is hypothetically segmented into equal number of grids. Then gradient-based features, inspired by Histogram of Oriented Gradients (HOG) feature descriptor, are extracted from each of the grids. For the selection of suitable classifier, five well-known classifiers are compared in terms of their recognition accuracies and finally the classifier Sequential Minimal Optimization (SMO) is chosen. The system has achieved 90.65% accuracy on 10,000 samples comprising of 20 most popular city names of West Bengal, a state of India.

Text and Non-text Separation in Handwritten Document Images Using Local Binary Pattern Operator

Development of an automated system for handwritten document analysis is being considered as an important research topic since last few decades. Digitized documents, either handwritten or printed, contain a mixture of text and non-text elements which need to be separated for designing a document layout analyzer or even an Optical Character Recognizer. In this paper, a technique is described to separate the text objects from the non-text objects present in a handwritten document image. For this purpose, a Rotation Invariant Local Binary Pattern (RILBP) based texture feature is used to represent the said components, at the feature space. Finally, the classification is carried out using an Artificial Neural Network based classifier called, Multi-layer Perceptron (MLP). The system provides an impressive result on a database comprising of 100 handwritten document images.

Word-Level Script Identification from Handwritten Multi-script Documents

In this paper, a robust word-level handwritten script identification technique has been proposed. A combination of shape based and texture based features are used to identify the script of the handwritten word images written in any of five scripts namely, Bangla, Devnagari, Malayalam, Telugu and Roman. An 87-element feature set is designed to evaluate the present script recognition technique. The technique has been tested on 3000 handwritten words in which each script contributes about 600 words. Based on the identification accuracies of multiple classifiers, Multi Layer Perceptron (MLP) has been chosen as the best classifier for the present work. For 5-fold cross validation and epoch size of 500, MLP classifier produces the best recognition accuracy of 91.79% which is quite impressive considering the shape variations of the said scripts.

Feature Selection for Handwritten Word Recognition Using Memetic Algorithm

Nowadays, feature selection is considered as a de facto standard in the field of pattern recognition where high-dimensional feature attributes are used. The main purpose of any feature selection algorithm is to reduce the dimensionality of the input feature vector while improving the classification ability. Here, a Memetic Algorithm (MA)-based wrapper–filter feature selection method is applied for the recognition of handwritten word images in segmentation-free approach. In this context, two state-of-the-art feature vectors describing texture and shape of the word images, respectively, are considered for feature dimension reduction. Experimentation is conducted on handwritten Bangla word samples comprising 50 popular city names of West Bengal, a state of India. Final results confirm that for the said recognition problem, subset of features selected by MA produces increased recognition accuracy than the individual feature vector or their combination when applied entirely.

Off-line Bangla handwritten word recognition: a holistic approach

Due to the cursive nature, segmentation of handwritten Bangla words into characters and also recognition of the same sometimes become a very challenging problem to the researchers. Presence of comparatively large character set along with modifiers, ascendants, descendants, and compound characters makes the segmentation task more complex. As holistic method avoids such character-level segmentation, it is generally useful for the recognition of words written in any such complex scripts. In the present work, a holistic handwritten word recognition method is developed using a feature descriptor, designed by combining different Elliptical, Tetragonal and Vertical pixel density histogram-based features. Recognition process is carried out separately using two classifiers, namely multi-layer perceptron (MLP) and support vector machine (SVM). For evaluation of the proposed method, a database of 18,000 handwritten Bangla word images, having 120 word classes, is prepared. The proposed system performs comparatively better with SVM than MLP for the prepared dataset. It has achieved 83.64% accuracy at best case and 79.38% accuracy on an average using fivefold cross-validation. The current method has also outperformed some recently reported holistic word recognition technique tested on the developed dataset. In addition to that the database, prepared in this work, is made freely available to fill the absence of a publicly available standard database for holistic Bangla word recognition.

Text/Non-Text Separation from Handwritten Document Images Using LBP Based Features: An Empirical Study

Isolating non-text components from the text components present in handwritten document images is an important but less explored research area. Addressing this issue, in this paper, we have presented an empirical study on the applicability of various Local Binary Pattern (LBP) based texture features for this problem. This paper also proposes a minor modification in one of the variants of the LBP operator to achieve better performance in the text/non-text classification problem. The feature descriptors are then evaluated on a database, made up of images from 104 handwritten laboratory copies and class notes of various engineering and science branches, using five well-known classifiers. Classification results reflect the effectiveness of LBP-based feature descriptors in text/non-text separation.

A Two-Stage Approach for Text and Non-text Separation from Handwritten Scientific Document Images

The presence of non-text components in the document image hinders the result of an optical character recognition (OCR)-based document analysis system. Thus, text and non-text separation has become an essential task in the domain of document image processing. To address this issue, in the present work, a simple two-stage method is developed to separate the text and the non-text components from the images of handwritten scientific documents. Before starting the actual process, connected components from the document pages are extracted. Then, in the first stage, some commonly occurred components are identified and separated out as graphics. In the second stage, remaining components are passed through feature extraction and subsequent classification processes. Evaluating the system on handwritten scientific document images, it is found that 87.16% components are classified correctly as text or non-text.