A new approach for the failure mode analysis of optical character recognition algorithms

Concerns the imprecise method of OCR algorithm development and improvement. While trial-and-error techniques must be employed, the evaluation of the failed attempts is the most crucial, tedious, and error prone task. The paper presents a precise and efficient solution to this problem. The overlooked cause of the problem stems from the passive nature of the analysis performed. The proposed approach is highly interactive by employing a tight integration between the analyst and the OCR code they are attempting to find fault with or improve upon. This close coupling can be achieved through the use of special tools built around the code which provide a level of control and precision previously unavailable to developers. By establishing a virtual awareness of the operations being performed and additionally providing the means to hypothesize scenarios with each of the OCR algorithms, an improved state of failure mode analysis (FMA) can be reached. This new approach to algorithm advancement circumvents the current inferior solution to a daunting problem and will allow the future of optical character recognition to rise to a new level at an excitingly fast pace.