JP2897649B2 - Motion compensated predictive coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-efficiency coding apparatus for digitizing an image with a smaller amount of information using inter-frame or inter-field prediction in a system for recording, reproducing and displaying image information. In particular, the present invention relates to a device that performs a motion compensation process.

[0002]

[Prior art]

<Motion Compensated Prediction Coding Device> In high-efficiency coding of a moving image, inter-frame prediction is performed using a high correlation between frames or between fields (hereinafter, represented by frames). In particular, recently, motion-compensated inter-frame prediction in which inter-frame prediction is performed after spatially moving a frame signal used for prediction in accordance with the motion of an image has become common.

[0003] Such motion-compensated inter-frame prediction (hereinafter, referred to as
FIG. 4 shows an example of an encoding device that performs motion compensation prediction. The image signal input from the image input 1 is provided to a prediction subtractor 2 and a motion vector (hereinafter abbreviated as MV) detector 23. The prediction subtractor 2 subtracts the inter-frame prediction signal given from the motion compensator 24 from the input image signal,
The prediction residual signal is provided to DCT3. 8 in DCT3
Two-dimensional discrete cosine transform (D
CT), and the transformed prediction residual signal (DCT coefficient) is provided to the quantizer 4. In the quantizer 4, the signal is quantized with a predetermined step width, and a fixed-length code is given to the inverse quantizer 13 and the variable-length encoder 5.

[0004] In the variable-length encoder 5, the fixed-length code is subjected to variable-length encoding and becomes a compressed code, which is provided to a multiplexer 6. A specific method of variable-length coding is to convert a two-dimensional array of DCT into a one-dimensional array of signal strings by zigzag scanning, code the non-zero value with its value, and use the Huffman code to code its continuous number (run-length) with 0. Become The multiplexer 6 multiplexes the code of the prediction residual and the code of the MV information supplied from the MV encoder 15 and outputs the multiplexed code from the code output 7 as a code string.

On the other hand, in the inverse quantizer 13, the fixed-length code is replaced with a representative value, which is supplied to the inverse DCT 12. Reverse D
In the CT 12, the inverse transform of the DCT 3 is performed, and the reproduced prediction residual signal is supplied to the adder 11. The adder 11 adds the inter-frame prediction signal supplied from the motion compensator 24 to the prediction residual signal, forms a reproduced image signal, and supplies the reproduced image signal to the image memory 10. The image memory 10 stores an image used for inter-frame prediction, and stores the MV as needed.
It is provided to a detector 23 and a motion compensator 24.

The MV detector 23 calculates the MV between the input image signal and the reproduced image signal stored in the image memory 10 and used for prediction. The obtained MV is the motion compensator 24 and M
It is provided to the V encoder 15. The MV encoder 15 encodes information of the MV value required for performing motion compensation in the decoding device, and supplies the code to the multiplexer 6. Motion compensator 24
Converts the reproduced image signal stored in the image memory 10 into an MV
It is moved spatially according to the value, and given to the prediction subtractor 2 and the adder 11 as an inter-frame prediction signal.

The operation of the MV accuracy and motion compensator 24 is as follows.
In order to perform the same motion compensation processing as that of the decoding device (that is, the accuracy on the decoding device side), it is assumed that transmission accuracy (that is, transmission standard) is matched. As a transmission standard, an international standard such as MPEG is generally used, and in this case, the MV precision is in units of 1/2 pixel. In the MV detection with 1/2 pixel accuracy, the MV is obtained once with the pixel accuracy, and only the MV around the MV is 1
Recalculate again with / 2 pixel accuracy. In addition, a method of forming a pixel at a half pixel position in the motion compensator 24 is also defined, and the pixel is formed by simple addition (Bi-Linear) of adjacent pixels.
Such MV detection and motion compensation processing are performed in units of 16 × 16 pixel macroblocks in which four DCT blocks are connected.

In the MV detection, if the reproduced image signal is degraded, an appropriate MV may not be obtained. Therefore, the input image signal of the same frame used for prediction is stored in another image memory. May be used instead of the reproduced image signal.

