JPH05308631A - Moving image encoder - Google Patents

Abstract

PURPOSE:To improve encode efficiency by effectively utilizing prediction performance using a low resolution locally decoded signal in the predictive encode of a high resolution image signal. CONSTITUTION:A moving image encoder which performs the predictive en-code of the high resolution image signal 11 and also, encodes a low resolution image signal obtained by passing the high resolution image signal through a down- sampling circuit 29 is provided with a prediction mode decision circuit 37 which renders decision considering a predictive error power when prediction signals 13, 16 are used in the predictive encode of the high resolution image signal 11 and the information quantity of added information such as motion vector information, etc., to select a prediction signal for the predictive encode of the high resolution image signal 11 from a high resolution prediction signal 13 obtained from a high resolution locally decoded signal 26 in which the encode result of the high resolution image signal 11 is decoded, and a low resolution prediction signal 16 obtained by passing a low resolution locally decoded signal 34 in which the encode result of the low resolution image signal is decoded through an up-sampling circuit 35.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture coding apparatus, and more particularly to a system for transmitting a moving picture through a line such as a TV conference and a TV telephone, and storing the moving picture on a storage medium such as an optical disk and a video tape. The present invention relates to a moving picture encoding device used in a system.

[0002]

2. Description of the Related Art Toward the start of a broadband network service in the future, currently, in the field of communication, a moving image having a picture quality equal to or higher than that of the current TV system is displayed in several Mbps to several tens of Mbps.
Standardization work on moving picture coding technology for transmission at a bit rate of about s is underway (H.26).
X). Similarly, in the fields of storage and broadcasting, the H.264 standard is also used. Two
The standardization work of the encoding method based on the image quality and the bit rate similar to that of 6X is in progress (MPEG2 for storage system, CMTT / 2 for broadcasting system). Then, in these standardization work, it is judged that it is advantageous to popularize the system in common as much as possible, and therefore work is being carried out while exchanging information with each other.

On the other hand, in the communication system and the storage system, the H.264 standard, which is a standard method of moving image coding, has already been targeted at a place where image quality and bit rate are lower. 261 and MPEG1
However, hardware is beginning to be supplied.
Therefore, it is pointed out that one of the problems is that the new standard system with higher image quality has the interoperability (compatibility) with these existing standard systems. Compatibility includes a forward method that enables a new decoder to decode a bit string created by an existing encoder.
There are two types of compatibility, backward compatibility that enables a decoder of the existing system to decode a part of the bit string created by the encoder of the new system. Considering that the decoder of the existing scheme is already defined while the decoder of the new scheme is undefined, backward compatibility is a stricter requirement for the scheme.

There are various possible backward compatibility schemes. Among them, as a method that can define compatibility as an option, include the locally decoded signal in the existing method as a candidate for the prediction signal to be used in the new method so that part of the coded bit string in the new method becomes the bit string in the existing method. A method of forming a coded bit string in a frame (called embedded coding using hierarchical coding)
Is proposed.

15 is a block diagram showing a schematic configuration of a moving picture coding apparatus based on this system, and an input image signal is input to a coding unit 1 based on a new system (for example, H.26X or MPEG2). Together with the existing method (for example, H.261 or MPE) via the down sampling circuit 2.
It is input to the encoding unit 3 based on G1). The encoded output from the encoding unit 1 is output as a higher layer bit string and is locally decoded by the local decoder 4. The encoded output from the encoding unit 3 is output as a lower layer bit string and is locally decoded by the local decoder 5.

The locally decoded signal from the local decoder 4 is fed back to the encoding unit 1 as a prediction signal, the locally decoded signal from the local decoder 5 is fed back to the encoding unit 3 as a prediction signal, and the upsampling circuit 6 is used. After that, it is also supplied to the encoding unit 1 as a prediction signal. The encoding unit 1 performs predictive encoding by selectively using either the locally decoded signal from the local decoder 4 or the prediction signal from the upsampling circuit 6.

[0007] Reference 1: Chapter 6 of "International Standard for Multimedia Coding" (published by Maruzen) describes one of the existing methods, M.
PEG1 is described, of which, in the section 6.3 Encoding Algorithm, Fig. 6.2 is used to describe I picture (intra prediction image), P picture (forward prediction image), B picture (forward and backward bidirectional prediction). Inter-picture prediction structure consisting of (images) is described. It is expected that the new method will have an inter-picture prediction structure similar to that of the existing method. However, in the existing system such as MPEG1, since the image signal has a non-interlaced structure, “image = picture” is a “frame”, whereas in the new system, this can be a “field” having an interlaced structure. Predictive structure is slightly complicated.

