Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
1999-09-28
2004-01-06
Johns, Andrew W. (Department: 2621)
Image analysis
Pattern recognition
Feature extraction
C382S269000
Reexamination Certificate
active
06674903
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to a method for improving the visual appearance of a digital image. In a specific embodiment, the invention relates to reducing staircasing effects as a result of electronic zooming of a low resolution image. The method is suitable for enhancing digital images obtained by an electronic camera.
BACKGROUND OF THE INVENTION
Original images are usually represented electronically by a rectangular array of picture elements or pixels, which together make up the electronic image. Each pixel represents a small rectangular portion of an original image. The length of each rectangular portion or inversely, the number of pixels per unit of length in the original image is referred to as the spatial resolution of the electronic image. The spatial resolution is expressed in pixels per millimeter. Usually the resolution is constant in each direction, such that each pixel represents a square portion. The electronic image may also have a constant maximum resolution in a first direction and a constant minimum resolution in a second direction orthogonal to the first direction. In that case, each pixel represents a non-square rectangular portion of the original image. The resolution is dictated by the size of the original image to be represented electronically and the maximum number of pixels available in a row or a column of the rectangular array for the electronic image. If the size of the original image is 1 meter and the number of pixels is 1000, then the resolution of the electronic image will be 1000 pixels per meter or 1 pixel per mm. That electronic image may be rendered on a device capable to print 10 pixels per millimeter, i.e. a device having a spatial device resolution of 10 pixels per mm. As such, the 1000 pixels of the electronic image will be rendered on 100 mm. The original image of 1 meter will thus be rendered on 1 decimeter, i.e. at a scale of {fraction (1/10)}. If however the original image must be rendered at a scale of ⅕, i.e. the electronic image must occupy 200 mm, then two options are available. As a first option, a device may be selected having a spatial device resolution of 5 pixels per mm. Then 1000 pixels will be rendered on 200 mm. Alternatively, as a second option, the device having a resolution of 10 pixels per mm may be used, to render a second electronic image of 2000 pixels, derived from the first electronic image having 1000 pixels. The first electronic image has a resolution of 1 pixel per mm, the second electronic image still refers to 1 meter of the original image, but represented by 2000 pixels. As such, the second electronic image has a spatial resolution of 2000 pixels per meter, i.e. 2 pixels per mm. Thus, the spatial resolution of the second electronic image is twice the resolution of the first electronic image. Several techniques are known in the art to convert a first electronic image having a first spatial resolution to a second electronic image having a second spatial resolution. Such techniques are referred to image scaling. The most common and easiest image scaling technique is referred to as pixel replication or nearest neighbour interpolation. To double the spatial resolution of a first electronic image according to pixel replication to obtain a second electronic image having a higher spatial resolution, in a first step each pixel is replicated horizontally in each one image line of the electronic image, i.e. the first pixel of the first electronic image is copied to the first and second pixels of an intermediate electronic image; the second pixel of the first electronic image is copied to the third and fourth pixels of the intermediate electronic image; etc. In a second step, each line of the intermediate electronic image is replicated vertically in the second electronic image, i.e. the first line of the intermediate electronic image is copied to the first and second lines of the second electronic image; the second line of the intermediate electronic image is copied to the third and fourth lines of the second electronic image; etc. By this scaling technique, a single black pixel on a white background in the first electronic image results in a block of 2 by 2 black pixels on a white background in the second electronic image. A black horizontal line having a width of one pixel on a white background in the first electronic image results in a black horizontal line having a width of two pixels on a white background in the second electronic image. The same applies for a vertical line. An oblique black line on a white background in the first low resolution electronic image, e.g. a line having a slope of 45° and having a width of 1 pixel in the first image, i.e. a set of single black pixels connected to each other by the upper right and the lower left corners of the pixel, will result in the second high resolution electronic image in a set of black squares of 2×2 pixels each, connected to each other by their upper right and lower left corners. This effect, which is even more conspicuous for oblique lines having slope different from 45°, is referred to as staircasing. The staircasing effect is also visible in the enlarged image where shorter oblique line segments are present in the original electronic image, or even where an abrupt transition from a white area to a black area is present and the line of transition is neither horizontal nor vertical. The above example describes a scaling factor of 2. Also other integer scaling factors such as 3, 4, etc. may be envisaged. Even non-integer scaling factors are possible, such as a factor of 1.5 where each even pixel of the first electronic image is replicated once and each odd pixel of the first electronic image is replicated twice.
By use of another scaling technique, referred to as bilinear interpolation, transitions are smoothed, in the sense that between a black and a white pixel of a transition, a grey pixel will be introduced by the interpolation technique. Also this may be an unwanted effect, since the originally sharp transitions tend to look fuzzy on the enlarged image.
A third scaling technique, referred to as cubic spline interpolation, creates the new pixels of the second high resolution electronic image by convolution of four pixels of one line in the first low resolution electronic image with a third order polynomial, which:
1. is zero outside the interval x=[−2,2],
2. takes the value y=1 in x=0,
3. takes the value y=0 in x=−2, −1, +1 and +2
4. is everywhere continuous, i.e. for x=(−∞,+∞); and,
5. has everywhere continuous first and second order derivatives.
This convolution technique applied horizontally and vertically substantially eliminates the staircasing effect and the smoothing effect on the grey values, but may introduce other artefacts known to be due to reconstruction, such as overshooting, where an abrupt transition from light grey pixels to dark grey pixels may introduce pixels having a greyness lighter than the light grey or darker than the dark grey.
Despite the above mentioned drawbacks of image scaling, this operation becomes increasingly necessary in electronic image processing applications for the following two reasons. First of all, low cost image scanners or one-shot electronic cameras may produce low resolution images. The size of the original image or scene may be fixed, e.g. 210 mm×297 mm on an A4 page to be scanned or 5 m horizontally by 3.75 m vertically of an object in the scene to be captured by an electronic camera. Also the number of pixels horizontally and vertically may be fixed, such as 1280 pixels horizontally and 960 pixels vertically as in the ePhoto 1280 (Trademark) electronic camera distributed by Bayer Corporation, Agfa division Wilmington, Mass. As such, the object of the scene will be stored in the electronic image at a resolution of 256 pixels per meter. If the user reproduces the object of the scene by printing the electronic image on a 400 dpi (15.75 pixels per mm) printer, the size
AGFA-Gevaert
Johns Andrew W.
Sabourin Robert A.
LandOfFree
Method for smoothing staircase effect in enlarged low... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for smoothing staircase effect in enlarged low..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for smoothing staircase effect in enlarged low... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3210288