<Decoding Apparatus> Next, a decoding apparatus corresponding to the encoding apparatus shown in FIG. 4 will be described. FIG. 5 is a configuration diagram showing an example of the decoding device. The image data given from the code input 51 is demultiplexed by the demultiplexer 52 into a code for the prediction difference and a code for the MV information. The code for the prediction difference is sent to the variable-length decoder 53, and the code for the MV value is MV decoded. To the container 55. In the variable length decoder 53, the variable length code is returned to the fixed length, and the obtained fixed length code is supplied to the inverse quantizer 13. The MV decoder 55 decodes the MV information, and supplies the obtained MV value to the motion compensator 24.

In the inverse quantizer 13, a representative quantized value corresponding to the fixed-length code is obtained and supplied to the inverse DCT 12. In the inverse DCT 12, the inverse transform of the DCT 3 in FIG. 4 is performed, and the reproduced prediction residual signal is added to the adder 11.
Given to. The adder 11 adds the inter-frame prediction signal supplied from the motion compensator 24, outputs a reproduced image signal from the image output 54, and outputs the image signal from the image memory 1
0 is given. The reproduced image used for the inter-frame prediction is stored in the image memory 10 and output to the motion compensator 24 as needed. The motion compensator 24 converts the reproduced image to MV
Motion compensation is performed according to the value to generate an inter-frame prediction signal, which is provided to the adder 11. The same MV value as that of the encoding device is used, and the inter-frame prediction signal becomes the same as that of the encoding device.

[0011]

In the motion compensation prediction coding, a prediction residual component is coded so that a decoded image is closer to an input signal. The prediction residual component includes an error due to an accuracy limit that occurs because the motion compensation processing is not always ideal, in addition to an essential image change in which a picture itself changes for each frame. Since the essential change in the image gradually changes from frame to frame, by faithfully encoding, the prediction error of the next frame can be reduced. However, since the error due to the accuracy limit of the motion compensation process changes for each frame, even if the encoding is performed, only the fidelity of the frame is improved and does not contribute to the prediction of the next frame. This is because an essential image change has a correlation between frames, but an error due to an accuracy limit has no correlation between frames.

The present invention has been made in view of the above points. By performing motion compensation with higher accuracy than the transmission accuracy, the error due to the accuracy limit of the motion compensation process is not coded, and the An object of the present invention is to provide a motion-compensated prediction encoding device that improves the next inter-frame prediction efficiency by encoding a change in an image more faithfully, thereby improving the overall encoding efficiency.

[0013]

According to the present invention, a motion vector between an input image signal and an image signal used for prediction is obtained with higher accuracy than transmission accuracy when a moving image is subjected to inter-picture prediction coding. Motion compensation that performs inter-frame prediction coding using a signal obtained by motion-compensating an image signal used for prediction according to a motion vector, and rounds the obtained motion vector to transmission accuracy, and encodes and transmits a motion vector value that has reached transmission accuracy. It is a prediction encoding device.

In addition, when a moving image is subjected to inter-picture prediction coding, a motion vector between an input image signal and an image signal used for prediction is obtained with transmission accuracy, and the image signal used for prediction is motion-compensated according to the motion vector. Obtain an inter-frame prediction signal of transmission accuracy, obtain a fine motion vector with higher accuracy than the transmission accuracy between the inter-frame prediction signal and the input image signal, and obtain a high-precision prediction signal or input image signal according to the fine motion vector. This is a motion-compensated predictive coding apparatus that performs inter-frame predictive coding using a signal that has been motion-compensated in (1) and uses a predicted signal of transmission accuracy in local decoding corresponding to this coding.

[0015]

In the coding apparatus according to the present invention, since motion compensation is performed with higher accuracy than transmission accuracy, an error due to an accuracy limit is reduced. Therefore, the code conventionally used for encoding the error is no longer needed, and the code can be used for encoding the original change of the image. However, since motion compensation on the decoding side can be performed with the same precision as the transmission precision, errors due to the precision limit exist as much as in the past, and since they are not encoded, the fidelity is poor.