By the way, in the above-mentioned embedded coding using the hierarchical coding, the prediction using the locally decoded signal in the existing system is added as an option, and this prediction is better than the prediction from the locally decoded signal in the new system. Only chosen at times.
That is, in FIG. 15, the better one of the locally decoded signal from the local decoder 4 and the locally decoded signal input from the local decoder 5 through the upsampling circuit 6 is selectively used as a prediction signal in the encoding unit 1. It
Therefore, it is said that there is no deterioration in coding efficiency due to the inclusion of prediction from the locally decoded signal in the latter existing method. However, conventionally, in the mode determination for selecting the prediction signal used in the encoding unit 1, only the prediction error power is taken into consideration and the prediction signal with the smaller prediction error power is selected. The prediction using the signal is not fully utilized. That is, the amount of information generated when the prediction error signal is coded is reduced, but the amount of information generated as a whole coded output is not necessarily reduced, which hinders improvement in coding efficiency.

Further, among the pictures classified by the prediction structure, the prediction from the existing method is effectively selected for the I picture, whereas the locally decoded signal corresponding to the coding result of the existing method is selected for the P picture and the B picture. There is also the problem that the predictions from the above do not work very well, and the selection rate is extremely low. As one of the causes of this, it is known that there is a factor that depends on a method of creating an image encoded by the existing method.

Further, in the conventional embedded coding using hierarchical coding, when a low-resolution image is created by dropping one field, the field not used to create the low-resolution image is the low-resolution prediction. Since it was not well predicted by the signal, it reduced the overall prediction efficiency.

[0011]

As described above, in the conventional embedded coding using the hierarchical coding, only the prediction error power is considered in the mode determination for selecting the prediction signal in the coding of the new method. Therefore, the prediction using the locally decoded signal in the existing system cannot be fully utilized, which hinders the improvement of the coding efficiency.

Further, in the conventional embedded coding using the hierarchical coding, among the pictures classified according to the prediction structure, the I-picture is selected with good prediction from the existing method, whereas the P-picture is selected. In the B picture, there is also a problem that the prediction from the locally decoded signal of the existing system is not so effective and the selection rate becomes very low.

Further, in the conventional embedded coding using hierarchical coding, when a low-resolution image is created by dropping one field, the field not used to create the low-resolution image is the low-resolution prediction. There is also a problem that the overall prediction efficiency is reduced because the signal is not well predicted.

The present invention solves the above-mentioned conventional problems and effectively utilizes the prediction capability using a low-resolution locally decoded signal for predictive coding of a high-resolution image signal to improve the coding efficiency. An object is to provide an image coding device.

[0015]

To solve the above problems, the present invention predictively encodes a high resolution image signal and encodes a low resolution image signal obtained by converting the high resolution image signal. In a moving picture coding device, a high-resolution prediction signal obtained from a high-resolution local decoded signal obtained by decoding the coding result of a high-resolution image signal and a low-resolution locally decoded signal obtained by decoding the coding result of a low-resolution image signal are uploaded. From the low-resolution prediction signal obtained from the sampled signal, the high-resolution prediction signal and the low-resolution prediction signal are the prediction code of the high-resolution image signal It is characterized in that the prediction error power and the amount of additional information such as motion vector information when used for each conversion are taken into consideration.

Further, the present invention has a mode for predictively encoding a high resolution prediction error signal, which is a difference between the predicted high resolution image signal and the high resolution prediction signal, by using the low resolution prediction error signal. Motion-compensated high-resolution prediction signal is generated using the motion vector between the high-resolution image signal and the high-resolution image signal, and the low-resolution locally decoded signal obtained by decoding the coding result of the low-resolution image signal A low resolution prediction error signal is obtained by generating a compensated low resolution prediction signal and taking the difference between the low resolution prediction signal and the low resolution local decoded signal.

Further, according to the present invention, when the low resolution prediction signal is used for predictively encoding the high resolution signal, a means for switching whether or not to use the prediction by the low resolution signal for each field forming the high resolution signal is provided. To have. Further, it comprises means for shifting the low-resolution locally decoded signal when generating the low-resolution prediction signal, and means for generating the shift amount based on the motion vector of the high-resolution image signal.

[0018]

As described above, when the prediction signal of the high resolution image signal is selected, not only the prediction error power but also the information amount of the additional information such as the motion vector information is taken into consideration. The prediction signal using the locally decoded signal is more effectively selected.

Further, since the motion compensation of the locally decoded signal in the new system alone cannot be sufficiently matched, the remaining high resolution prediction residual signal is further motion compensated predictive coded by using the low resolution locally decoded signal according to the existing system. By
The low-resolution locally decoded signal according to the existing method is effectively used.

As described above, in the present invention, it is more effective to use the low-resolution local decoded signal in the existing system as a prediction signal not only for I pictures but also for P pictures and B pictures, and the overall coding efficiency is improved. To do.

Further, by switching whether or not the prediction by the low resolution image signal is used for each field, the prediction from the low resolution image is poor in the conventional one field prediction, so that the prediction from the low resolution image is performed in both fields. Even if it is not selected, this is selected for the field that is better predicted from the low resolution image, and the overall coding efficiency can be improved. Further, by introducing the shift, it is possible to improve the prediction of the field in which the prediction is bad, and further improve the coding efficiency.

[0022]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing a first embodiment of a moving picture coding apparatus according to the present invention. In this embodiment, the image encoded by the new method is an interlaced image,
The case where an image encoded by the existing method is a non-interlaced image created by removing one field from the interlaced image will be described.

