Image analysis – Applications – Personnel identification
Reexamination Certificate
1999-07-28
2003-09-30
Mariam, Daniel G. (Department: 2621)
Image analysis
Applications
Personnel identification
C382S137000, C382S181000, C382S217000, C382S306000, C382S321000, C235S380000, C705S044000
Reexamination Certificate
active
06628808
ABSTRACT:
BACKGROUND
There is a need in the Optical Character Verification (OCV) industry for an OCV apparatus and method with increased reliability of verification of each character or image scanned into a computer system.
Specifically, there is a need for an OCV method that is different from traditional Optical Character Recognition (OCR) and OCV algorithms. Traditional algorithms are based on analysis of pure bi-level images. Bi-level images contain only two colors or two levels of intensity. Typically these are visualized as black-and-white images. Everything is black (one level) or white (the other level). There exist no in-between levels. Furthermore, traditional OCR and OCV approaches rely mainly on statistical analyses of the foreground pixels of the scanned image.
An example is the case of OCR of black text on white paper. When these types of documents are scanned (or whenever these images are otherwise obtained in computer format), a gray-scale image is normally obtained. The gray-scale image has no “color,” but rather contains varying intensities of gray. White is represented as a very bright gray, black is represented as a very dark gray. There are many intermediate levels of gray between these two extremes. It is well known in the art how to convert the raw gray-scale image into a bi-level form by choosing a cutoff value. All pixels which have an intensity that is greater than the cutoff are made white (the brightest possible gray-scale intensity) and all other pixels are made black (the darkest possible gray-scale intensity).
A “Gray-Scale Card Image” is an image of a card that is scanned with any conventional scanning system that is generally commercially available (e.g. an UltraChek I system). Also, executing an intensity normalization algorithm to a card makes the card image visually appear slightly better than the raw UltraChek I scanned image. However, it does not fundamentally affect the character image processing.
The card image has varying intensities of gray on it. Although the card background is not quite perfectly white, it is generally close to being perfectly white, with a few scattered gray pixels throughout. Heavy characters appear as a very dark gray, with intermediate intensities of gray appearing near their edges. Small characters will appear significantly lighter in intensity that the large characters. This effect is normal and is predominantly due to the resolution of the camera used for scanning the card image. Also, neighboring pixels will generally contribute part of their intensity to a central pixel, resulting in an intensity smearing effect.
A histogram of the pixel intensities for the gray-scale card image is illustrated in FIG.
1
. The white background accounts for most of the pixel data of the scanned image. This is indicated as the large spike
20
centered around the 220 mark. Moreover, the spike is so large that it is off the scale. There is also a large spike
22
around the 0 mark accounting for the majority of the black text that appears on the source card. In addition, minor spikes of intermediate intensities are registered as the pixels fade from the full black color of the text to the full white color of the background.
The image represented by the histogram may be converted to a black and white image by choosing an appropriate cutoff value (e.g. 128). Choosing the appropriate cutoff value and marking all intensities above the cutoff as white, all others as black, yields a bi-level, black and white, image.
This technique enables the OCR algorithm to easily recognize the larger characters. However, since many of the smaller characters will become significantly distorted, the OCR algorithm will not be able to accurately recognize these characters. One contributing factor to the quality of the bi-level image is the distortion introduced by the UltraCheck-I image capturing system. The image capturing system tends to blend pixels together, such that physically adjacent pixels on the card actually contribute to logical pixels in the resultant scanned image. For example, the dots used to form the colon characters on the small text will be distorted in intensity by the background pixels that surround them. The blending effect results in fuzzy cutoff values between foreground and background pixels. The distortion can be somewhat reduced by evaluating the pixel intensity histogram in the immediate vicinity of the character itself, rather than making the cutoff decision by considering the entire card.
Regardless of the approach, whenever a cutoff value is used, there typically exists a significant noticeable distortion in the character images once they are converted to bi-level format. However, for a solid color text printed on a solid color background, the bi-level form of the image is typically adequate for a person to recognize the characters.
Once the bi-level form of the image has been created, conventional character processing algorithms will attempt matching the character images that appear on the source image (e.g. the card) with a set of reference character images (e.g. a reference template of characters or images stored in the memory circuits of a computer). Character matching is based on some correlation between the source and reference images. Several different correlation approaches are used by conventional software programs.
In many cases, conventional software programs attempt to isolate individual characters appearing on the source image. This produces a series of discrete character images that may be individually processed. There are other systems that operate based on recognizing larger sequences of characters, not just recognizing one character at a time. However, the majority of software programs operate on one character at a time.
At this point, there is a divergence in the approaches taken by conventional OCR and OCV software programs. Conventional OCR software programs recognize the text without having knowledge of what text should actually be present, whereas conventional OCV software programs verify the text according to a known set of text data.
Typically, OCR software programs operate on a substantial amount of text (e.g. a printed page). Accordingly, the OCR software program must read all the characters on a page fairly quickly. It cannot spend a lot of time on any single character. In contrast, OCV software programs typically operate on a very short string of text (e.g. a dozen characters) and it generally knows what text is supposed to be present at a particular character coordinate location. Therefore, the OCV software program can spend more time analyzing individual characters than the OCR software program can.
Once an OCR software program isolates an individual source character, some type of comparison is made between the scanned character from the source image and a reference character image stored in the memory of a computer, or controller. The comparison is often performed in the frequency domain, rather than in the spatial domain, using the Fourier transform or other similar frequency transformations to convert the scanned image into a frequency domain representation. Character images typically exhibit less variance in appearance when transformed into the frequency domain rather than the spatial domain. Therefore, individual missing pixels are not as relevant in the frequency domain. Therefore, the matching process is able to occur more quickly and reliably.
Although frequency domain comparisons of character images are common and well known in the art, there exists some systems that use spatial comparison techniques. These techniques include computing basic characteristics about the source character image, such as the number of connectors, the number of closed curves, etc. The comparisons are then used to narrow the source characters to a reasonable set of characters that may actually be present in the source image. Subsequently, a statistical match is then performed on the individual pixels to actually recognize the set of scanned characters.
In both frequency domain and spatial domain comparison cases, the
Bach George
Johnson Brian P.
Nichols Dean R.
Datacard Corporation
Mariam Daniel G.
LandOfFree
Apparatus and method for verifying a scanned image does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Apparatus and method for verifying a scanned image, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Apparatus and method for verifying a scanned image will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3029150