Image analysis – Image compression or coding – Transform coding
Reexamination Certificate
1999-10-25
2003-03-04
Mehta, Bhavesh (Department: 2621)
Image analysis
Image compression or coding
Transform coding
C382S232000, C382S233000, C382S240000, C382S251000, C375S240110, C375S240130, C375S240150
Reexamination Certificate
active
06529636
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to a digital image processing technology, and in particular to an image coding device for coding image data with a high efficiency and an image decoding device for decoding the coded data which has been coded by the image coding device.
BACKGROUND ART
Recently, subband coding techniques have been proposed as highly efficient image coding and decoding techniques. Among the subband coding techniques, a technique for decomposing an image into bands as shown in
FIG. 16
in which analysis of an input image is carried out by means of a band decomposing filter bank has been generally known as a technique having a high coding efficiency. Such a technique is described by, for example, Fujii and Nomura “Topics on Wavelet Transform”, technical report, IEICE, Institute of Electronics, Information and Communication Engineers, IE 92-11(1992).
FIG. 16
shows subband images which are obtained by conducting two dimensional subband decomposition for an input signal three times. A horizontal high frequency and vertical low frequency subband which is obtained by the first decomposition is designated as HL
1
. A horizontal low frequency and vertical high frequency subband is designated as LH
1
. A horizontal high frequency and vertical high frequency subband is designated as HH
1
. Subbands HL
2
, LH
2
and HH
2
are obtained as similarly to the foregoing by conducting second two-dimensional subband decomposition for the horizontal low frequency and vertical low frequency subband.
Subband HL
3
, LH
3
and HH
3
are obtained similarly to the foregoing by conducting third two-dimensional subband decomposition for the horizontal low frequency and vertical low frequency subband which has been obtained by second decomposition. A horizontal low frequency and vertical low frequency subband at this time is designated as LL
3
. The filter bank which is used for decomposing band may use a filter bank for wavelet transformation and a subband decomposing synthesizing filter bank and the like. The image which has been decomposed into subbands in such a manner has a hierarchical structure.
As a recent technique having the highest coding efficiency which is capable of adapting to the subband images, a ZTE (Zero Tree Entropy coding) technique using the above-mentioned hierarchical structure has been proposed (ISO/IEC JTC/SC29/WG11/MPEG95/N0441,ISO/IEC JTC1/SC29/WG11/MPEG96/M0637, ISO/IEC JTC1/SC29/WG11/MPEG96/M1539).
Now, the ZTE technique will be described. In the ZTE technique, a block structure which is shown in
FIG. 18
is formed by collecting subband coefficients (hereinafter referred to as coefficients) corresponding to the same spacial positions which are linked with each other by arrows as shown in
FIG. 17
from the image which has been decomposed in subbands. It has already known that there is a correlation between coefficients which are linked with each other by arrows in
FIG. 17
excepting the highest frequency subbands.
The whole relation of the coefficients which are linked with each other by arrows in
FIG. 17
is referred to as “trees”. One coefficient of each of the subbands (LH
3
, HL
3
, HH
3
) having a frequency one level higher than that of one coefficient of the lowest frequency subband (LL
3
) corresponds thereto (for example, a
1
, a
2
and a
3
correspond to a
0
in FIG.
17
). Four coefficients of each of the subbands (LH
2
, HL
2
, HH
2
) having a frequency one level higher than that of each of these coefficients correspond thereto (for example, a
10
, a
11
, a
12
, a
13
correspond to a
1
in FIG.
17
). Sixteen coefficients of each of the subbands (LH
1
, HL
1
, HH
1
) having a frequency one level higher than that of each of four coefficients correspond thereto. Trees with respect to coefficient a
0
is shown in FIG.
19
. White circle ◯ and solid black circle &Circlesolid; in
FIG. 19
denote coefficients in each subband. The trees in upper area comprise coefficients of the subbands having a lower resolution while the trees in lower area comprise coefficients of the subbands having a higher resolution.
