This paper describes the derivation of probability of correctness from scores assigned by most recognizers. Motivation for this research is three-fold: (i) probability values can be used to rerank the output of any recognizer by using a new set of training data; if the training data is sufficiently large and representative of the test data, the recognition rates are seen to improve significantly, (ii) derivation of probability values puts the output of different recognizers on the same scale; this makes comparison across recognizers trivial, and (iii) word recognition can be readily extended to phrase and sentence recognition because the integration of language models becomes straightforward. We have conducted an extensive set of experiments. The results show a reranking of recognition choices based on the derived probability values leading to an enhancement in performance. The performance of many different digit recognizers improved by 1-4% points on a blind set of images. 1 Introduct...

In several tasks from different fields, we are encountering sparse events. In order to provide with probabilities for such events, researchers commonly perform a maximum likelihood (ML) estimation. However, it is well-known that the ML estimator is sensitive to extreme values. In other words, configurations with low or high frequencies are respectively underestimated or overestimated and therefore nonreliable. In order to solve this problem and to better evaluate these probability values, we propose a novel approach based on the probabilistic logic (PL) paradigm. For a sake of illustration, we focuss on this paper on events such as word trigrams (w 3 ; w 1 ; w 2 ) or word/pos-tag trigrams ((w 3 ; t 3 ); (w 1 ; t 1 ); (w 2 ; t 2 )). These latter entities are the basic objects used in speech or handwriting recognition. In order to distinguish between for example: "replace the fun" and "replace the floor" an accurate estimation of these two trigrams is needed. The ML estimation is equival...