JPH02200084A - High-efficiency encoding system - Google Patents

Abstract

PURPOSE:To reduce a prediction error by performing in-frame encoding at every (2n+1) frames. CONSTITUTION:A (2n+1) frame memory 21 is provided to store (2n+1) frames which become objects to be encoded to perform the encoding of the continuous frame of a moving image signal at every (2n+1) frames comprehensively (n is >=1 integer). Also, two one-frame memories 22 and 23 and a change-over switch 24 to supply data read out from those memories by switching to a motion compensation circuit 6 are provided to form a prediction signal (predicted value) based on preceding and succeeding frames. And the entire picture of one frame can be refreshed in a single action by performing the in-frame encoding at every (2n+1) frames, and also, inter-frame prediction in two way, forward and backward, directions can be performed by performing the inter-frame prediction encoding of the preceding and succeeding frames setting encoded and decoded data on which the in-frame encoding are applied as reference. In such a way, it is possible to reduce distance between frames to the half, and to reduce the prediction error by that share.

Description

Detailed Description of the Invention (Field of Industrial Application) The present invention is useful for processing moving image signals in various devices such as recording, transmission equipment, and other display devices that perform signal processing of digital signals. This relates to interframe predictive coding, which is a high-efficiency coding method for compressing and encoding and transmitting or recording information, and is particularly useful for information recording of general moving image signals rather than specific moving image signals such as video conferences. It is useful for recording on storage media such as disks and information recording tapes.

(Conventional technology) Video conferencing is an example of high compression technology for moving image signals.

There is a known encoding technology (high-efficiency encoding technology) that uses a smaller amount of code for videophone calls, but in this case, it is difficult to encode a special video signal in which there is movement of a person in the same background image. It becomes encoding.

Here, to explain the encoding technique, at the time of the initial screen or scene change (screen change), the frame is frozen (for example, a still image between 5 frames), and one rule is inserted in between. The image signal of the frame is intra-frame encoded, and then each frame is subjected to inter-frame predictive encoding (eg, motion compensated inter-frame predictive encoding) using the encoded/decoded data of the previous frame as a reference.

In interframe predictive coding, since each frame of a normal moving image is quite similar, the signal of the frame to be encoded is predicted from the signal of the previous frame, which has been widely encoded, and is based on the 8I standard. Only the error (residual) is encoded.

At this time, one frame is divided into blocks each having a size of about 8 x 8 pixels (if one frame is 360 x 240 pixels, it becomes 45 x 30 blocks), and each block is divided into blocks each having a size of about 8 x 8 pixels. (coded decoded image data) and motion compensated interframe prediction i1! ! lf
J difference signal is encoded and transmitted.

Further, at this time, data of blocks having a strong correlation with the previous frame is not coded and transmitted in order to compress the data.

In addition, in order to prevent a block that cannot be decoded due to an error during transmission from being decoded and reproduced thereafter, each block is encoded little by little by intra-frame encoding and transmitted. A method is used in which all blocks are refreshed about once every second.

FIG. 4 is a block diagram showing a typical conventional configuration of motion compensated interframe predictive coding.

In the same figure, the moving image signal that is continuously input from the image signal input terminal 1 has a predicted signal (predicted value) subtracted by the predicted signal subtracter 2, and the difference is the pre-3 standard + error (residual). is encoded by the encoder 3. At this time,
The changeover switch 7 is connected to the b side.

The predicted signal (predicted value) is locally decoded by the local decoder 4,
It is obtained by motion compensation in the motion compensation circuit 6 from the previous frame signal (encoded/decoded data) stored in the one frame memory 5.

Initial screen (initial state) and scene change (screen change)
At this time, the changeover switch 7 is connected to the a side (at this time, O (zero) is subtracted in the predicted signal subtracter 2, that is, there is no subtraction), and the entire frame is intra-frame encoded. Ru. The situation is shown in Figure 5, where frames F1 to F4 should originally be encoded.
That time is allocated to encoding frame Fo. This is because of the time required for encoding and the problem of data transmission. Therefore, frame Fo is 5 frames.

- continues (freezes) for 5 frames, becomes a still image for 5 frames, and then frames F5. FB, . . . are then encoded.

