Image analysis – Image compression or coding – Adaptive coding
Reexamination Certificate
2000-03-22
2003-12-09
Chen, Wenpeng (Department: 2624)
Image analysis
Image compression or coding
Adaptive coding
C382S240000, C382S238000, C382S131000, C375S240110
Reexamination Certificate
active
06661925
ABSTRACT:
I. BACKGROUND OF THE INVENTION
The present invention relates to methods and apparatuses for compressing data representing video or digital images. More specifically, the present invention relates to compressing data representing a set of images having a certain degree of similarity.
There are numerous areas of technology in which increasing volumes of image data must be processed and stored. This is particularly true with regard to medical radiological image data. The radiology department of a hospital may generate more than 500,000 digital images per year. One source estimates that a 1500-bed hospital may generate approximately 20 Terabytes of image data per year. Techniques for efficient classification, image processing, image database organization, and image transmission for analysis via telemedicine, have become important research areas.
The medical art has developed various Picture Archiving and Communication Systems (PACS) in an attempt to alleviate the problems associated with the creation of increasing volumes of medical image data. The functionality of most PACS is derived from the international Digital Imaging and Communication in Medicine (DICOM) standard. The DICOM standard specifies communications protocols, commands for image query and retrieval, storage standards, and similar parameters. However, prior art PACS still have many shortcomings. One serious disadvantage of conventional PACS is their insufficiency at reducing the large amounts of homogeneous data accumulating in radiological image databases. One method of reducing the amount of data in a database is to efficiently compress the image data prior to placing it in the database. Image data compression is also an important consideration in transferring the image from one location to another via network.
Typically, medical image data are compressed through conventional compression algorithms such as the standard Joint Photographic Experts Group (JPEG) algorithm. JPEG is presently the DICOM standard for data compression. JPEG utilizes Discrete Pulse Code Modulation (DPCM) for lossless and Discrete Cosine Transform (DCT) for lossy image compression. However, lossy compression is often avoided in medical imaging because of potential information loss, while lossless JPEG provides no means for compressing a series of similar images.
Alternate types of compression have been carried out with wavelet transforms. A wavelet transform decomposes a signal into a series of related waveforms similar to Fourier and DCT transforms decomposing a signal into a series of sine and cosine waveforms. The wavelet transform is based on a “scaling function” &phgr;(x) and derived from it is a “mother wavelet” &psgr;(x) satisfying the recursion relationship:
φ
(
x
)
=
∑
k
q
k
φ
(
2
x
-
k
)
,
(1a)
ψ
(
x
)
=
∑
k
h
k
φ
(
2
x
-
k
)
,
(1b)
where constants h
k
uniquely define functions &phgr;(x), &psgr;(x) and constants q
k
, and different choices of admissible h
k
(i.e. an h
k
for which a solution exists) produce different solutions to eqn. (1a) and eqn. (1b). Given &psgr;(x), a basis of sub-band waveforms or wavelets is constructed as successive translations and dilations of the mother wavelet. Thus, each wavelet in the &psgr;(x)−generated wavelet basis is represented by the equation:
&psgr;
j,k
(
x
)=2
j/2
&psgr;(2
j
x−k
), (2)
where k represents the degree of translation and j represent the degree of dilation. When an image I(x) is transformed with wavelet &psgr;(x), it is projected onto the v(x)-generated wavelet basis:
I
(
x
)
=
∑
j
,
k
w
j
,
k
(
x
)
=
∑
j
,
k
w
j
,
k
2
j
/
2
ψ
(
2
j
x
-
k
)
,
(
3
)
where w
j,k
is the coefficient associated with each wavelet In signal compression (and particularly in image signals), the common use of integer precision requires the coefficients w
j,k
to be integers, which can be always achieved in eqn. (3) with “wavelet coefficient lifting.” Wavelet coefficient lifting is well-known in the prior art as can be seen from publications such as “Wavelet Transforms that Map Integers to Integers” by R. C. Calderbank, Ingrid Daubechies, Wim Sweldens, and Boon-Lock Yeo, in “Applied and Computational Harmonic Analysis” (ACHA), Vol. 5, Nr. 3, pp. 332-369, 1998. The number of wavelets N representing an image is generally equal to the number of elements (e.g. pixels) making up the image. A less formal, but more convenient manner of representing the wavelet transform of an individual image I(x) is to use a single index i instead of index pairs (j,k):
I
(
x
)
=
∑
i
=
1
N
w
i
ψ
i
(
x
)
,
(
4
)
where &psgr;
i
(x) still represents a particular wavelet, w
i
represents the wavelet coefficient associated with that wavelet, i is an integer from 1 to N, and the variable x represents each individual pixel position of the image.
FIG. 1
illustrates conceptually a patient's head
1
and a series of horizontal brain CT images I
(1)
, I
(2)
, to I
(m)
taken therefrom. Each image may be represented by a wavelet equation such as equation (4). Thus, image I
(1)
may be represented as I
(1)
=w
1
(1)
&psgr;
1
+w
2
(1)
&psgr;
2
+. . . +w
N
(1)
&psgr;
N
, and image I
(m)
may be represented as I
(m)
=w
1
(m)
&psgr;
1
+w
2
(m)
&psgr;
2
+. . . +w
N
(m)&psgr;
N
, as indicated in FIG.
1
. Decomposing images into a wavelet transform with the above equations is well known in the art. One conventional software program which will determine the coefficients of an image for a given mother wavelet is MATLAB®, produced by The MathWorks, Inc. of 24 Prime Park Way, Natick, Mass. MATLAB® implements a “fast wavelet transform” technique which is well know in the art. The fast wavelet transform also orders the wavelet sub-bands in a sequence of lowest spatial frequency to highest spatial frequency. It will be understood that given the mother wavelet form and the wavelet coefficients w
1
, w
2
, w
3
. . . w
N
, the inverse transform may be applied and the original image reconstructed. However, transforming an image from the original pixel into the wavelet coefficient representation (eqn. (4)), often reduces redundant information typically present in images, thus providing the means for more compact image representation, i.e. compression. If the mother wavelet is known, the image is easily recovered from its wavelet coefficients with the inverse wavelet transform.
However, despite developments of new PACS and the introduction of wavelet transforms in data compression, the prior art has still failed to efficiently represent the inter-image informational redundancy that is usually present in any image database. This redundancy arises from the fact that radiological images typically comprise a set or “study” of different views or numerous “slices” of a particular organ. For example, a CT scan study of the brain may consist of approximately 150 vertical slices and approximate 30 horizontal slices. Each of these views or slices is a digital image requiring a significant number of bytes to represent: one view in an X-ray study of the chest may require as much as 10MB and thousands of such X-ray studies could be expected in a hospital's database.
It has been found that, statistically, the mean difference between images of the same organ in different individuals is much smaller than the difference between two unrelated images, such as images of two famous paintings. The lack of difference (or the similarity) of two images can be quantified by the correlation between the two images, where a correlation of zero reflects absolutely no similarity and a correlation of one reflects identical images. It was observed that the correlation between many images of the
Pianykh Oleg S.
Tyler John M.
Board of Supervisors of Louisiana State University & Agricultura
Chen Wenpeng
Jones Walker Waechter Poitevent Carrere & Denegre, LLP
LandOfFree
Functional set compression does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Functional set compression, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Functional set compression will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3170121