Estimating the readability of handwritten text - a Support Vector Regression based approach

This paper presents a new approach to estimating the readability of handwritten text. The estimation task is posed as a regression problem. A novel support vector regression (SVR) system is used to estimate the recognition rate of a text recognizer on a given text. The estimated recognition rates are used to classify text as either readable or unreadable. Unreadable text can then be filtered out prior to recognition, thus avoiding needless recognition attempts or a high cost caused by manual correction. The system is systematically evaluated on a data set of 1,830 text lines from 50 writers.