Image analysis – Image enhancement or restoration
Reexamination Certificate
1998-12-14
2002-02-26
Rogers, Scott (Department: 2624)
Image analysis
Image enhancement or restoration
C382S275000, C382S132000, C382S191000, C382S195000, C382S205000, C382S206000, C382S279000
Reexamination Certificate
active
06351571
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates in general to image enhancement in digital image processing and in particular to a method for rapidly analyzing the high- and low-frequency components of pixels in an arbitrarily shaped region of an image without producing artifacts at edges of the region.
BACKGROUND OF THE INVENTION
It is a common desire to enhance images acquired from imaging devices such as those used for medical imaging diagnostics. Such enhancement may involve amplifying detail and local contrast and sharpening edges. Enhancement algorithms, however, typically produce undesirable artifacts at the edges of the image that result from the analysis of the image into high- and low-frequency components. It is therefore desirable to provide a method for enhancing images while suppressing these artifacts.
Methods for the spatial frequency decomposition of images using the unsharp masking approach have been described by many authors. Two methods for convolution of signals of finite extent, circular convolution and linear convolution, are described by Oppenheim and Schafer (“Digital Signal Processing,” pp. 105-113, Prentice Hall, Inc., Englewood Cliffs, N.J., 1975). In circular convolution the image is assumed to be periodic so that for example the pixel to the right of the rightmost pixel is taken to be the leftmost pixel of the same line. This results in image data from the left edge of the image influencing convolutions near the right edge of the image and image data from the bottom of the image influencing convolutions near the top of the image. This can and often results in image artifacts near the edges of the image. Linear convolution takes data outside the image to be of constant value (commonly zero) for purposes of convolution. This causes artifacts within half the kernel width of the edge of the image. Circular and linear convolution are still commonly used.
In many cases the precise method for computing the convolution near the edges of the image is unspecified. This is the case, for example, in the book “Digital Image Processing” by Gonzales and Woods. The descriptions are in terms that are clearly applicable to only the central portion of the image, and not applicable to the edges of the image without further specification.
Therefore the techniques disclosed in these references do not provide a satisfactory solution for efficiently analyzing the high- and low-frequency components of entire images or specified by arbitrary selected regions-of-interest in images. Nor do they address efficient algorithms for these calculations.
SUMMARY OF THE INVENTION
According to the present invention there is provided a method of digital image enhancement, especially image enhancement of medical diagnostic (radiographic) digital images that solves these problems.
According to the feature of the present invention, there is provided a method for efficiently analyzing the low- and high-frequency components of a digital image. The algorithm is first defined in general terms, then specifically for three cases: an entire image, an arbitrarily specified region-of-interest, and an image with multiple arbitrary connected regions of interest.
A specific example is given for the case of the entire image, where the method of the present invention comprises the steps of: providing a rectangular digital image; specifying the size of a rectangular kernel; dividing that rectangular image into nine sub-regions based on the kernel size; applying the modified convolution operator (described herein) in each sub-region; and thereby computing the convolution at each pixel of the original image to obtain the low-frequency component of the original image. The high-frequency component is obtained by subtracting the low-frequency component image from the original image.
For the case of the arbitrarily specified region-of-interest, the method of the present invention comprises the steps of: providing an original digital image and a digital (mask) image describing a region of interest; specifying the size of a rectangular kernel; optionally analyzing the original image to find the smallest rectangle aligned with the pixel coordinates that inscribes the region of interest; dividing that rectangle or the full image into nine sub-regions based on the kernel size; applying the region-of-interest mask-weighted convolution operator (using one of the two methods described below) in each sub-region; and thereby computing the convolution at each pixel of the original image to obtain the low-frequency component of the original image. The high-frequency component is obtained by subtracting the low-frequency component image from the original image.
For the case of the multiple regions of interest, the method of the present invention comprises the steps of: providing an original rectangular digital image; creating an intermediate image for each region of interest of equal size or smaller than the original image; specifying the size of a rectangular kernel; optionally analyzing each intermediate image to find the smallest rectangle aligned with the pixel coordinates that inscribes the region of interest; dividing that rectangle or the full image into nine sub-regions based on the kernel size; applying the region-of-interest mask-weighted convolution operator (using one of the two methods described below) in each sub-region; and thereby computing the convolution at each pixel of the original image to obtain the low-frequency component of the intermediate image. The high-frequency component for each region of interest is obtained by subtracting said low-frequency component image from the intermediate image for that region of interest.
ADVANTAGEOUS EFFECT OF THE INVENTION
The invention has the following advantages.
1. The present invention does not exhibit edge profile artifacts near the borders since only original image pixels within the kernel are used to compute the convolution.
2. No arbitrary pixel value need be assumed for “padding” the image and image data from the opposite border is not included in the calculation.
3. The present invention is capable of limiting the computation of the convolution to any arbitrarily shaped region of interest or to a region of interest described by a convex hull. In the former case it only requires that the region be defined by a “mask” image. In the latter, the convex hull must be described.
4. The methods described in the present invention avoid the artifacts associated with the edges of the region of interest and can speed the computation as a smaller area is used for the convolution.
REFERENCES:
patent: 4794531 (1988-12-01), Morishita et al.
patent: 5457754 (1995-10-01), Han et al.
patent: 5524162 (1996-06-01), Levien
“Digital Signal Processing”, by Oppenheim and Schafer pp. 105-115, Prentice Hall, Inc., Englewood Cliffs, NJ, 1975.
Kohm Kevin S.
VanMetter Richard L.
Wang Xiao-hui
Eastman Kodak Company
Noval William F.
Rogers Scott
LandOfFree
Uniform convolution algorithm for frequency decomposition of... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Uniform convolution algorithm for frequency decomposition of..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Uniform convolution algorithm for frequency decomposition of... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2959230