Input sensitive thresholding for ancient Hebrew manuscript

In this paper, we describe an input sensitive thresholding algorithm for ancient Hebrew calligraphy documents. Usually, historical document images are of poor quality since the documents have degraded over time due to storage conditions. However, the distribution of noise in one document is not uniform and the characters quality may vary. We develop tools to identify noisy characters and apply more sophisticated tools to process them. First, we use a global thresholding method to obtain an initial binary image. This suffices for noise free characters. Then we evaluate the document characters and invoke an accurate local method only on the noisy characters. Results show that our method detects a very high percent of the noisy characters, and that the local method achieves very accurate results.