Searching for TELUGU Script in Noisy Images Using SURF descriptors
Kesana Mohana Lakshmi, Tummala Ranga Babu · 2016
Day by day scanned documents are increasing to preserve the old literatures digitally, especially Telugu literatures. During this scanning so much noise comes into the scanned documents, in addition to noise, other geometry distortions, scale changes will take place while scanning. In this scenario, general optical character recognition (OCR) systems fail to extract the text due to these problems. So, an efficient mechanism is needed to extract TELUGU text. Towards this, in this work, an efficient algorithm to search for the TELUGU text has been proposed, which uses Speeded Up Robust Feature Transform (SURF) to extract descriptors and uses the KD tree algorithm for matching purpose. The presented results prove the robustness of the system under different noise and partial occlusion by strip-lines, orientation changes like rotation etc.