Candidate word generation for OCR errors using optimization algorithm

OCR post-processing is an important step to improve text accuracy. It includes two main tasks, error detection and correction. Hill climbing algorithm a heuristic search method used for solving optimization problems. In this paper, we present novel correction approach using adapted version of the algorithm. Correction candidates errors are explored by random character edits evolved with climbing. The edit patterns obtained from training data. proposed model evaluated on benchmark dataset in post-correction competition International Conference Document Analysis Recognition 2017. shown that our outperforms various baseline approaches competition. addition, randomness analyzed verify its stability under parameter configurations.