Image analysis – Image compression or coding – Interframe coding
Reexamination Certificate
2000-12-18
2004-11-09
Boudreau, Leo (Department: 2621)
Image analysis
Image compression or coding
Interframe coding
C382S107000, C382S239000, C375S240020, C348S407100
Reexamination Certificate
active
06816617
ABSTRACT:
BACKGROUND OF THE INVENTION
1) Field of the Invention
This invention relates to a motion vector searcher and a motion vector search method as well as a moving picture coding apparatus suitable for use for motion vector detection, for example, in digital moving picture coding.
2) Description of the Related Art
In recent years, a service of distributing images of a high picture quality in multiple channels is carried out in broadcasting media such as satellite broadcasting and a CATV (Cable Television of Community Antenna Television System), and a compression technique called MPEG (Moving Picture Experts Group) has been developed.
A moving picture has various redundancies. For example, according to an image pickup method called panning, since an image of an object which is stationary on a screen is picked up while the orientation of a camera is varied with the object fixed, the object appearing on a screen at present and the object appearing on a succeeding screen exhibit no significant difference of data regarding the image. Accordingly, the image is first converted into data only with regard to a portion which relates to motion information of the object.
Here, several terms used in the present specification are defined in the following manner. The screen is what a physical display apparatus displays, and the image is an object such as a body, a landscape and so forth which are included in the screen. Further, the picture element is a point on the screen which corresponds to the resolution of the image and has a value called luminance value or color difference value. A pixel of, for example, a personal computer corresponds to the picture element.
A still picture has a spatial redundancy that adjacent pixels are similar to each other. For example, where the pattern of the screen is plain and in full white, the color of an adjacent picture element can be analogized more readily than where the pattern of the screen includes two different colors of white and red.
On the other hand, a moving picture has a temporal redundancy that similar pictures successively appear with respect to time. The temporal redundancy has a characteristic that the luminance value of a certain picture element has a high correlation to the luminance value of a neighboring picture element in time. For example, in the movie, images which are different a little from each other are successively displayed to present a dynamic image, and therefore, preceding and succeeding pictures in time are similar to each other.
Therefore, the MPEG system has two different functions of DCT (Discrete Cosine Transform) for removing a spatial redundancy of a still picture and motion compensation (MC) for removing a temporal redundancy of a moving picture. Here, to remove a spatial redundancy signifies to convert an image into frequency components using orthogonal transform (for example, DCT) to eliminate high frequency components to reduce the amount of information to be transmitted.
Meanwhile, to remove a temporal redundancy signifies to determine differences between frames without using compression within a frame (intra-frame compression which is hereinafter described) and encode the difference information. Encoding of such difference information is more effective to reduce the amount of information than use only of the intra-frame compression. In particular, where a motion search which is hereinafter described is performed to select a position at which the difference is comparatively small, the information amount can be reduced more effectively than where the difference of a block at the same position between frames is calculated. It is to be noted that the difference is represented, for example, as (10, 15) and also called motion vector, and is expressed, for example, from a variable length code (VLC) having a value ranging from −16 to +15 and an f code.
The DCT is a process of dividing an original image into square macro blocks (a macro block may be hereinafter referred to simply as MB) of 16×16 picture elements and decompose each MB into different frequency components from a low frequency component to a high frequency component. The DCT can convert a distribution of luminance values which are random on the frequency axis into another distribution in which the frequency components appear in a concentrated manner in a low frequency region. Further, the DCT can eliminate high frequency components and thus allows information compression. Further, the compression ratio in such information compression can be changed as hereinafter described.
The motion compensation signifies a method of determining a difference between the position shifted by a fixed value from a picture (also called coded picture), which is an image processing unit, and a coded picture of an object of compression. Here, the picture signifies an individual still picture from still pictures of a moving picture and specifically indicates a frame or a field. The frame and the field arise from a difference in scanning method, and in the following description, the picture is used with a higher concept than the frame and the field. A difference between the frame and the field is hereinafter described above.
A method of obtaining a motion vector is called motion search and is performed by comparison between an original image and a reference image. First, an original image MB is determined from a single original image, and a quadrilateral region of a reference image different from the original image is determined as a search range. Then, a search MB (reference MB) of the same size as that of the original image MB is selected from within the search range, and the search MB is shifted within the range from −16 to +15 both horizontally and vertically. Then, differences between the original image MB and the shifted search MBs are calculated each as a sum of absolute values (absolute value sum) or a square sum, and a relative position of that one of the search MBs at which the lowest difference value is obtained within the ranges is determined. Accordingly, the arithmetic operation of the difference is performed, for example, by 256 times.
It is to be noted that the search range in the following description is, for example, from −16 to +15 MB distances in both of the horizontal and vertical directions. However, the motion search number may otherwise be set to a value different from 256. The MPEG system provides, for example, also a scheme called half pel motion compensation, by which also a motion vector (−3.5, 5.5) or the like can be produced. In the following description, the difference amount is sometimes used to signify the difference value.
FIG. 16
illustrates a motion search. Referring to
FIG. 16
, an MB
100
a
is provided in an original image picture
100
, and a search range
110
a
is provided in a reference picture
110
. A motion search is performed centered at the position of the MB
10
a
, and difference calculation between the MB
100
a
and a shifted MB
110
b
for each picture element in the search range
110
a
is performed.
FIG. 17
illustrates a motion vector. An outer framework shown in
FIG. 17
is a range (the search range
110
a
) of the reference picture
110
for which a search is performed. Here, an arrow mark which is directed in the left upper direction is a motion vector
111
, and an MB
110
c
is at the same position as that of the MB
100
a
in the original image picture
100
(refer to
FIG. 16
) while the MB
110
b
is at a position at which the motion search evaluation value exhibits its smallest value within the search range
110
a
. Such motion search evaluation values are calculated based on individual difference amounts obtained by raster scanning by which the search position is successively shifted one by one MB distance from the left upper corner of the search range
110
a
, and the lowest value of the motion search evaluation values is determined as a motion vector.
FIG. 18
illustrates block matching, and the block matching is described with reference to
FIG. 18. A
motion searcher
1
Sakai Kiyoshi
Yamori Akihiro
Akhavannik Hussein
Boudreau Leo
Fujitsu Limited
Staas & Halsey , LLP
LandOfFree
Motion vector searcher and motion vector search method as... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Motion vector searcher and motion vector search method as..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Motion vector searcher and motion vector search method as... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3279637