Chayan Halder

PHDIndic_11: page-level handwritten document image dataset of 11 official Indic scripts for script identification

Without publicly available dataset, specifically in handwritten document recognition (HDR), we cannot make a fair and/or reliable comparison between the methods. Considering HDR, Indic script’s document recognition is still in its early stage compared to others such as Roman and Arabic. In this paper, we present a page-level handwritten document image dataset (PHDIndic_11), of 11 official Indic scripts: Bangla, Devanagari, Roman, Urdu, Oriya, Gurumukhi, Gujarati, Tamil, Telugu, Malayalam and Kannada. PHDIndic_11 is composed of 1458 document text-pages written by 463 individuals from various parts of India. Further, we report the benchmark results for handwritten script identification (HSI). Beside script identification, the dataset can be effectively used in many other applications of document image analysis such as script sentence recognition/understanding, text-line segmentation, word segmentation/recognition, word spotting, handwritten and machine printed texts separation and writer identification.

Writer Identification from Handwritten Devanagari Script

This paper presents analysis of Devanagari characters for writer identification. Being originated from Brahmic script, Devanagari is the most popular script in India. It is used by over 400 million people around the world. Application of writer identification of Devanagari handwritten characters covers a vast area such as The Questioned Document Examination (QDE) is an area of the Forensic Science with the main purpose to answer questions related to questioned document (authenticity, authorship and others). Signature verification in banking, in Graphology (study of handwriting) a theory or practice for inferring a person’s character, disposition, and attitudes from their handwriting. Here we collect 5 copies of handwritten characters to nullify intra-writing variation, from 50 different people mainly students. After preprocessing and character extraction, 64-dimensional feature is computed based on gradient of the images. Some manual processing is required because some noises are too difficult to remove automatically as they are much closer to the characters. We have used LIBLINEAR and LIBSVM classifiers of WEKA environment to get the individuality of characters. We have done the writer identification with all the characters and obtained 99.12 % accuracy for LIBLINEAR with all writers. Features collected from this work can be used in the next level to identify writers from their cursive writing.

Separating Indic Scripts with matra for Effective Handwritten Script Identification in Multi-Script Documents

We present a novel approach for separating Indic scripts with ‘matra’, which is used as a precursor to advance and/or ease subsequent handwritten script identification in multi-script documents. In our study, among state-of-the-art features and classifiers, an optimized fractal geometry analysis and random forest are found to be the best performer to distinguish scripts with ‘matra’ from their counterparts. For validation, a total of 1204 document images are used, where two different scripts with ‘matra’: Bangla and Devanagari are considered as positive samples and the other two different scripts: Roman and Urdu are considered as negative samples. With this precursor, an overall script identification performance can be advanced by more than 5.13% in accuracy and 1.17 times faster in processing time as compared to conventional system.

Transform based approach for Indic script identification from handwritten document images

In a multi-script country like India script identification from document images is an essential step before choosing appropriate script specific OCR (Optical Character Recognizer). The problem of handwritten script identification is more challenging compared to printed one due to uneven variations with respect to writers, time, content etc. Increasing efforts are coming day by day from document image processing researchers to develop standard techniques for Indic script identification. But most of the works is found to be considering printed script document images. In this paper a simple, robust and segmentation free technique based on different image transform methods and statistical features to identify any one of the four popular Indic scripts namely Bangla, Roman, Devanagari and Oriya is proposed. A dataset of total 101 handwritten document images comprising of more than 11000 words and 1300 lines with almost equal distribution of each type of scripts are built, which were collected from different writers with varying age, sex and educational qualification. On experimentation, an average accuracy rate of 88.1% is found for Four-scripts combination by MLP (Multilayer Perceptron) classifier after five fold cross validation. The average Tri-Scripts and Bi-Scripts accuracy are found to be 89.7% and 94.3% respectively. The outcome of this work is really impressive considering inherent complexities of handwritten Indic scripts.