In FIG. 5, the 101 part indicates interframe predictive coding, and the r-hatched J part indicates intraframe coding (
partial refresh).

In addition, in the normal state other than the initial screen or scene change, partial refresh is performed as shown in Figure 6.
Only then, the changeover switch 7 is connected to the a side, and in this figure, six frame periods L frames F n , Fnn
+Fn+2, Fn+3, Fn+4.

Full refresh is performed at F + 5], but in reality it is refreshed at a frame rate of 30 frames/second for 1 second (3
In many cases, the entire image is refreshed over a period of about 0 frames).

Further, the encoding in the encoder 3 of FIG. 4 is performed first in the 8×8 DCT circuit 13 as shown in FIG.
Each 8-pixel block is two-dimensional! 1tr1. D sine transformation [D CT; Discrete CCo51
neTransform], and each of the obtained 8×8 coefficients is then multiplied by coefficient by the coefficient-by-coefficient quantizer 14.

Furthermore, the transmission encoder 8 converts the coefficient value from 0 (zero) to non-O (
(zero) coefficient value is obtained by hanging OCT as shown in Figure 8.
run-length encoded by scanning the coefficients,
These run-length codes are transmission-encoded by entropy coding (variable-length coding such as Huffman coding), and are output as variable-length digital data via a buffer 7 and memory 9 from a data output terminal reference, and are transmitted or recorded.

Here, the OCT coefficients after hanging are often 0 (zero), so the run-length encoding described above is effective.Furthermore, the probability distribution is set in advance as shown in Figure 9, and Huffman Data can be effectively compressed by encoding and transmitting it.

Note that the intra-frame/inter-frame coding switching control circuit 10 detects a scene change based on the output of the motion compensation circuit 6, and controls the switching of the connection of the changeover switch 7 at that time and at the time of the above-mentioned partial refresh. Intra/inter frame coding identification information is supplied to the transmission encoder 8, and intra/inter frame coding is switched.

The transmission encoder 8 is supplied with motion vector i~ information from the motion compensation circuit 6, and is also supplied with intra-frame/inter-coding identification information from the intra-frame/inter-coding switching control circuit 10. It is also transmission encoded.

When performing variable-length encoding, buffer memory is unavoidable, but if the same encoding was performed for each frame, the amount of code generated would depend on the type of image (for example, when there are large movements of people during a video conference). data cannot be sent on a transmission line with a certain transmission capacity.

Therefore, when the image has a small amount of fluctuation, it is smoothed in the buffer 7 memory, and when a large amount of fluctuation is about to occur, the buffer memory is monitored for maximum occupancy, and then the encoder 3 encodes the image via the encoding control circuit 12. control.

Note that the encoding control is specifically performed by the encoder 3 described above.
The filtering step of the 8×8 DCT coefficients is made rougher, or only the low-order coefficients are sent without sending all the coefficients.

<Problems to be Solved by the Invention> However, in the conventional high-efficiency encoding method described above, that is, in the cyclic interframe predictive encoding method using the previous frame, it is difficult to use continuous Although there are no major disadvantages when transmitting images, there are the following problems, especially when applied to recording on storage media such as information recording disks and information recording tapes.

■ Scene changes are common in general moving images, but it is unnatural for the image to freeze every time.

■ When recording and reproducing data on storage media, special measures against errors are required, but it takes up to one second to recover the block area where the error occurred.

■ Playback from the middle of the media is frequently performed (random access), but each time a playback image cannot be obtained for one second until the entire screen is refreshed.

■ When recorded on storage media, visual search (fast forwarding or rewinding while monitoring the rough screen) is required, but this is not possible.

■ Interframe predictive coding of blocks that are temporally distant from refresh will result in a large prediction error, and the compression effect of interframe predictive coding will decrease slightly.

Therefore, an object of the present invention is to provide a highly efficient encoding method that solves the problems of the conventional techniques described above.

<Means for Solving the Problems> In order to achieve the above object, the present invention collectively encodes consecutive frames of continuously input image signals into 2n+1 frames [n is an integer of 1 or more]. This is a high-efficiency encoding method, in which 2n+1 frames to be encoded are sequentially encoded by F-n, F-n+1°-, F-2, F-1, Fo.
, F +, F2. -, Fn-+.