In FIG. 1, a high resolution image signal is input to the terminal 10 as an input image signal 11. This input image signal 11 is input to the subtractor 12 and the high resolution prediction signal 1
3 is taken to generate the high resolution prediction error signal 14. The first selection switch 15 is the prediction mode determination circuit 3
7, and selects any one of the input image signal 11, the high resolution prediction error signal 14 and the low resolution prediction signal 16.

The signal selected by the selection switch 15 is D
DCT coding is performed by a CT (discrete cosine transform) circuit 17. The DCT coefficient data obtained by the DCT circuit 17 is quantized by the quantization circuit 18. The signal quantized by the quantizing circuit 18 is branched into two, one of which is subjected to variable length coding by the variable length coding circuit 19 and further passed through the buffer 20 from the output terminal 21 to the transmission system or the storage system at a predetermined bit rate. Sent out. The subtracter 12, the DCT circuit 17, the quantization circuit 18, and the variable length coding circuit 19 constitute a first coding means.

The other of the signals quantized by the quantization circuit 18 and branched into two is the inverse quantization circuit 22 and the inverse DCT circuit 2.
3 sequentially receives a process reverse to the process in the quantizing circuit 18 and the DCT circuit 17, and then the adder 24 adds the signal to the signal selected by the second selection switch 25 to obtain a high-resolution locally decoded signal. 26 is generated. The inverse quantization circuit 22, the inverse DCT circuit 23, and the adder 24 constitute a first local decoding means.

The selection switch 25 is the prediction mode determination circuit 3
7 is controlled in cooperation with the selection switch 15, and is “0” when the selection switch 15 selects the input image signal 11, and the high resolution prediction signal 13 and the low resolution prediction signal when the high resolution prediction error signal 14 is selected. When 16 is selected, the low resolution prediction signal 16 is selected.

The high resolution local decoded signal 26 output from the adder 24 is written in a frame memory 27 having a storage capacity for a plurality of frames. The output of the frame memory 27 is input to the predictor 28 and used to create the high resolution prediction signal 13. The predictor 28 detects a motion vector between the input image signal 11 which is a high resolution image signal and the high resolution local decoded signal from the frame memory 27, and uses this motion vector to perform motion compensation between frames (between fields). By performing the prediction, the high resolution prediction signal 13 is generated and the motion vector information 38 is output. The prediction signal thus obtained includes three types of forward prediction, backward prediction, and bidirectional prediction as described in the above-mentioned document (1), and the optimum one is selected from these.

On the other hand, the input image signal 11 is also input to a down-sampling circuit 29 which is a means for converting a high resolution image signal into a low resolution image signal, and is down-sampled there, such as CIF or SIF. It is converted to a low resolution image signal in the format. As an example of processing in the downsampling circuit 29, FIG.
The state of down-sampling by dropping a single field is shown in (a). The low resolution image signal indicated by the small rectangle is discarded after discarding the even (Even) field of the high resolution image signal indicated by the horizontally long rectangle and passing the odd number (Odd) field through the band-limiting filter for aliasing removal. It is created by subsampling.

The low resolution image signal of the CIF or SIF format obtained by the downsampling circuit 29 is an existing method (for example, H.2) which is the second encoding means.
61 or MPEG1), and the data is encoded by the encoder 30 and transmitted through the buffer 31 from the output terminal 32 to the transmission system or the storage system at a predetermined bit rate. The encoded output from the encoding unit 30 is locally decoded by the local decoder 33 based on the existing method (H261 or MPEG1) which is the second local decoding means. The low-resolution local decoded signal 34 output from the local decoder 33 is input to create a prediction signal used in the predictive coding in the coding unit 30, and is up-sampled by the up-sampling circuit 35 to generate the second signal. It is also used by the predictor 36 to create the low resolution prediction signal 16.

The predictor 36 detects a motion vector between the input image signal 11 which is a high resolution image signal and the low resolution local decoded signal from the upsampling circuit 35 which is a low resolution image signal, and obtains this motion vector. By applying motion compensation to the low resolution locally decoded signal using
The low resolution prediction signal 16 is generated and the motion vector information 39 is output.

With reference to FIG. 2 (b), the effect of applying motion compensation to the low resolution local decoded signal in the predictor 36 will be described. The low resolution image signal generated by down-sampling as shown in FIG. 2A is passed through the encoding unit 30 and the local decoder 33 to obtain the low resolution local decoded signal 3
4 is generated and up-sampled by the up-sampling circuit 35 as shown in FIG.
When it is used for field prediction, it is possible to perform good prediction because it matches in position with the original Odd field, but when it is used for Even field prediction, it may move due to mutual time lag. It cannot be predicted well for areas with However, as shown in FIG. 4, if the low-resolution local decoded signal is shifted by the shift amount corresponding to the movement within the field period, the position is matched with the Odd field. Can be used to make correct predictions. Therefore, the motion compensation of the predictor 36 in this case may be basically performed only when the EVen field is predicted. However, if the motion compensation is performed, the prediction error in the Odd field is also reduced. Therefore, the motion compensation may be performed for both the Odd field and the Even field.

