Handwritten word recognition using lottery ticket hypothesis based pruned CNN model: a new benchmark on CMATERdb2.1.2
- PDF / 894,481 Bytes
- 12 Pages / 595.276 x 790.866 pts Page_size
- 101 Downloads / 214 Views
(0123456789().,-volV)(0123456789(). ,- volV)
ORIGINAL ARTICLE
Handwritten word recognition using lottery ticket hypothesis based pruned CNN model: a new benchmark on CMATERdb2.1.2 Samir Malakar1 • Sayantan Paul2 • Soumyadeep Kundu2 • Showmik Bhowmik3 • Ram Sarkar2 Mita Nasipuri2
•
Received: 16 July 2019 / Accepted: 19 March 2020 Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Handwritten word recognition, a classical pattern recognition problem, converts a word image into its machine editable form. Mainly two basic approaches are followed to solve this problem, one is segmentation-based and the other is holistic. A number of research attempts have shown that the holistic approach performs better than its counterpart when the lexicon is predefined, fixed and small in size. Relying on this, initial benchmark recognition accuracy on CMATERdb2.1.2, a publicly available database consists of handwritten city names in Bangla, was reported following a holistic word recognition protocol. In the present work, we have followed the same trend to recognize the word samples of the said database and set a new benchmark recognition accuracy. A sparse convolutional neural network (CNN)-based model which is a low-cost trainable model has been developed for this. We have relied on a recently proposed hypothesis, known as lottery ticket hypothesis for pruning the layers of CNN model methodically, and derived a low-resource model having much less number of training parameters. This model competently surpasses the previously reported recognition accuracy on the said database by a significant margin with an axed training cost. Keywords Handwritten word recognition CNN model Lottery ticket hypothesis Bangla script CMATERdb2.1.2
1 Introduction & Samir Malakar [email protected]; [email protected] Sayantan Paul [email protected] Soumyadeep Kundu [email protected] Showmik Bhowmik [email protected] Ram Sarkar [email protected] Mita Nasipuri [email protected] 1
Department of Computer Science, Asutosh College, Kolkata, India
2
Department of Computer Science and Engineering, Jadavpur University, Kolkata, India
3
Department of Computer Science and Engineering, Ghani Khan Choudhury Institute of Engineering and Technology, Malda, India
The process of handwritten word recognition (HWR) aims to convert a handwritten word image into its machine editable form generally in Unicode representation. Variations in writing style of different writers make HWR a challenging task. Writing style of the writers may vary based on their sex, age, profession, educational qualification and mood. In addition to these, the type and quality of writing medium increase the complexity of the recognition task, i.e. words written on paper using pen/pencil in offline mode are more complex to recognize than writing on tablet, mobile phone, etc., using stylus or digital pen in online mode [1]. Irrespective of sample collection mode, HWR systems follow either segmentation-based approach or holistic approa
Data Loading...