As a result, in the prediction residual components, essential image changes are more faithfully encoded, and errors due to accuracy limits are not faithfully encoded. The error due to the accuracy limit does not affect the inter-frame prediction of the next image, while the fact that the essential image changes are more faithfully encoded contributes to the improvement of the prediction efficiency, and the overall prediction efficiency is reduced. Be improved.

[0017]

【Example】

<Encoder 1> FIG. 1 is a block diagram showing a first embodiment of the encoder. The same portions as those in the conventional example of FIG. 4 are denoted by the same reference numerals. 4 differs from the conventional example of FIG. 4 in that a fine MV detector 8 and a fine motion compensator 9 are provided instead of the MV detector 23 and the motion compensator 24, and an MV value rounder 14 is present. In FIG. 1, the operation different from the conventional example of FIG. 4 is the inter-frame prediction processing, and the encoding operation of the prediction residual is the same as the conventional example.

The image signal input from the image input 1 is subtracted by the prediction subtractor 2 from the inter-frame prediction signal supplied from the fine motion compensator 9 to become a prediction residual signal, which is supplied to the DCT 3. DCT 3, quantizer 4, variable length encoder 5, multiplexer 6, inverse quantizer 13, inverse DCT 12, adder 1
1. The operation of the image memory 10 is the same as that of the conventional example.

On the other hand, the fine MV detector 8 and the fine motion compensator 9
Is basically the same as the conventional MV detector 23,
It is the same as the motion compensator 24, but differs in its accuracy and the like. The accuracy of the motion vector (MV) is 1/2 pixel accuracy which is the transmission accuracy (the accuracy of the transmission standard) in the conventional example.
In this embodiment, 1/6 pixel accuracy, which is higher than the transmission accuracy (transmission standard), is used. FIG. 6 shows this state. 1/6
In the MV detection with the pixel accuracy, the MV is obtained once with the pixel accuracy, and the MV around the MV is obtained again with the 1/6 pixel accuracy. The high-precision MV information obtained by the fine MV detector 8 is provided to the fine motion compensator 9 and the MV rounder 14.

In the fine motion compensator 9, the precision of the MV is made finer in accordance with the fine MV detector 8, and the method of forming pixels at positions other than the original pixels is also improved. In particular,
As shown in FIG. 7, the conventional example has a Bi
-Linear (simple quadratic linear interpolation) ((A) in the figure), and is created by high-precision (higher-order) resampling processing using peripheral pixels ((B) in the figure). ).

The MV rounder 14 discards (rounds) lower information in order to match a high-precision motion vector value with the transmission accuracy (transmission standard) and lowers the accuracy. , MV value encoder 15. Specifically, nine MVs with １ ／ pixel accuracy surrounded by a broken line in FIG. 6 are regarded as the same MV with 画素 pixel accuracy. The MV value encoder 15 encodes the MV value as in the conventional example.

In this embodiment, the motion compensation operation differs between the encoding device and the decoding device. Since inter-frame prediction is a cyclic operation, differences between the encoding device and the decoding device are cumulative. Therefore, it is necessary to prevent the error due to the rounding of the MV value from being biased to one side, and the precision is set to 1/6 pixel which is an even multiple of the transmission precision.
When 1/6 pixel precision is rounded to 1/2 pixel precision, 1
Each of the even number of 1/6 pixel points that are between the / 2 pixel precision is divided into two to obtain 1/2 pixel precision on both sides, and the error is balanced. Since this error has no correlation between frames, even if it accumulates, no major problem occurs.

The motion compensation method includes bidirectional prediction (B-pict) used in MPEG and the like in addition to unidirectional prediction.
ure).

<Encoding device 2> FIG. 2 shows a second example of the encoding device.
FIG. 3 is a configuration diagram showing an example of the embodiment. The same portions as those in the conventional example of FIG. 4 are denoted by the same reference numerals. 4 compared to the conventional example of FIG.
The configuration differs in that a V detector 21 and a minute motion compensator 22 are provided. In the figure, only the micro MV detector 21 and the micro motion compensator 22 are different from the conventional example in operation.
The configuration and operation of the other parts including the motion compensator 24 are the same as in the conventional example.

