City name recognition for Indian postal automation: Exploring script dependent and independent approach

Postal documents are often used for official communication, online shopping, etc. At times, the delivery gets delayed due to multiple scripts leading to the need for postal sorting facilities. Understanding the destination city name plays a major part in solving automatic sorting problems as the same becomes more challenging due to the presence of handwritten documents. In order to develop an autonomous system to solve the problem, a Deep Learning-based system is proposed to recognize handwritten city names written in 6 major scripts namely Tamil, Roman, Devanagari, Bangla, Gurumukhi, and Arabic. Experiments were performed in both script-dependent (bi-stage) and independent approaches. In the bi-stage framework, we have obtained an average accuracy of 97.58% along with a back-end script recognition rate of 99.07% while in the script-independent approach, an accuracy of 97.03% was obtained on a dataset consisting of 807 classes.