As Fn, a means for intra-frame encoding the frame Fo therein, a means for inter-frame encoding the frames F-1 and F+ before and after this frame Fo, based on the encoded/decoded data of this frame Fo, and this frame. Fo, F-1, F
means for interframe encoding frames F-2 and F2 before and after the encoded/decoded data of + as a reference; and frames F o , Fl, F + , .
”, F-n+1. Provides a high-efficiency encoding method characterized by comprising means for inter-frame encoding the frames F-n and Fn before and after the encoded/decoded data of Fn-+ as a reference. It is something to do.

(Function) In the high-efficiency encoding system with the above configuration, (a) 2
By performing intra-frame encoding every n+1 frames, the entire screen of one frame is refreshed at once. (c) Using the encoded/decoded data of the intra-frame encoded frame as a reference, the frames before and after it are inter-frame encoded to perform inter-frame prediction in both forward and backward directions. With this technology, up to 2n
What used to be inter-frame encoding of frames separated by frames becomes inter-frame encoding of frames separated by a maximum of n frames, which reduces the inter-frame distance by half and reduces the error on the numerator side.

Historically, in (a) above, refresh is performed several times per second, so (in practical terms) the amount of intra-frame encoding increases, and therefore the amount of code increases. (2) improves the compression rate of interframe coding, and solves problems (1) to (4) of the prior art without significantly increasing the amount of code as a whole.

(Embodiment) FIG. 1 is a block diagram showing the configuration of an embodiment of the high-efficiency encoding system according to the present invention.

The basic configuration in this figure is similar to the conventional example, and the same components in FIG. 4 mentioned above are given the same numbers.

In FIG. 1, consecutive frames of a continuously input moving image signal are collectively converted into r:f0 every 2n+1 frames [n is an integer of 1 or more], so 2 is the encoding target.
It has (2n+1) frame memories 21 that store n+1 frames.

In addition, in order to form a predicted signal (predicted value) based on the previous and subsequent frames, two 1-frame memories 22 and 23,
There is a changeover switch 24 for switching the data read from these memories and supplying the data to the motion compensation circuit 6.

A case will be described in which consecutive frames of a continuously input moving image signal are collectively encoded every five frames using the above configuration.

[1 As shown in FIG. 2, the first five frames to be encoded among the consecutive frames of the moving image signal that are continuously input are F-2, F-1, Fo, F+.

As F2, first, these five frames are stored in the (2n+1) frame memory 21.

[Frame F in the middle order of the 215 frames
o is read from the (2n+1) frame memory 21, and this frame Fo is intraframe encoded. At this time, the changeover switch 7 is connected to the a side. And encoder 3
The local decoder 4 locally decodes the encoded data of the frame FO obtained from the
．． 23 respectively.

[3] Frame F1, which is the frame after frame Fo
(2n+1) is read from the frame memory 21, and the difference between this frame F1 and the motion compensated predicted value is calculated based on the data stored in the one frame memory 22 (that is, the encoded/decoded data of frame Fo),
Perform interframe predictive coding. At this time, selector switch 7
is connected to the b side, and the changeover switch 24 is connected to the a side. Then, the encoded data of frame F1 obtained from encoder 3 is locally decoded by local decoder B4, and it is
The frame memory 22 is rewritten and stored.

[4] Frame F-, which is the frame before frame Fo
1 (2n+1>) is read from the frame memory 21, and the difference between this frame F−1 and the motion compensated predicted value is calculated based on the data stored in the 1 frame memory 23 (i.e., the encoded/decoded data of frame Fo). Then, the frame F obtained from the encoder 3 is
-1 encoded data is locally decoded by the local decoder 4, and is rewritten and stored in the 1-frame memory 23.

[5] Frame F2, which is the frame after frame F1
(2n+1) is read from the frame memory 21, and the difference between this frame F2 and the motion compensation predicted value is calculated based on the data stored in the one frame memory 22 (that is, the encoded/decoded data of frame F+),
Perform interframe predictive coding. At this time, selector switch 7
is connected to the b(l! reference), and the changeover switch 24 is connected to the a side.