The image signal input from the image input 1 is given to the prediction subtractor 2, the minute MV detector 21, and the MV detector 23. The prediction subtracter 2 subtracts an inter-frame prediction image on which high-precision motion compensation has been performed from the input image, and supplies a prediction residual to the DCT 3. DCT 3, quantizer 4, variable length encoder 5, multiplexer 6, MV encoder 15, inverse quantizer 13, inverse DCT 12, adder 11, image memory 1
0, the operation of the MV detector 23 is the same as in the conventional example. The operation of the motion compensator 24 is the same as that of the conventional example, but the inter-frame prediction signal from the motion compensator 24 is given to the minute MV detector 21 and the minute motion compensator 22 instead of the prediction subtractor 2.

The minute MV detector 21 obtains a minute movement deviation between the inter-frame prediction signal and the input signal as a minute MV. Here, since the inter-frame prediction signal has already been motion-compensated with half-pixel accuracy of the transmission accuracy, the minute MV becomes finer than that, and the accuracy is made finer than the transmission accuracy. Assuming that the specific accuracy is 1/6 pixel accuracy as in the first embodiment, the MV is -2/6, -1 in the horizontal and vertical directions.
Five values of / 6, 0, + ／, +2/6 are set, and MV is selected from 25 types of 5 × 5 vectors.

The minute motion compensator 22 is a minute MV detector 21
The motion compensation is performed by higher-order resampling with the precision of the small MV given by As described above, in the present embodiment, the fine MV detector 8 and the fine motion compensator 9 of the first embodiment are used.
In the processing performed in step (1), the processing part having higher precision than the transmission precision is separated and applied only to the predictive subtractor 2. In this case, since the inter-frame prediction signal from the motion compensator 24 is input to the adder 11, the locally decoded image becomes the same as the conventional example, and no error occurs between the encoding device and the decoding device. .

<Encoding device 3> FIG. 3 shows a third example of the encoding device.
FIG. 3 is a configuration diagram showing an example of the embodiment. The same parts as those in the second embodiment of FIG. 2 are denoted by the same reference numerals. In FIG. 3, the insertion position of the minute motion compensator 22 differs from that of the second embodiment in terms of configuration.

An image signal input from the image input 1 is supplied to a minute motion compensator 22, a minute MV detector 21, and an MV detector 23. In the minute motion compensator 22, the minute MV detector 2
According to the minute MV value given from 1, minute motion compensation of one pixel or less is performed in the same manner as in the case of FIG. 2, and the result is given to the prediction subtractor 2. In the prediction subtractor 2, instead of the input image as it is, the inter-frame prediction image is subtracted from the input image on which the minute motion compensation has been performed, and the prediction residual is DC.
Provided to T3. DCT 3, quantizer 4, variable length encoder 5, multiplexer 6, MV encoder 15, inverse quantizer 1
3, inverse DCT 12, adder 11, image memory 10, MV
The operations of the detector 23 and the motion compensator 24 are the same as in the second embodiment.

The minute MV detector 21 is basically the same as the case of FIG. 2, but the input is reversed because the MV for moving the input image signal is calculated based on the prediction signal. In the present embodiment, similarly to the second embodiment, the locally decoded image is the same as the conventional example, and no error occurs between the encoding device and the decoding device. Furthermore, by applying the resampling process to both the prediction signal and the input signal, there is an advantage that the frequency characteristics can be balanced. This is related to the frequency characteristic degradation caused by the resampling process of the motion compensation, and occurs only in the prediction signal in the conventional example and the first embodiment.
This is because what is likely to cause a mismatch can be balanced by being applied to both the prediction signal and the input signal.

Further, since the aliasing distortion component which exists independently in each signal is suppressed due to the deterioration of the frequency characteristics of both signals, the error component which is not the motion of the original image is reduced. It is a suitable process.

<Decoding Device> The decoding device corresponding to the motion-compensated predictive coding device shown in FIGS. 1 to 3 is the same as the conventional example shown in FIG. 4, and does not necessarily require a new decoding device. Absent.

