Image analysis – Image segmentation
Reexamination Certificate
2001-08-21
2004-11-02
Boudreau, Leo (Department: 2621)
Image analysis
Image segmentation
C382S236000, C375S240080, C375S240120
Reexamination Certificate
active
06813379
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to an image segmentation apparatus which is suitable for the case where segmentation according to an encoding rate is conducted in optimizing segmentation encoding which minimizes the generated entropy.
DESCRIPTION OF THE RELATED ART
In optimal segmentation for encoding, it is necessary to previously estimate generated entropies or data quantities from areas for each of segments. As a prior art of this optimal segmentation method, for example, there is a technique described in Japanese Patent Application Laid-Open No. 11-41611 assigned to the present assignee.
According to this prior art, it is possible to conduct such segmentation as to minimize the total generated entropy which is a sum total of a motion vector entropy, a shape entropy, and a texture entropy (which is equal to a prediction error entropy in the case of motion compensative encoding).
As for the motion vector entropy included in the generated entropies, a certain entropy can be estimated fixedly per motion vector. Furthermore, the shape entropy can be estimated by using a method described in Japanese Patent Application Laid-Open No. 11-41596 assigned to the present assignee.
Furthermore, as for the estimation of the texture entropy, the following method is conceivable. In other words, assuming that a texture signal of an area to be encoded can be approximated by, for example, Laplace distribution, its signal variance is &sgr;
2
, and the number of pixels in an area is N, an estimated value E
1
of a texture entropy from that area is represented by the following equation (1).
E
1
=N/
2·log
2
2
e
2
&sgr;
2
[bits] (1)
This equation is based on a texture signal of that area before quantization.
As a matter of fact, however, there is a restriction in the entropy which can be transmitted. In other words, the transmission rate is finite. Therefore, the entropy of the texture signal is typically reduced by quantization. Especially in the case of a low bit rate, it happens that the signal becomes 0 in some portions of an image, as a result of “truncation”, i.e., quantization.
From the equation (1), it is usual for the texture entropy to differ depending upon the area shape. If truncation occurs, however, the texture entropy eventually becomes zero no matter what shape is assumed. It is desirable in this case to adopt such a shape as to reduce the shape entropy as much as possible.
As a matter of fact, there is a limit in the entropy which can be transmitted. In some cases, therefore, the texture entropy of the equation (1) becomes zero. In this case, however, it is not considered in the prior texture entropy estimation method of the equation (1) to adopt such a shape as to reduce the shape entropy as much as possible. Especially in a low bit rate, therefore, if segments become too fine, the shape entropy occupies a large part of the total generated entropy. Eventually, this results in a problem that the image quality degradation becomes remarkable.
SUMMARY OF THE INVENTION
An object of the present invention is to provide an image segmentation apparatus capable of conducting such optimizing segmentation as to minimize the generated entropy while taking a limit of the entropy which can be transmitted (encoding rate) into consideration.
In order to accomplish the object, a feature of this invention resides in that an image segmentation apparatus comprises first means for estimating a shape entropy of a certain area of an image signal; second means for estimating a motion vector entropy of the region; third means for estimating a texture entropy based on a signal variance of the area and an allowed value of encoding degradation determined according to an encoding rate; and means for adding up the entropies derived by the first to third means and thereby estimating a generated entropy of the area.
According to this feature, an encoding degradation value determined according to the encoding rate is added to estimation of the texture entropy. As a result, segmentation according to a given encoding rate can be conducted.
REFERENCES:
patent: 5144426 (1992-09-01), Tanaka et al.
patent: 5969764 (1999-10-01), Sun et al.
patent: 6005980 (1999-12-01), Eifrig et al.
patent: 6377309 (2002-04-01), Ito et al.
patent: 11-41611 (1999-02-01), None
patent: 11041611 (1999-12-01), None
Eryurtlu et al., Very low-bit-rate segmentation-based video coding using contour and texture prediction, IEEE, vol. 142, ISSN 1350-245X, pp. 253-261.
Kawada Ryoichi
Koike Atsushi
Matsumoto Shuichi
Boudreau Leo
KDDI R&D Laboratories Inc.
Sherali Ishrat
Westerman Hattori Daniels & Adrian LLP
LandOfFree
Image segmentation apparatus does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Image segmentation apparatus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image segmentation apparatus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3342134