[6] Frame F, which is the frame before frame F-1
-2 is read from (2n+1) frame memory 21,
For this frame F-2, the difference between the motion compensation predicted value is calculated based on the data stored in the frame memory 23 in the space 1 (that is, the encoded/decoded data of frame F-1), and the difference between the frames is calculated. Predictive encoding. At this time, the changeover switch 7 is connected to the b side, and the changeover switch 24 is connected to the b side.

As mentioned above, the 5
Among frames F-2, F-1, Fo, Fl, F2,
As shown in FIG. 3, first, the frame Fo in the middle of the five frames is intra-frame encoded, and then, using the encoded/decoded data of this frame Fo as a reference, the frames F-1° before and after it are encoded. Interframe predictive coding of Fl,
Furthermore, based on the encoded/decoded data of frames F-1, Fo, F+, frames F-2, F
2 is subjected to interframe predictive coding.

In addition, the next five frames F3 to be encoded among the continuous frames. F4. F5. As for FB and F7, in the same way as above, first, frame F in the middle order of the 5 frames is
5 is intra-frame encoded, and then the frames before and after it are subjected to inter-frame predictive encoding.

Thereafter, the same encoding as above is performed sequentially for every five frames.

Note that the encoding is not limited to every 5 frames as described above, but also 2n+1 such as every 7 frames or 9 frames.
When encoding each frame at once, the same encoding procedure as above is used.

As described above, in the present invention, by performing intra-frame encoding every 2n+1 frames, the entire screen of one frame is refreshed at once, and based on the encoded/decoded data of the intra-frame encoded frame, By performing interframe predictive coding on the previous and subsequent frames, interframe prediction is performed in both forward and backward directions. Therefore, in conventional technology, interframe predictive coding was performed on frames that are up to 2n frames apart. However, interframe predictive coding is performed for frames that are separated by a maximum of 9 frames, and the interframe distance can be halved and the numerator error can be reduced.

(Effects of the Invention) As described above, in the method of the present invention, the interframe distance for interframe encoding can be halved compared to the conventional technique, and the numerator i! 1! It can reduce errors and solve the problems of conventional technology without significantly increasing the amount of code. It is especially useful when applied to recording on storage media, enables random access and visual search, and allows scene changes and It is also possible to encode general moving images that involve movement with high efficiency.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of the high-efficiency encoding method according to the present invention, FIGS. 2 and 3 are diagrams for explaining the encoding of the present invention, and FIG. 4 is a conventional A block diagram showing the configuration of an example of a high-efficiency encoding method, Figures 5 and 6 are diagrams for explaining the conventional encoding method, and Figure 7 shows the configuration of the encoder shown in Figure 4. A detailed diagram,
FIG. 8 and FIG. 9 are diagrams for explaining encoding. 1... Image signal input terminal, 2... Prediction signal subtractor,
3... Encoder, 4... Local decoder, 6... Motion compensation circuit, 7.24... Changeover switch, 8-... Transmission encoder, 9... Buffer memory, 10 ...Intra-frame/inter-frame coding switching υIt reference circuit, reference...data output terminal, 12...encoding control circuit, 13...8
X8DCT circuit, 14... quantizer for each coefficient, 21...
- (2n+1) frame memory, 22, 23...1 frame memory. Patent Applicant: Japan Victor Co., Ltd. Representative Kakiki
Kunio

Claims (1)

[Claims] Continuous frames of image signals that are continuously input are 2n+1.
This is a high-efficiency encoding method that collectively encodes each frame [n is an integer of 1 or more], and the 2n+ to be encoded
1 frame in order F-n, F-n+1,..., F-2,
F-1, F_0, F_1, F_2,..., F_n_-_1
. Based on the encoded/decoded data of frames F_0, F-1, F_1, frames F-2, F_2 before and after that
means for interframe encoding the frames F_0, F-1, F_1, ..., F- in the following order:
A high-efficiency encoding method comprising means for interframe encoding frames F-n and F_n before and after the encoded/decoded data of n+1 and F_n_-_1 as a reference.