The prediction mode judgment circuit 37 is a circuit for judging the prediction mode when the input image signal 11 is encoded, and controls the selection switches 15 and 25 according to the judgment result. The prediction mode determination circuit 37 also outputs a prediction mode signal 40 indicating which prediction mode is selected.
The prediction mode signal 40 is input to the variable length coding circuit 19, and the quantized signal from the quantization circuit 18 and the motion vector information 38, 3 output from the predictors 28 and 36.
Variable length coded together with 9.

FIG. 3 shows a function for determining which of the high resolution prediction signal 13 from the predictor 28 and the low resolution prediction signal 16 from the predictor 36 is selected in the prediction mode determination circuit 37. When using the high resolution prediction signal 13,
The information of one or two motion vectors must be transmitted as the motion vector information 38, but when the low resolution prediction signal 16 is used, only the information of 0 or one motion vector is transmitted as the motion vector information 39. It's done.

The motion vector information 38 and 39 are variable length coded by the variable length coding circuit 19 as described above and transmitted. The number of bits (code amount of variable length code) used for transmission of the motion vector information 38 and 39 is estimated, and the difference between them is set as MVbit diff. Here, it is considered that the prediction error signal of the low resolution image signal is quantized again by this number of bits.

MSE is the prediction error power of the prediction error signal 14 when the high resolution prediction signal 13 from the predictor 28 is used.
_{Let pred1} be the MSE of the prediction error power of the prediction error signal 14 when the low resolution prediction signal 16 from the predictor 38 is used.
_{As pred2} , MSE _pred1 and MSE _pred2 x 2
^-Judge the magnitude relationship of ^{2MVbit diff} . As a result of this judgment,

[0037]

[ _Expression 1] When MSE _pred1 <MSE _pred2 × 2 ^{-2MVbit diff} (1), the high resolution prediction signal 13 is selected by the selection switches 15 and 25.
Select

[0038]

[ _Expression 2] When MSE _pred1 > MSE _pred2 × 2 ^{-2MVbit diff} (2), the low resolution prediction signal 16 is selected by the selection switches 15 and 25.
If is selected, the optimum predictive coding of the input image signal 11 can be performed while suppressing the amount of information to be transmitted. This is particularly effective when it is not necessary to transmit the motion vector information for the low-resolution prediction signal 16, for example, when the low-resolution prediction signal 16 is created and encoded from field-superimposed frames.

As the decision function α in FIG. 3, for example, α = 2 ^{−2 MVbit diff} , that is, a function whose slope changes with MVbit diff may be used. This function can also be used in the comparison between prediction signals of high resolution images.

FIG. 4 is a block diagram showing a concrete example of the configuration of the prediction mode determination circuit 37 in FIG. 1 based on the above principle. Input image signal (high resolution image signal) 11
And the high resolution prediction signal 13 from the predictor 28 is input to the high resolution prediction mode determination circuit 41, where one of the high resolution prediction signals 13 is selected. The difference between the selected high resolution prediction signal and the input image signal 11 is taken by the subtractor 44 to generate a high resolution prediction error signal. on the other hand,
The low resolution prediction signal 16 from the predictor 36 is input to the subtractor 45, where the difference from the input image signal 11 is taken and a low resolution prediction error signal is generated. In addition, the predictor 28,
The motion vector information 36, 39 from 36 is input to the motion vector information amount estimating circuits 42, 43, and the motion vector information amount, that is, the generated information amount from the variable length coding circuit 19 corresponding to the motion vector information 38, 39 Presumed.
The subtractor 46 obtains the difference in the estimated motion vector information amount, that is, the MVbit diff.

In the operation determination circuit 47, the prediction error power MSE of the high resolution prediction error signal and the low resolution prediction error signal
_{In addition to obtaining pred1} and MSE _pred2 , the determination shown in equations (1) and (2) is performed from the difference MVbit diff between the two motion vector information amounts using the determination function shown in FIG. 3, and according to the determination result While controlling the selection switches 15 and 25 of FIG. 1, the prediction mode determination signal 40 is output.

As described above, according to the embodiment shown in FIG. 1, not only the prediction error power but also the additional information such as the motion vector information is taken into consideration for each of the high resolution prediction error signal and the low resolution prediction error signal. According to the conditions, a prediction signal used for predictive coding of an input image signal which is a high resolution image signal is selected from the high resolution prediction signal and the low resolution prediction signal, and therefore there is an advantage that the coding efficiency is further improved.

The above-described embodiment can be implemented with various modifications as follows.

(1) In FIG. 2A, the low-resolution image signal by dropping one field is created from the Odd field which precedes in time, but it may be created from the Even field. As a result, regarding the prediction of the Odd field, the prediction from the low-resolution local decoded signal becomes the backward prediction, so that further improvement in prediction efficiency can be expected.

