Image Repair

The Problem The Check 21 law has ushered in a new set of challenges for check imaging technology. Image quality that was sufficient for sending out images of returned checks in monthly statements is not sufficient for accurate data extraction. To accurately extract MICR data and CAR/LAR data from check images, the check images need to be optimized for the respective OCR and ICR engines. Images that were previously required to only “look good” for inclusion in bank statements, now need a new criteria applied - usability. The real world is full of “bad” check images that are difficult to read by people and machines. Example types of difficult to read imaged checks are: Postal money orders Traveler checks Business and personal checks with lines Graphics and art as background that are not printed to specification In a bank imaging system, automatic data extraction from check images is one of the primary criteria for projecting a short return on investment from the financial institution investment in check imaging technology. The alternative is expensive manual re-keying and repairing of check images, thus turning the automatic process into a manual process. Examples of where automatic image repair can increase the usability of data extraction are: Remote and distributed capture – remote, low-speed scanners can be misaligned, with dirty, dusty lenses, threshold levels incorrectly set, poorly trained operators, etc. Recent studies have shown a 1% substitution error rate from low speed remote check scanners. Click here to read a presentation about this study Limiting Your Liability When Imaging Checks When printing IRDs the MICR line of the substitute check must by law be the same as on the check image. Click here to read how data from On-Us fields are sometimes not included in the electronic exchange data base. http://www.allmypapers.com/solutions_4.htm Extracting accurate read rates can be a challenge when the source data is poorly imaged checks. The goal is to extract the most reliable read rates with the highest confidence values. All My Papers technology is able to image process the difficult images and accurately and speedily extract data separating the data from the interfering background automatically. We solve the hard problems and minimize the exceptions so there is nominal need for manual repair of defects. All My Papers Has the Experience Over 10 years ago, a large U.S. financial institution wanted to automate the process of selling stock certificates by scanning the certificates and transforming them into a document images. Embedded in a stock certificate are the serial number and number of shares that the certificate represents. Each of these numbers is printed over a security scroll. A security scroll background is designed to minimize alteration of these numbers. If someone attempts to alter the number of shares or the serial number the back ground would also be altered as well. Since the numbers were printed over the background it was difficult for machine reading OCR software to differentiate between the number and lines touching the numbers from the background security scroll. All My Papers developed a set of intelligent filters to leave the numbers on the certificate, while removing the interfering background. Today an updated version of these filters are applied to check images to remove background patterns, lines and check art so the MICR data and CAR/LAR information can be successfully and accurately extracted with high confidence. All My Papers has two different types of image processing and repair suites of functions. The programmer has the option of retaining these processed images or just using them to obtain the best data extraction. The two types of image processing are: Image Prep – which comes standard and performs the following functions Deskew Auto- rotation Right side up positioning Thinning or thickening of bi-tonal characters – morphological filters Image Repair – an optional function which automatically repairs the image for better data extraction and readability Excessively Dark Background patterns Void stamps that interfere with data read Black triangles from improper skew correction