Many companies rely on a combination of automatic pattern recognition technology and manual data processing. Even with automated forms processing, bad data can get through. And accuracy levels for automated recognition have been under criticism since the technology was first introduced. Performance is typically determined by the amount of information that is 100% correctly processed at an ‘x’ percentage of the time. Manual keying, in itself, is not error-proof, and is the baseline for comparison.#ContentManagement #datacapture #Recognition #risktollerance #formsprocessing #dataentry #patternrecognition #documents #OCR #ICR #accuracy #ScanningandCapture
Recognition technology does not encounter human factors such as boredom and repetition, but performance varies depending upon other issues, such as the application, quality of the images, etc. Therefore, it is important to fully understand the application and its limitations in order to understand overall performance.
For their specific applications, optical Character recognition (OCR) and intelligent character recognition (ICR) demonstrate high levels of accuracy when working with constrained text (i.e., lines, boxes and combs). Working with high-quality machine print, OCR provides nearly 99.9 % recognition accuracy — high enough to be acceptable without additional controls for most OCR applications. With different machine fonts, this high level of accuracy can vary.
When the need for unconstrained handprint or cursive writing recognition is required, the need for more complex analysis arises. Standard read accuracy rates become more an art than a science and depend heavily on the types of form data, the instructions typically included (such as the instructions “please print” or “enter date as MM/DD/YYYY”), the cost of human intervention, and desired processing throughput rate. Again, accuracy can be in the 80% to mid-90% range depending on a number of variables and tolerance thresholds.
Accuracy can also be further increased through improved practices in form design and preparation, context and sensitivity, database cross validation, and other techniques.
In the end, it is critical to understand the accuracy required vs. risk tolerance for each particular application. Recognition technologies can be leveraged to dramatically reduce costs, yet the up-front configuration and tolerances identified by the organization are critical to a successful implementation. A thoroughly considered approach will reduce both the usage of manual processing and the costs of fixing errors caused by a manual process.
For more on Intelligent Character Recognition and how it can simplify forms processing, join me on January 19th, where I will address the topic in a live webinar, Keys to successful handwriting recognition. It is free to register.