(2) The low resolution prediction signal for the prediction of the Even field can be obtained not by the motion compensation prediction but by the spatiotemporal interpolation as shown in FIG.
That is, each pixel of the low resolution prediction signal of the Even field is a pixel of an Odd field adjacent to the low resolution local decoded signal in the time axis direction t, and is also a signal of a pixel adjacent in the spatial axis direction (vertical direction) V. Interpolate by. In this case, since it is not necessary to transmit the motion vector information corresponding to the low resolution image signal, the coding efficiency is improved.

(3) The prediction using the low resolution local decoded signal may be prohibited for the Even field and may be performed only for the Odd field.

(4) Even when motion compensation is performed on the Even field, it is possible to estimate the motion vector from the past motion vector or the motion vector at the time of low-resolution image coding and use the motion vector. No transmission is needed.

(5) It is possible to add to the predictor 28 a function of generating a prediction signal using both the high resolution local decoded signal and the low resolution local decoded signal. In this case, as shown in FIG. 6, a high-pass filter that allows the high-resolution signal (high-resolution local decoded signal) and the low-resolution signal (low-resolution local decoded signal) to have complementary pass bands. After passing through 1 and the low pass filter 2, the predicted signal is formed by adding them. Quantization noise is reduced in the prediction signal thus obtained.

(6) Input image signal 1 in the predictor 28
1 (high-resolution image signal) and the high-resolution local decoded signal, when detecting the motion vector, the motion vectors of the low-resolution image signals obtained by the encoding unit 30 are expanded around the motion vector corresponding to the high-resolution image. By limiting the search range, the amount of calculation can be reduced.

Next, a second embodiment of the moving picture coding apparatus according to the present invention will be described with reference to FIG. In this embodiment, four selection switches 51 for selecting the prediction mode,
54, 55, and 56 are provided, and these are controlled by the prediction mode determination circuit 37. The prediction mode determination circuit 37 may have the configuration shown in FIG. 4 or the same configuration as the conventional one. The selection switch 51 selects either the input image signal 11 or the high resolution prediction error signal 14 output from the subtractor 14. The selection switch 54 selects either the signal 52 selected by the selection switch 51 or the low resolution prediction error signal output from the subtractor 53 (the difference signal between the signal 52 and the low resolution prediction signal 16). The signal selected by the selection switch 54 is input to the DCT circuit 17.

The selection switch 55 is the prediction mode determination circuit 3
7 is controlled in cooperation with the selection switch 51, and is “0” when the selection switch 51 selects the input image signal 11.
When selecting the high resolution prediction error signal 14, the high resolution prediction signal 13 is selected. Similarly, the selection switch 56 is controlled by the prediction mode determination circuit 37 in association with the selection switch 54, and the selection switch 54 is operated by the selection switch 5.
When selecting a signal from 1, "0" is selected, and when selecting a low resolution prediction error signal, a low resolution prediction signal 16 is selected.

The signal selected by the selection switch 55 is added to the signal from the inverse DCT circuit 23 in the adder 24a, and the signal selected by the selection switch 56 is added by the adder 24b.
In, the local decoded signal 26 is generated by adding the output from the adder 24a.

The predictor 28 detects the motion vector between the input image signal 11 which is a high resolution image signal and the high resolution local decoded signal from the frame memory 27, as in the embodiment of FIG. By performing motion compensation inter-frame (field-to-field) prediction using a vector, a high-resolution prediction signal 13 is generated and motion vector information 3
8 is output. The prediction signal obtained in this way includes three types of prediction including forward prediction, backward prediction, and bidirectional prediction.
The optimum one is selected from these.

Motion vector information 3 obtained by the predictor 28
8 is sent to the variable length coding circuit 19 and also to the predictor 36. The predictor 36 is the local decoder 3
The predictor 2 is accessed by accessing the frame memory 57 for storing a plurality of low-resolution locally decoded signals from
According to the motion vector information sent from the frame 8, the signal inputted from the frame memory 57 through the upsampling circuit 35 is motion-compensated, and the motion-compensated low resolution prediction error signal 16 is output.

FIG. 8 is a diagram showing a state of predictive coding of a high resolution image signal in this embodiment. The predictor 28 outputs the high-resolution predicted signal 16 obtained by performing motion compensation on the high-resolution image signal (high-resolution local decoded signal) at t = t ₁ , and the subtractor 12 outputs the high-resolution predicted signal 16. A high resolution prediction error signal 14 which is the difference between the resolution prediction signal 16 and the high resolution image signal 11 at t = t ₀ is output. On the other hand, from the predictor 36, a low-resolution predictive signal obtained by performing motion compensation with respect to the low-resolution image signal (low-resolution local decoded signal) at _{t = t 1, t = t} 0
The low-resolution prediction error signal, which is the difference from the low-resolution locally decoded signal in, is output as the low-resolution prediction signal 16.
Then, the low resolution prediction signal 16 is supplied to the subtractor 53 as a prediction signal for interframe (or interfield) predictive coding of the high resolution prediction error signal 14.

As described above, in the embodiment shown in FIG. 7, since the high resolution prediction error signal 14 has a mode in which the low resolution prediction error signal is further predictively coded using the prediction signal as the prediction signal, the locally decoded signal in the existing system is converted into the high resolution. It can be used more effectively in predictive coding of image signals, and has an advantage of improving coding efficiency.

