Techniques to determine document recognition errors

#Various endeavors result in the creation of paper documents. These paper documents frequently comprise text, which may include one or more of handwritten text, typewritten text, and computer-generated text. These documents may further comprise images. These paper documents may be digitized, converted into a computerized form, and captured as a digital image. These digital images may be exchanged via various systems, such as telefacsimile, the Internet or other network, or the physical exchange of storage devices. As such, people may receive digital images representing physical documents that contain various kinds of text and images.

#The following presents a simplified summary in order to provide a basic understanding of some novel embodiments described herein. This summary is not an extensive overview, and it is not intended to identify key/critical elements or to delineate the scope thereof. Some concepts are presented in a simplified form as a prelude to the more detailed description that is presented later.

#Various embodiments are generally directed to techniques to determine document recognition errors. Some embodiments are particularly directed to techniques to determine document recognition errors using comparisons between received document images and reconstructed document images. In one embodiment, for example, an apparatus may comprise an image reception component operative to receive an incoming document image; a document recognition component operative to perform a document recognition process on the incoming document image to generate a recognized document; and collect document recognition process information for the document recognition process of the incoming document image; a document reconstruction component operative to generate a reconstructed document image based on the document recognition process information; and an image comparison component operative to compare the reconstructed document image to the incoming document image to generate document recognition performance information for the recognized document. Other embodiments are described and claimed.

#FIG. 1 illustrates an embodiment of a document recognition system.

#FIG. 2A illustrates an embodiment of an original document image.

#FIG. 2B illustrates an embodiment of a known document form.

#Institutions regularly receive documents that, while generated using computer or typewriter fonts, are represented as computer images or in a physical form like paper, such that the text content is represented visually rather than as machine-encoded text. For instance, a document that has been scanned or sent as a facsimile may be received. However, the institutions may prefer to ultimately represent the document as machine-encoded text so as to empower the computer-aided processing of the text content of the document.

#As such, received document images may be processed to generate recognized documents with machine-encoded text. Optical character recognition (OCR) techniques may be used for this process, including known OCR techniques. Particularly where form data is being recognized, the highest possible OCR accuracy is desirable. As such, techniques to improve the accuracy of an OCR process are beneficial to people relying on OCR in their handling of forms.

#Particularly where important form data is being recognized, human operators may be used to check the accuracy of the machine-encoded text in recognized documents. For example, a single error in a recognized account number could result in significant problems if not caught. However, as forms may be long, human operators may use significant time to review recognized documents and may struggle to retain focus across dozens of document pages. As such, the operators may be aided by an OCR evaluation process that identifies possible problems in the recognized document, specific sub-portions of the document with identified potential errors, for them to compare to the original document image and possibly correct recognition errors.