Image analysis – Image compression or coding – Pyramid – hierarchy – or tree structure
Reexamination Certificate
2000-04-24
2003-08-26
Do, Anh Hong (Department: 2624)
Image analysis
Image compression or coding
Pyramid, hierarchy, or tree structure
C382S254000, C382S274000
Reexamination Certificate
active
06611627
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to digital image processing methods for improving the apparent sharpness of images. In particular, the invention is directed to a class of such methods known as edge reconstruction. The invention can be used in various applications where image signals are processed and displayed or printed, such as in digital printing, photofinishing, medical imaging, video imaging, motion pictures, and graphic arts.
BACKGROUND OF THE INVENTION
Many factors degrade image sharpness during their capture and display processes. For example, focus error and camera shake cause image blur in the capture process, and the non-ideal system transfer functions of monitors or printers cause image blur in the display process. Prior art techniques for deblurring an image and enhancing its sharpness roughly fall into four main approaches: unsharp masking, inverse filtering, local histogram modification, and edge reconstruction. Unsharp masking works by decomposing an input image into a low frequency component and a high frequency component. The latter is then amplified and added back to the low frequency component to create a sharper-appearing output image. The inverse filtering approach is aimed at explicitly reversing the blurring process by applying a filtering operation which is an inverse of the image blurring operation. Because of its sensitivity to noise, inverse filtering is often applied with some smoothness constraint on the output image signal. The local histogram modification method is based on the idea that if the gray level differences among neighboring pixels are stretched out by modifying the histogram of a local window, the contrast of the image details will be more visible and the image will appear sharper. The edge reconstruction method extracts edge information from an input image at many spatial scales. At each scale, the edges are manipulated to enhance their sharpness. The final output image is then reconstructed from the modified edge signals.
U.S. Pat. No. 6,005,983 teaches an edge reconstruction method of image enhancement based on a Laplacian pyramid. At each resolution, the Laplacian of the image signal is clipped and then processed through a high-pass filter to sharpen the Laplacian response. The output image is reconstructed from the modified Laplacian pyramid. One drawback with this technique is that the Laplacian is sensitive to noise. Furthermore, because clipping affects every spatial frequency, it is difficult to control the way edges are sharpened in the high-pass filtering.
U.S. Pat. No. 5,963,676 teaches a method of image enhancement by a different implementation of the Laplacian pyramid. An input image is filtered with a series of low-pass filters. At each filter output, four one-dimensional Laplacian operators at different angles are applied to the low-passed image signal. The Laplacian response (unsharp masked signal) is then amplified through a nonlinear function and then added to the input image to create sharper-appearing edges. The nonlinear function has an upper clipping limit, a lower undershoot limit, and a middle linear-gain range. For its intended application in x-ray angiography, the linear gain is set very high so that the directional Laplacian can pick up very fine vessel structures and render them with very high contrast. The method is obviously designed for x-ray angiography for its emphasis on curve structures, but not for other types of edge or texture. Adding the Laplacians of an image to enhance the image sharpness is basically an unsharp masking approach and suffers from the problems of creating overshoots along edges and failing to differentiate between different types of edges, such as texture, shading and occlusion boundaries in an image.
Mallat and Zhong teaches a method of edge-wavelet multiresolution analysis for image representation and compression in their paper, “Characterization of signals from multiscale edges,” published in
IEEE Transactions on Pattern Analysis and Machine Intelligence
, 14, 7, 710-732, 1992. Image compression is achieved in their work by finding and keeping only the local extrema in their edge-wavelet coefficients. Finding local extrema of image gradients as an edge detector is commonly known as Canny's edge detector proposed by F. Canny, in his paper, “A computational approach to edge detection,”
IEEE Transactions on Pattern Analysis and Machine Intelligence
, vol. PAMI-8, 6, 679-698, November 1986. Mallat and Zhong also teaches a method of image reconstruction from edge-wavelet extrema by iterative projection onto feasible solution sets. The iterative projection method is too computationally intensive, and hence too time-consuming to be of practical use on commonly available image processing computers such as personal computers. It currently takes several hours to process a single image using the iterative projection method.
Lu, Healy, and Weaver teach a method of image enhancement by increasing the amplitude of the edge-wavelet coefficients at the local extrema in their paper, “Contrast enhancement of medical images using multiscale edge representation,” published in
Optical Engineering
, 33, 7, 2151-2161, 1994. Their method relies on an iterative projection based on the self-reproduction kernel corresponding to the edge-wavelet. This approach requires a large amount of computation and on relatively low speed image processing hardware, such as a personal computer, it is very slow because it takes several long iterations for the algorithm to converge. Another weakness of this method is that the same gradient amplification factor is applied to all edges, independent of the edge properties. As a consequence, edges corresponding to noise and shading are enhanced as well as those corresponding to object boundaries and other discontinuities.
Vuylsteke and Schoeters also teach a method of contrast enhancement in their paper, “Multiscale image contrast amplification,” published in
SPIE Proceedings
, Volume 2167, 551-560, 1994. Their method uses a power function for amplifying the edge contrast.
U.S. Pat. No. 5,717,791 teaches a method of image contrast enhancement based on the combination of Mallat and Zhong's edge-wavelet transform and Lu, Healy, and Weaver's idea of amplifying wavelet coefficients. A constraint is imposed on the amplification factor such that the factor is a function of the average wavelet maxima. The larger wavelet coefficients get amplified less than the smaller wavelet coefficients. The motivation was to amplify the low contrast edges more than the high contrast edges. However, the amplitude of wavelet coefficient is not an intrinsic edge property, because it depends on the image metric. For example, when a digital image is scaled up by a factor of 10, all the wavelet coefficients are also scaled up by a factor of 10. Therefore, wavelet amplitudes alone do not specify the nature of the edges. This reconstruction transform is also slow when implemented on relatively low speed computers because of its iterative processes.
In summary, the prior edge reconstruction methods for image enhancement suffer from two common drawbacks: (1) the iterative reconstruction process is too slow, and (2) the amplification of the edge contrast does not take into account the intrinsic properties of the edge. There is a need therefore for an improved method of image processing using edge reconstruction to enhance the appearance of the image that avoids these problems of the prior art.
SUMMARY OF THE INVENTION
The need is met according to the present invention by providing a digital image processing method that includes the steps of: transforming the digital image using an edge sensitive wavelet transform to produce a plurality of wavelet coefficients at various resolutions and a residual image; modifying the wavelet coefficients as a function of the image gradient and its rate of change at a resolution corresponding to the respective wavelet coefficient; and inverse transforming the modified wavelet coefficients and the residual image to produce a processed digi
LaRossa Gina N.
Lee Hsien-Che
Do Anh Hong
Woods David M.
LandOfFree
Digital image processing method for edge shaping does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Digital image processing method for edge shaping, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Digital image processing method for edge shaping will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3126148