Next, the above-described (3) prediction using a low-resolution local decoded signal is
Prohibit the field, and only the Odd field.

Now, the third embodiment of the present invention will be specifically described.

In this embodiment, an image coded by the new method is an interlaced image, and an image coded by the existing method is a non-interlaced image created by dropping one field from the interlaced image. The following description will be made on the assumption that the interlaced image is encoded by overlapping the frames.

FIG. 9 is a block diagram of this embodiment.
The same reference numerals are given to the same components as.

FIG. 10 is a diagram showing the relationship between encoded images of various sizes, and FIG. 11 is a diagram showing candidates for switching low-resolution prediction in this embodiment.

First, the input field image is field-merged by the field merging circuit 100 and treated as a frame image. The situation is shown in FIG. After that, it is divided into blocks by the blocking circuit 101 for encoding. After being converted into a frame image, the blocked data is Odd field data, one line at a time, as shown at the left end of FIGS.
The data of d appear alternately. (For the sake of simplicity, the case of 4 × 4 blocks is shown. ◯ indicates Oddfire.
ld data, Δ indicates Even field data. ) The predictor + prediction mode determiner 104 processes the input data in blocks into various inter-frame predictions (see images in the frame memory 27) from high-resolution images considered as candidates for prediction. A signal for prediction from the resolution image (refer to an image in the frame memory of the existing system local decoder 33 up-sampled by the up-sampling circuit 35) is created, an error from the input signal is taken, and a certain evaluation criterion The prediction mode that minimizes the error is selected. Prediction from a high-resolution image is not the gist of the present invention, so a detailed description thereof will be omitted, but it is, for example, a method currently being studied in MPEG2. Now, in the present invention, as shown in FIG.
A prediction image is created from a low resolution image for both the dd line and the Even line, and a prediction image is created from the low resolution image for the Odd line and the high resolution image for the Even line as shown in FIG. Can be switched. (The black circles in the figure indicate the low resolution prediction signals.) The predictor 104 is used to generate these signals.
First, the input signal is separated into the odd line and the even line, and the optimum one is obtained from all possible high resolution prediction signals for each. (Double triangle in the figure)
Next, the pixel at the position corresponding to the current coding block of the frame memory in the local decoder 33 of the existing method is taken out, up-sampled by the up-sampling circuit 35, and finally these are alternately merged for each line to obtain a prediction signal. And A prediction signal from only the low-resolution image is created by up-sampling the image in the existing frame memory.

Furthermore, matching is performed while shifting the low-resolution image in the upsampled image of the low-resolution image separately for the O and Δ pixels of the input image, and the optimum shift amount is searched for. On the other hand, a signal obtained by merging line by line alternately and a signal obtained by merging line by line with the prediction signal of the high resolution image are generated in the same manner.

In any case, three low resolution prediction signal candidates are created ((1) Odd: low resolution ev).
en: high resolution (2) Odd: high resolution even: low resolution (3) both Odd and even have low resolution). The prediction mode determiner 104 selects the one that optimizes the prediction error from these three and the candidates for the high-resolution prediction signal, gives it as the prediction signal, and sends the prediction mode to the variable length coding circuit 19. The prediction mode is variable length coded and multiplexed as type information for each prediction unit (for example, macroblock in MPEG). In the present embodiment, different type information is assigned to all of the prediction signal candidates.

Alternatively, whether or not the prediction from the low resolution image is included may be set as a 1-bit flag.

FIG. 12 is a block diagram showing a configuration example of the predictor + prediction mode determiner 104 of FIG. The operation will be described in correspondence with the description of FIG.
And the line of △ are separated (even-odd line separation circuit 130)
For each field, the optimum prediction signal double triangle is obtained from the high resolution image (high resolution prediction selection circuit 131).
The motion vector corresponding to this is the prediction mode determiner 135.
Sent to. Next, an optimum prediction signal ● from the low-resolution image for each field is selected (low-resolution prediction selection circuit 132) and, if necessary, the corresponding motion vector is sent to the prediction mode determiner 135.

Three combinations are performed by these combinations (even / odd line merge circuit 133), and prediction signals corresponding to (a) and (b) of FIG. 11 are created and sent to the prediction mode determiner 135. In the high-resolution prediction circuit 134, other high-resolution prediction signals (for example, those corresponding to the prediction currently under consideration in MPEG2) are created and sent to the prediction mode determiner 135 together with the corresponding motion vector. The prediction mode determiner 135 selects the prediction signal with the smallest prediction error from all the prediction signals sent, and transmits it to the difference circuit 12. Also, a motion vector corresponding to the selected motion vector is selected and transmitted to the variable length coding circuit 19.

Here, the optimum shift amount detected for the low-resolution image is transmitted as motion vector information, separately from the motion vector information of the high-resolution image, in a variable-length coding circuit by being variable-length coded and multiplexed. You may. In the present embodiment, the mode in which only the field dropped by the field skip is predicted by the low resolution signal is low in selection rate, and therefore, it is considered that this mode can be excluded from the selection branches. The shift can also be limited to prediction of fields dropped by field skipping.