[0033]

According to the encoding apparatus of the present invention, motion compensation is performed with higher accuracy than the transmission accuracy (transmission standard). Therefore, in the prediction residual component, an essential image change is more faithfully encoded. The error due to the accuracy limit is not faithfully encoded.
Errors due to the accuracy limit do not affect the inter-frame prediction of the next image, while improving the inter-frame prediction by encoding the essential image changes more faithfully and improving the overall prediction efficiency I do. By improving the prediction efficiency, the prediction residual to be coded is reduced, and the coding efficiency is improved.

On the other hand, in each frame, an increase in the error due to the accuracy limit leads to a spatial displacement, but if the motion compensation accuracy of the transmission accuracy (transmission standard) is 1/2 pixel, it is visually apparent. Because it is less of a problem and the essential image change fidelity is improved, subjective image quality is also desirable. As a result, it is possible to obtain a required reproduction image quality with a smaller code amount.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a configuration example of an encoding device according to a first embodiment of the present invention.

FIG. 2 is a diagram illustrating a configuration example of an encoding device according to a second embodiment of the present invention.

FIG. 3 is a diagram illustrating a configuration example of an encoding device according to a third embodiment of the present invention.

FIG. 4 is a diagram illustrating a configuration example of a conventional encoding device.

FIG. 5 is a diagram illustrating a configuration example of a conventional decoding device.

FIG. 6 is a diagram showing a fine motion vector and a state of a rounding process thereof.

FIG. 7 is a diagram illustrating a state of a resampling process according to a conventional example and an embodiment.

[Explanation of symbols]

REFERENCE SIGNS LIST 1 image input, 2 predictive subtractor, 3 DCT, 4 quantizer, 5 variable length encoder, 6 multiplexer, 7 code output, 8 fine MV detector, 9 fine Motion compensator, 10 image memory, 11 adder, 12 inverse DCT, 13 inverse quantizer, 14 MV rounder, 15 MV encoder, 21
... minute MV detector, 22 ... minute motion compensator, 23 ... MV detector, 24 ... motion compensator, 51 ... code input, 52 ... demultiplexer, 53 ... variable length decoder, 54 ... image output, 55 ...
MV decoder.

Claims (3)

(57) [Claims] An apparatus for motion-compensated inter-picture predictive coding of a moving image, comprising: means for obtaining a motion vector between an input image signal and an image signal used for prediction with higher accuracy than transmission accuracy; Motion compensation means for performing motion compensation on an image signal used for prediction in accordance with the above, and obtaining a prediction signal with higher accuracy than transmission accuracy, and means for rounding the motion vector to transmission accuracy and encoding a motion vector value having reached transmission accuracy. And a predictive coding means for coding a residual obtained by subtracting the predictive signal from the input image signal. 2. An apparatus for motion-compensated inter-picture predictive coding of a moving image, comprising: means for obtaining a motion vector between an input image signal and a reproduced image signal used for prediction with transmission accuracy; A motion compensating unit that motion-compensates an image signal to be used and obtains a prediction signal of transmission accuracy; Means for determining, a motion compensation of the prediction signal of the transmission accuracy by resampling according to the micro motion vector, and a micro motion compensation means for obtaining a micro motion compensation prediction signal; and a residual obtained by subtracting the micro motion compensation prediction signal from the input image signal. A predictive coding unit for coding the difference, and a decoding process corresponding to the predictive coding unit performed by using the predicted signal of the transmission accuracy. Motion compensated predictive coding apparatus characterized by comprising a local decoding means for obtaining. 3. An apparatus for motion-compensated inter-picture predictive coding of a moving image, comprising: means for obtaining a motion vector between an input image signal and an image signal used for prediction; A motion compensating means for compensating and obtaining a prediction signal; and a means for obtaining a fine motion vector having a pixel interval or less between the input image signal and the prediction signal obtained by the motion compensating means with higher accuracy than the motion vector. A motion compensation means for re-sampling the input image signal in accordance with the motion vector to obtain a motion compensation signal; and a prediction encoding means for encoding a residual obtained by subtracting the prediction signal from the motion compensation signal. A motion-compensated predictive coding apparatus comprising: