<p>Accurate statistical information provided in real-time during sports broadcasts serves as a key element that significantly enhances viewers’ understanding and immersion. Especially in data-driven sports like baseball, it is more important to deliver precise information to viewers in real-time. Optical Character Recognition (OCR) technology, which automatically recognizes text information displayed on screens, has been widely adopted in sports for various applications such as highlight extraction and statistical analysis. However, despite its usefulness, OCR technology still faces various challenges in the context of sports broadcasts. Complex backgrounds, diverse fonts, and rapidly changing screens often cause OCR systems to make errors. These OCR errors not only provide viewers with incorrect information but also decrease the accuracy of automated data analysis systems. To overcome these limitations of OCR, we propose a novel OCR system that incorporates a prior knowledge-based post-processing technique. The proposed post-processing method utilizes similarity functions and ensemble mechanisms to enhance OCR recognition accuracy. We derived optimal combinations of various similarity functions considering language-specific characteristics and constructed the specialized post-processing systems for each language and error type. We conducted various analyses and experiments using the actual Major League Baseball (MLB) and Korea Baseball Organization (KBO) game videos, and our approach achieved performance improvements ranging from 7.14% to 60.71% compared to the existing technologies. Moreover, the results validated that the proposed technique can flexibly respond to various languages and error types by leveraging the combination of diverse similarity functions.</p>