Further, another embodiment of the present invention will be described.

FIG. 13 is a diagram showing the configuration of the predictor + prediction mode determiner in this embodiment. The same parts as those in FIG. 12 are designated by the same reference numerals.

First, similar to the embodiment of FIG. 12, the high resolution prediction circuit 134 creates high resolution prediction signal candidates. This includes forward and backward motion vector detection. Further, even and odd lines of the input are separated by the even-odd line separation circuit 130, and the high-resolution prediction selection circuit 131 selects the optimum high-resolution image when the high-resolution image is predicted for each field. In this embodiment, it is performed by the low resolution prediction selection circuit 132. The method of creating a low-resolution prediction signal for predicting a field dropped by field skip differs from that in the above embodiment. In the above-described embodiment, the motion vector is detected and motion compensation is used, but in this embodiment, the motion vector obtained for the high-resolution prediction candidate is converted into one-field interval. This state is shown in FIG. Using the motion vector given as a candidate for high-resolution prediction for the even field prediction, for example, the vector of V _{1 in} the figure (shown by the dotted line), the vector of the motion vector V ₁ ′ for the low-resolution prediction (Indicated by the bold line in the figure). For example, when the predicted even field and the fields to be used are n fields apart as shown in the figure,

[0072]

[Equation 3] Is obtained by

In the case of this embodiment, since this motion vector is directly used for low resolution prediction, it is not necessary to send the motion vector for low resolution prediction.

Here, a predetermined search range around this motion vector is searched to re-determine the optimum motion vector for low resolution prediction. In the case of this example, if this optimum motion vector is, for example, the vector of V ₁ ″, the difference between this vector and the vector of the reference vector V ₁ ′.

[0075]

[Equation 4] Is sent as a difference vector.

The prediction signal candidates created on the low resolution side are averaged with the prediction signal candidates on the high resolution side by the averaging circuit 140 and merged by the even-odd line merge circuit 133 to become the low resolution prediction signal candidates. Of these candidates and the candidates of the high resolution prediction signal, the prediction mode determiner 135 selects the one with the smallest prediction error.

In this embodiment, since the motion vector of the low resolution image is created from the motion vector of the high resolution image, both of them must correspond to each other. That is, for example, when the one using the vector of V ₁ is selected in FIG.
₁ 'vector of, when those vectors of V ₂ are used is chosen V _2' is a vector of used in low resolution side.
Since there are three types of prediction on the high resolution side, the forward direction, the backward direction, and both directions, in the case of both directions, for example, a motion vector on the field side close to the predicted field may be used.

[0078]

As described above, according to the present invention, in the moving picture coding apparatus by the embedded coding method using the hierarchical coding, the prediction ability using the low resolution local decoded signal can be improved. It can be effectively used for predictive coding to improve coding efficiency.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of a moving picture coding apparatus according to the present invention.

FIG. 2 is a diagram showing a method of creating a low-resolution image signal by dropping one field and effectiveness of applying motion compensation to a low-resolution locally decoded signal.

FIG. 3 is a diagram showing a determination function when selecting a low resolution prediction signal and a high resolution image prediction signal in the prediction mode determination circuit in FIG.

4 is a block diagram showing a configuration example of a mode determination circuit in FIG.

FIG. 5 is a diagram showing another method for creating a signal for predicting a dropped one field when a low resolution image signal is created by dropping one field.

FIG. 6 is a diagram for explaining a method of using both a low resolution image signal and a high resolution image signal for prediction.

FIG. 7 is a block diagram showing a second embodiment of the moving picture coding apparatus according to the present invention.

FIG. 8 is a diagram for explaining a predictive coding operation for a high resolution image signal in the embodiment of FIG.

FIG. 9 is a block diagram showing a third embodiment of the moving picture coding apparatus according to the present invention.

FIG. 10 is a diagram showing the relationship between encoded images of various sizes in the embodiment of FIG.

FIG. 11 is a diagram showing candidates for switching low resolution prediction in the embodiment of FIG.

12 is a diagram showing the configuration of a predictor + prediction mode determiner in the embodiment of FIG.

13 is a diagram showing another configuration of the predictor + prediction mode determiner in the embodiment of FIG.

FIG. 14 is a diagram showing how low-resolution shift amounts are calculated in the embodiment of FIG.

FIG. 15 is a block diagram showing a conventional embedded coding method using hierarchical coding.