In such a tree structure, the coefficients having lower resolution are referred to as “parents” and the coefficients having next higher resolution in the same spacial position as designated by arrows are referred to as “children”. In
FIG. 19
, for example, coefficient a
0
is a parent for coefficients a
1
, a
2
and a
3
, which are in turn children for coefficient a
0
. Coefficient a
1
is a parent for coefficients a
10
, a
11
, a
12
and a
13
and, coefficients a
10
, a
11
, a
12
and a
13
are children for coefficient a
1
.
All coefficients having higher resolution in the same spacial position which are linked with each other by arrows with respect to one parent are referred to as “descendants” and all coefficients having a lower resolution in the same spacial position which are linked with each other by arrows with respect to one child are referred to as “ancestors”. In
FIG. 19
, for example, the coefficients encircled with a dotted line are descendants for coefficient al and coefficients a
10
, a
1
and a
0
are ancestors for coefficient a
100
.
Then, the coefficients are quantized in the block basis. Three symbols are assigned to each node of the trees for representing whether the quantization coefficient is zero or non-zero. Definition of the symbol will now be described. The coefficient having the lowest frequency among the coefficients in which one coefficient in a tree is zero and the coefficients of its descendants are all zero is referred to as zero-tree-root (ZTR). Since this coefficient and the coefficients having a higher resolution than that of the former coefficient are all zero at this time, it would be unnecessary to code the coefficients of its descendant if ZTR appear on a tree. When any one coefficient in a tree is not zero, but the coefficients of its descendant are all zero, the coefficient in interest is referred to as valued zero-tree root (VZTR). If there is any one non-zero coefficient in the descendant, its coefficient is referred to as “Value”.
White and solid black circles denote the coefficients which the quantizing value is zero and non-zero, respectively in FIG.
19
. In this case, the coefficients which require coding are shown in FIG.
20
. Since a
0
has a quantizing value which is not “zero” in
FIG. 20
, the symbol Value is assigned to code the quantizing value. Since a
1
and its descendants (a
10
through a
13
, a
100
through a
103
through a
133
) are all zero, symbol ZTR is assigned to a
1
and it is not necessary to code the quantizing value. Since it can be found that the value of a
1
is zero due to the fact that a
1
is ZTR, it is never necessary to code the information on the descendants of a
1
.
Since a
2
has a quantizing value which is not zero, but its descendants all have a quantizing value which is zero, symbol VZTR is assigned for coding only the quantizing value of a
2
. Concerning the descendants of a
2
, same as those of a
1
, it is not necessary to code their information. Since a
3
has a quantizing value which is not zero and there are some descendants which have a quantizing value which is not zero, symbol Value is assigned for coding the quantizing value. VZTR is assigned for a
30
. ZTR is assigned for a
31
. Value is assigned for a
32
and a
33
. Only the quantizing values of the coefficients having the highest frequency (a
320
through a
333
) are coded without assigning a symbol to the coefficients. As mentioned above, the information to be coded on this block comprises:
symbol information including Value, ZTR, VZTR, Value, VZTR, ZTR, Value, Value, Value, Value, Value, . . . , Value and coefficient information including Q(a
0
), Q(a
2
), Q(a
3
),
Q(a
30
), Q(a
32
), Q(a
33
), Q(a
320
), Q(a
321
), Q(a
322
), . . . , Q(a
333
), wherein Q(a) denotes the quantizing value of the coefficient a. The contents of coded data are shown in FIG.
21
.
When the symbol is VZTR or Value, it is necessary to code the quantizing values of the coefficients. Since there are gen
Aono Tomoko
Ito Norio
Katata Hiroyuki
Kusao Hiroshi
Bayat Ali
Mehta Bhavesh
Sharp Kabushiki Kaisha
LandOfFree
Picture coding device and picture decoding device does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Picture coding device and picture decoding device, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Picture coding device and picture decoding device will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3036227