Image analysis – Image enhancement or restoration – Image filter
Reexamination Certificate
1998-06-03
2001-01-09
Lee, Thomas D. (Department: 2724)
Image analysis
Image enhancement or restoration
Image filter
C382S299000
Reexamination Certificate
active
06173084
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The invention relates to a method of processing an image. The invention also relates to an image processor and to an x-ray examination apparatus.
2. Description of Related Art
The article Image filtering using multiresolution representation in the IEEE Transactions of Pattern Analysis and Machine Intelligence, 13(1991)426-440, by S. Ranganath, discloses a method for processing an image, wherein the image is decomposed into detail images at successive resolution levels, and the detail images are filtered so as to generate filtered detail images. The processed image is reconstructed from the filtered detail images.
The cited reference discloses image filtering using multiresolution representation. Image filtering is implemented as the convolution of the image and a two-dimensional mask. Preferably, a pyramid structure is applied for the multiresolution representation. In the cited reference the detail images of the multiresolution representation implemented by the pyramid structure are named ‘coded images’. The cited reference mentions that the ‘coded images’ are filtered. The known method aims at achieving a good compromise between noise removal and loss of resolution. It has been found that the known method is not satisfactory with respect to reducing noise while retaining edges in the image. In particular, the known method is not suitable for producing a processed medical image having a high diagnostic quality, i.e. an image in which small details having low contrast are displayed so as to properly visible.
SUMMARY OF THE INVENTION
It is an object of the invention to provide a method of processing an image, which reduces noise and retains relevant image information more effectively than the known method.
This object is achieved by the method of processing an image according to the invention, which method comprises the steps of
decomposing the image into detail images at successive resolution levels,
filtering detail images to generate filtered detail images,
filtering a current detail image in dependence on at least one subsequent detail image, said subsequent detail image having a resolution level which is lower than the resolution level of the current detail image, and
reconstructing a processed image from the filtered detail images.
The detail images at successive resolution levels contain image information at respective spatial scales. The detail images are ordered according to decreasing resolution levels and increasing coarseness, that is the subsequent detail image contains coarser image features than the current detail image. Fine details are prominent in the detail image having the finest (highest) resolution level; progressively, coarser features become more prominent at the coarser (lower) resolution levels. Preferably, the decomposition of the image into detail images is carried out by means of a pyramid structure. In such a pyramid structure, the finest resolution level is at the base of the pyramid, the coarsest resolution level is at the top of the pyramid. In the pyramid structure, the detail images are derived as frequency-band images, i.e. spatial-frequency band-pass filtered versions of the image. The filtering involves a local averaging of the detail images so as to reduce differences between pixel-values that are due to noise. As an alternative, the filtering may include a non-linear transformation of pixel-values. Pixel-values represent brightness values such as grey-values or colour-values of the image. Because the filtering of the current detail image depends on one or several detail images at coarser resolution levels, the method according to the invention distinguishes between image structures that appear at several resolution levels and image structures at only one or a few resolution levels. For example, pixel-values of the current detail image are retained when there is a significant corresponding structure at coarser resolution levels, but pixel-values of the current detail image are locally smoothed when there is hardly any significant structure at coarser resolution levels. In particular, a steep edge in the image appears at almost each of the resolution levels, but an extended slope in the pixel-values of the image appears almost only at one or a few subsequent resolution levels. It has been found that filtering in dependence on several lower resolution levels produces good results when said filtering is controlled by a sum of gradients of pixel-values of detail images having several resolution levels. The method of the invention retains relevant image information, notably directional structure-like edges in the image, and reduces noise even in portions of the image containing variations of pixel-values on a coarse scale. Because the filtering is adjusted on the basis of image information with spatial frequencies lower than the spatial frequencies of the current detail image, the filtering is notably insensitive to noise. The method of processing an image is particularly suitable for processing a medical diagnostic image, such as an x-ray image, a magnetic resonance image or an ultrasound image. Such a processed medical diagnostic image has a very small noise component and renders small details having little contrast properly visible.
A preferred embodiment of a method of processing an image according to the invention controls the filtering on the basis of the next detail image. Image structures, such as edges, relating to relevant image information appear most strongly at the next coarser resolution level. Since filtering of the current detail image is controlled on the basis of the correlation between image information in the current and the next detail images, it is effectively achieved to reduce noise while retaining relevant image information in the current detail image.
A preferred embodiment of a method of processing an image according to the invention controls filtering on the basis of differences between pixel-values in the subsequent detail image. Image structures are predominantly represented by differences between pixel-values. Notably image structures in the form of edges gives rise to differences between pixel-values at different positions in the image. The current detail image is strongly filtered, i.e. smoothed, at locations corresponding to locations in a detail image of the subsequent resolution level where there is a relatively small local difference between pixel-values, especially a small gradient of pixel-values. The pixel-values of the current detail image are retained at locations corresponding to locations in a detail image of the subsequent resolution level where there is a relatively large local difference between pixel-values, especially a large gradient of pixel-values. Image structures are predominantly represented by oriented structures like edges or lines. These structures have a strong variation in intensity in one direction in the image whereas the intensity variation in the direction at right angles thereto is small. In the vicinity of such an image structure, the gradient of pixel-values has a high magnitude and a direction perpendicular to the orientation of the image structure. Local averaging perpendicularly to the gradient, i.e. along the elongate image structure hardly affects the strength of visually important intensity transitions, notably variations along the image structures are mainly due to noise.
Edges or other steep transitions result in strong local gradients in detail images at several resolution levels. Moreover, the direction of maximum local variation and therefore also the direction of the gradient will be substantially equal in detail images at several resolution levels. The amplitude of the gradient may vary from level to level depending on the steepness of the transition and the size of the structure. Therefore the information contained in the gradients for different resolution levels is similar for adjacent resolution levels. The information in the coarser resolution levels is less affected by noise and is therefore more reliable.
Aach Til
Koch Dietmar U.
Kunz Dietmar W.
Brinich Stephen
Lee Thomas D.
Renfrew Dwight H.
U.S. Philips Corporation
LandOfFree
Noise reduction in an 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 Noise reduction in an image, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Noise reduction in an image will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2538118