[Explanation of symbols]

 11 ... High resolution image signal 12 ... Subtractor 13 ... High resolution prediction signal 14 ... High resolution prediction error signal 15, 25, 51, 54-56 ... Selection switch 16 ... Low resolution prediction signal 17 ... DCT circuit 18 ... Quantization circuit 19 ... Variable-length coding circuit 20, 31 ... Buffer 22 ... Inverse quantization circuit 23 ... Inverse DCT circuit 24 ... Adder 26 ... High resolution local decoded signal 27, 57 ... Frame memory 28, 36 ... Predictor 29 ... Down sampling Circuit 30 ... Encoding unit of existing system 33 ... Local decoder 34 ... Low resolution local decoded signal 35 ... Upsampling circuit 37 ... Prediction mode determination circuit 38, 39 ... Motion vector information 40 ... Prediction mode determination signal 41 ... High resolution prediction mode Judgment circuit 42, 43 ... Motion vector information amount estimation circuit 47 ... Operation comparison circuit

Front Page Continuation (72) Inventor Toshinori Otaka 1 Komukai Toshiba-cho, Sachi-ku, Kawasaki-shi, Kanagawa Inside the Toshiba Research Institute Co., Ltd. (72) Inventor Tadahiro Oku 1 Komu-shi Toshiba-cho, Kawasaki, Kanagawa Inside the Toshiba Research Institute

Claims (8)

[Claims] 1. A first for predictively encoding a high resolution image signal
And a first local decoding means for decoding a coding result by the first coding means to obtain a high resolution local decoded signal, and a first local decoding means for obtaining a high resolution prediction signal from the high resolution local decoded signal. 1 prediction means, a conversion means for converting the high resolution image signal into a low resolution image signal, a second encoding means for encoding the low resolution image signal, and an encoding by the second encoding means. Second local decoding means for decoding the result to obtain a low-resolution locally decoded signal, up-sampling means for up-sampling the low-resolution locally decoded signal, and a low-resolution prediction signal from the signal up-sampled by this means. Second prediction means, prediction error power when the high-resolution prediction signal and low-resolution prediction signal are used for predictive coding by the first coding means, respectively, and Prediction signal selecting means for selecting a prediction signal to be used by the first coding means from the high resolution prediction signal and the low resolution prediction signal according to a determination condition considering the amount of additional information generated by the coding means. A moving picture coding device comprising: 2. The high-resolution image signal is an interlaced signal, and the second predicting means shifts the signal upsampled by the upsampling means by an amount corresponding to a motion within a field period. 2. The moving picture coding apparatus according to claim 1, wherein the low resolution prediction signal is obtained from the shifted signal and the high resolution local decoded signal. 3. The first predicting means outputs a signal obtained by adding a high frequency component of the high resolution local decoded signal and a low frequency component of the low resolution local decoded signal as the high resolution prediction signal. The moving picture coding apparatus according to claim 1 or 2, further comprising: 4. A first encoding means having a mode for predictively encoding a high resolution prediction error signal, which is a difference between a high resolution image signal and a high resolution prediction signal, using a low resolution prediction error signal, and the first encoding means. First local decoding means for decoding the coding result by the coding means to obtain a high-resolution local decoded signal, and detecting a motion vector between the high-resolution local decoded signal and the high-resolution image signal, A first obtaining the motion-compensated high resolution prediction signal using the motion vector;
Predicting means, a converting means for converting the high resolution image signal into a low resolution image signal, a second encoding means for encoding the low resolution image signal, and an encoding result by the second encoding means. Second local decoding means for decoding a low-resolution locally decoded signal, up-sampling means for up-sampling the low-resolution locally decoded signal; A second prediction for obtaining a low-resolution prediction error signal that is motion-compensated using the obtained motion vector and for obtaining the low-resolution prediction error signal by taking the difference between the low-resolution prediction signal and the low-resolution local decoded signal. A moving picture coding apparatus, comprising: 5. A first encoding means for predictively encoding a high resolution interlaced image as a frame image, and a first encoding means for decoding the encoding result by the first encoding means to obtain a high resolution locally decoded signal. Local decoding means, first prediction means for obtaining a high resolution prediction signal from the high resolution local decoded signal, conversion means for converting the high resolution interlaced image into a low resolution non-interlaced image, and the low resolution non-interlaced image. Second encoding means for encoding the interlaced image signal; second local decoding means for decoding a result of the encoding by the second encoding means to obtain a low-resolution local decoded signal; and the low-resolution local part Up-sampling means for up-sampling a decoded signal, second predicting means for obtaining a low-resolution prediction signal from the signal up-sampled by this means, and the frame image A prediction signal to be used for predictive encoding by means of a first prediction means and a second prediction means, and a prediction using the low-resolution prediction signal is performed for each field forming the frame image. A moving picture coding apparatus characterized by switching whether or not to do so. 6. The second predicting means adds a signal obtained by shifting the signal up-sampled by the up-sampling means by an amount corresponding to a motion within a field period to the candidates for the low-resolution predicted signal. The moving picture coding device according to claim 5. 7. A shift amount corresponding to a motion within the field period is created by dividing a shift amount used to create the high resolution prediction signal by an inter-field distance, and a low resolution created thereby. 7. The moving picture coding apparatus according to claim 6, wherein a signal created from a prediction signal and the high-resolution prediction signal is added to a prediction signal candidate. 8. A search is performed around a reference shift amount created by dividing the shift amount corresponding to the movement within the field period by the inter-field distance by the shift amount used to create the high resolution prediction signal. 7. The moving picture coding apparatus according to claim 6, wherein the moving picture coding apparatus is created by obtaining a shift amount that optimizes a certain evaluation function, and the difference between the optimum shift amount and the reference shift amount is transmitted.