Summary & Purpose
In handwritten text skew and slant are inevitably introduced, but to varying degrees depending on several factors, such as the writing style, speed and mood of the writer. Therefore skew and slant detection in offline handwritten text words and their subsequent correction have become the critical pre-processing steps in Document Analysis and Retrieval systems to neutralize the variability in writing styles to improve the performance of word and character recognition systems. In this paper, we present two new methods for the estimation of slope and slant angles of offline handwritten word images along with two novel core-region detection techniques for both skewed and non-skewed text words. Besides this, we also prepare multilingual datasets comprising both real and synthetic handwritten word images along with ground truth information related to slope and slant of the same to address the lack of standard datasets in this regard. These datasets are made publicly available as word-level slope and slant datasets are scarce, especially words written in Bangla and Devanagari. Extensive experimental results prove the efficiency of the proposed methods compared to contemporary state-of-the-art methods. Moreover, the method is robust, efficient, and easily implementable.
Date of Publication or Submission
Users are free to share, copy, distribute and use the dataset; to create or produce works from the dataset; to adapt, modify, transform and build upon the dataset as long as the user attributes any public use of the dataset, or works produced from the dataset, referencing the author(s) and DOI link. For any use or redistribution of the dataset, or works produced from it, the user must make clear to others the license of the dataset and keep intact any notices on the original dataset.
Data Source Credits
IAM dataset: http://www.fki.inf.unibe.ch/databases/iam-handwriting-database
Single Dataset or Series?
This is an open access database that contains four different sets of handwritten text words. The text words are written by several writers belonging to different age groups, starting from school-going children to aged people. There are four databases along with corresponding ground truth information about their writing slope and slant - of these three are new benchmark datasets having words written in Bangla, Devanagari and Roman. Each dataset consists of 500 handwritten words. Fourth one is a synthetic dataset having 180,500 words created from the IAM database. The newly collected datasets are named as Dataset A, Dataset B and Dataset C which are written in Bangla, Devanagari and Roman scripts respectively. The fourth dataset, Dataset D, is prepared by using 500 word images taken from the IAM database, and through image processing techniques calibrated slope and slant have been added. To collect the word samples for Datasets A, B & C, we made 30 equal sized blocks on an A4-size white sheet and writers were requested to write words naturally inside the blocks. Then the A4 sheets were scanned at a 300 dpi resolution and saved as RGB images. After that the individual words were cropped programmatically. Finally, we selected 500 words from each of the three scripts to prepare 1500 benchmark text words. It has been noticed that even though words are written within boxes, most of the word samples are skewed as well as slanted naturally. In order to prepare the database D, we selected 500 words from a large set of IAM. The chosen words were visually checked to be de-skewed as well as de-slanted and have a length of more than one character. The selected words are then binarized and cropped to confine all the text pixels and reduce extra manipulation of background portions. Then the cropped words are slanted in 19 different angles from -45° to 45° with a step of 5° by using the shear transform. In this way we generated 9,500 synthetic words with known ground truth. Finally, each of the slanted words are skewed in 19 different angles -45° to 45° with steps of 5° by rotating the word image about its centroid. This procedure generates 180,500 synthetic words for which the ground truth skew and slant values are known. The ground truth information for each word image is provided within the image name itself. Database A: WBEN_Sequence number (001)_Slant Angle (-18)_Slope Angle(14).jpg Database B: WHIN_ Sequence number (001)_Slant Angle (-17)_Slope Angle (-15).jpg Database C: WENG_Sequence number (001)_Slant Angle (-28)_Slope Angle(-7).jpg Database D: HWE_Slant Angle (-10)_Slope Angle (40)_ Sequence number (001).png
Privacy and Confidentiality Statement
Boise State is explicitly compliant with federal and state laws surrounding data privacy including the protection of personal financial information through the Gramm-Leach-Bliley Act, personal medical information through HIPAA, HITECH and other regulations. All human subject data (e.g., surveys) has been collected and managed only by personnel with adequate human subject protection certification.
Disclaimer of Warranty
BOISE STATE UNIVERSITY MAKES NO REPRESENTATIONS ABOUT THE SUITABILITY OF THE INFORMATION CONTAINED IN OR PROVIDED AS PART OF THE SYSTEM FOR ANY PURPOSE. ALL SUCH INFORMATION IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY KIND. BOISE STATE UNIVERSITY HEREBY DISCLAIMS ALL WARRANTIES AND CONDITIONS WITH REGARD TO THIS INFORMATION, INCLUDING ALL WARRANTIES AND CONDITIONS OF MERCHANTABILITY, WHETHER EXPRESS, IMPLIED OR STATUTORY, FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT.
IN NO EVENT SHALL BOISE STATE UNIVERSITY BE LIABLE FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF INFORMATION AVAILABLE FROM THE SYSTEM.
THE INFORMATION PROVIDED BY THE SYSTEM COULD INCLUDE TECHNICAL INACCURACIES OR TYPOGRAPHICAL ERRORS. CHANGES ARE PERIODICALLY ADDED TO THE INFORMATION HEREIN. COMPANY AND/OR ITS RESPECTIVE SUPPLIERS MAY MAKE IMPROVEMENTS AND/OR CHANGES IN THE PRODUCT(S) AND/OR THE PROGRAM(S) DESCRIBED HEREIN AT ANY TIME, WITH OR WITHOUT NOTICE TO YOU.
BOISE STATE UNIVERSITY DOES NOT MAKE ANY ASSURANCES WITH REGARD TO THE ACCURACY OF THE RESULTS OR OUTPUT THAT DERIVES FROM USE OF THE SYSTEM.
Bera, Suman Kumar; Chakrabarti, Akash; Lahiri, Sagnik; Barney Smith, Elisa H.; and Sarkar, Ram, "Dataset for Normalization of Unconstrained Handwritten Words in Terms of Slope and Slant Correction" (2019). Signal and Image Processing Lab. 2.