Indic script identification from handwritten document images — An unconstrained block-level approach

In a multi-script country like India, prior identification of script from document images is an essential step before choosing appropriate script specific OCR. The problem becomes more complex and challenging in case of HSI (Handwritten Script Identification). An automatic HSI technique for document images of six popular Indic scripts namely Bangla, Devanagari, Malayalam, Oriya, Roman and Urdu is proposed in this paper. A Block-level approach is followed for the same and initially 34-dimensional feature vector is constructed applying transform based (BRT, BDCT, BFFT and BDT), textural and statistical techniques. Finally using a GAS (Greedy Attribute Selection) method 20 attributes are selected for learning process. Total 600 unconstrained document image blocks of size 512×512 each, are prepared with equal distribution of each script type. The whole dataset is divided into 2:1 ratio for training and testing. Extensive experimentation is carried out for Six-scripts, Tetra-scripts, Tri-scripts and Bi-scripts combinations. Experimental result shows promising and comparable performance.

Effect of writer information on Bangla handwritten character recognition

Handwritten character recognition has various potential in the field of document image processing. It is one of the important aspects for systems like handwritten optical character recognizer, writer identification/verification, automatic document sorter etc. In Bangla only few attempts are made towards character recognition. In this current study a relatively new attempt is made towards finding the dependency of writer information on character recognition by varying the inputs. This study will provide a better understanding of the input data for character recognition. Also it will help to know the Bangla characters better for writer identification/verification. Here, highest accuracy of 100% is achieved in case of numeral 7 applying LibSVM classifier.

A Corpus of Word-Level Offline Handwritten Numeral Images from Official Indic Scripts

Dataset development is one of the most imperative tasks in document image processing research. The problem becomes more challenging when it comes about Numeral Image Database (NIdb) for official Indic scripts. Few efforts are made so far but they were restricted on single script which is basically a local script of the fellow researcher who prepared the database. In this paper, a technique for development of a handwritten NIdb of four popular Indic scripts namely Bangla, Devanagari, Roman and Urdu is proposed. Initially data were collected in unconstrained manner at Word-level from different writers with varying age, sex and educational qualification. All the images are stored in grey-level at .jpg format so that the data can be used in various ways as per need. A benchmark result on the present dataset is proposed using a novel hybrid approach with respect to Handwritten Numeral Script Identification (HNSI) problem.

Offline Writer Identification and Verification—A State-of-the-Art

In forensic science different unique bio-metric information of humans are being used to analyses forensic evidence like finger print, signature, retina scan etc. The same can be used applied on handwriting analysis. The Automatic Writer Identification and Verification (AWIV) is a study which combines forensic analysis field and computer vision and pattern recognition field. This paper presents a survey of literature on the offline handwritten writer identification/verification with the type of data, features and classification approaches attempted till date in different languages and scripts. The analysis of the approaches has been described for further enhancement and adaptation of these techniques in different languages and scripts.

Individuality of Bangla numerals

This paper presents analysis of individuality of handwritten Bangla numerals. It has a great prospect in Writer Identification, Writer Verification, Forensic Science etc. After collecting and extracting characters from filled in forms, 400 dimensional feature vectors is computed based on gradient of the images. A total of 450 documents were used for this work. In our experiment we have used LIBLINEAR classifier of WEKA environment. We have computed and analyzed the Individuality of each numeral and observed that the numeral 5 has the most individuality property than other numerals and 0 has the least. We have also done the writer identification with all the numerals and obtained 96.5% accuracy with all writers.

Word & Character Segmentation for Bangla Handwriting Analysis & Recognition

Segmentation of unconstrained handwritten word into different zones (upper middle and lower) and characters is more difficult than that of printed documents. This is mainly because of variability in inter-character distance, skew, slant, size and curved like handwriting. Sometimes components of two consecutive characters may be touched or overlapped and this situation complicates the segmentation task greatly. In Indian languages such touching or overlapping occurs frequently because of modified characters of upper-zone and lower-zone. In this paper we propose a simple method to segment unconstrained handwritten Bangla word. We achieved 81.41% success rate in the proposed system.