JPH09163376A - Inter-frame coding, decoding method and device for moving image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve considerably the coding efficiency by predicting a fade-in and fade-out mixture images. SOLUTION: Coefficients ak-n , bk-n are multiplied to decoded images of n-th and m-th frames stored in frame memories 101, 102, they are added by an adder section 111, which produced a prediction image 112. A difference between an input image 113 and the prediction image 112 is taken by a subtractor processing section 114, which produces a prediction error 115. The prediction error 115 and the coefficients ak-n , bk-n are coded by a prediction error coding processing section 116 and a coefficient coding processing section 117 and the result is fed to a decoder. Coefficient coding data 119 are decoded and multiplied with n-th and m-th frame coding images 203, 204 stored in frame memories 201, 202 and the products are added by an adder section 211, which produces a prediction image 212. A prediction error coding data 118 are decoded and added to the prediction image 212 by an adder section 216, which generates a decoded image 217.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an interframe coding / decoding method for moving images.

[0002]

2. Description of the Related Art Conventional high-efficiency coding of moving images (for example, M
PEG; Moving Picture Expert
s Group, ISO / IEC 11172 standard), a method of predicting a current frame from a previous frame (forward prediction), a method of predicting a current frame from a subsequent frame (backward prediction), and predictions thereof. It is possible to appropriately switch and use the three types of methods (bidirectional prediction) for predicting using the average value of the values. Forward and backward prediction is called unidirectional prediction. Bidirectional prediction is a method of mixing two types of predicted images to generate a predicted image, but the mixing ratio was limited to 1: 1.

[0003]

A method of editing a moving image is often such that one image gradually disappears over time (fade out) and at the same time another image gradually appears (fade in). Used.

In the conventional inter-frame differential encoding method, it is impossible to predict an image in which fade-in and fade-out are mixed in either the unidirectional prediction mode or the bidirectional prediction mode. As a result, there is a drawback that a large amount of prediction error occurs and the coding efficiency is reduced.

An object of the present invention is to provide an interframe coding / decoding method and apparatus for a moving image, which has a high coding efficiency for an image in which fade-in and fade-out are mixed, that is, a high compression rate. is there.

[0006]

In the interframe coding / decoding method for a moving image according to the present invention, the decoding side linearly sums the decoded image of the nth frame and the decoded image of the mth (n <m) frame. To generate a prediction image of the k-th (n <k <m) frame, encode the prediction error of the k-th frame, and send it together with the coefficient used for the linear sum. generating a prediction image of the k-th (n <k <m) frame by using the coefficient for the decoded linear sum with respect to the decoded image of the m frame, decoding the prediction error, and adding to this prediction image To decode the k-th frame.

Further, the moving picture interframe coding / decoding apparatus of the present invention stores the decoded images of the nth frame and the mth frame (n <m), respectively. First and second frame memories, a first multiplication unit that multiplies a decoded image of the nth frame by a coefficient _akn , a second multiplication unit that multiplies a decoded image of the mth frame by a coefficient _bkn , Coefficient a _k-in the decoded image of n frames
Multiplied by _n and the decoded image of the m-th frame, coefficient b
An addition processing unit that adds a product of _kn and adds a prediction image by a linear sum, and obtains a difference between the input image of the k-th (n <k <m) frame and the prediction image to obtain a prediction error. Subtraction processing unit, and a prediction error encoding processing unit that encodes the prediction error and outputs prediction error encoded data, and encodes the coefficients a _kn and b _kn to output coefficient encoded data Coding unit including a coefficient coding processing unit, a prediction error decoding processing unit that decodes the prediction error coded data, a coefficient decoding processing unit that decodes the coefficient coded data, the nth frame, and the mth frame, respectively. Third and fourth frame memories in which the decoded image of the frame is accumulated, a third multiplication unit that multiplies the decoded image of the nth frame by the decoded coefficient a _kn, and the decoded image of the mth frame multiplied by a coefficient b _kn decoded against A fourth multiplication unit that,
A second addition processing unit that adds a decoded image of the n-th frame multiplied by a coefficient a _kn and a decoded image of the n-th frame multiplied by a coefficient b _kn to generate a predicted image by a linear sum; And a decoding unit including a third addition processing unit that adds the prediction error decoded by the prediction error decoding processing unit to the prediction image to obtain a decoded image.

According to the present invention, when performing inter-frame prediction,
It is assumed that bidirectional prediction is available.

At the encoding side and the decoding side, before processing the frame to be bidirectionally predicted and encoded,
It is necessary that the two frames (previous frame and subsequent frame) to be referred to have already been decoded.

On the encoding side, the mixing ratio of the two reference frames is changed to calculate the linear sum, and the coefficient is changed to obtain the coefficient that minimizes the prediction error. Prior to this search, it is assumed that the coefficients for the linear sum have been quantized. The coefficient with the smallest prediction error is encoded and sent to the decoding side.
Also, the prediction error is encoded and sent to the decoding side.

FIG. 1 shows an example of bidirectional interframe prediction by linear sum. Image F _k of the k-th frame sandwiched decoded image C _n and decoded image C _m of the m-th frame of the n-th frame
The predicted image P _{k for} is calculated by the linear sum P _k = a _kn Cn + b _kn C _{m with} the coefficients a _kn and b _kn .

On the decoding side, the linear sum coefficient is decoded. The two already decoded reference images are multiplied by this coefficient to generate a predicted image. Further, the prediction error is decoded and added to the prediction image to obtain a decoded image.

As described above, according to the present invention, a coefficient of the linear sum of two images is changed, for example, for each frame (every other frame may be used) to generate a prediction image, thereby predicting a fade-in / fade-out mixed image. Is possible. Further, by encoding the coefficient of the linear sum and giving it to the decoding side, it is possible to obtain exactly the same predicted image as the encoding side on the decoding side.

Therefore, since the prediction error becomes very small, the amount of coded information is small and the coding efficiency is greatly improved. Conversely, when the same amount of coded information is available, the S / N ratios of the two base images are improved, and the quality of the decoded image is improved as a whole.

[0015]

Next, an embodiment of the present invention will be described.

2 (1) and 2 (2) are block diagrams of an encoding unit and a decoding unit, respectively, which are an embodiment of the present invention.

First, the encoding unit (FIG. 2 (1)) will be described. In FIG. 2A, it is assumed that the reference frame used for prediction has already been decoded. Now the kth
A process of predictively encoding the frame input image F _k (113) is shown.

Frame memory 101, frame memory 1
02, the decoded image C _n (10
3), the decoded image C _m (104) of the m-th frame is accumulated. The multiplication processing unit 107 multiplies the decoded image 103 of the nth frame by the coefficient a _kn (105), and obtains the decoded image 109 of the nth frame by which the coefficient is multiplied. Similarly, the multiplication processing unit 108 adds the coefficient b to the decoded image 104 of the m-th frame.
_kn (106) is _multiplied to obtain the decoded image 110 of the mth frame multiplied by the coefficient. The addition processing unit 111 applies the coefficient to the decoded image 109 of the n-th frame and the coefficient to the m-th frame.
The decoded image 110 of the frame is added to generate a predicted image P _k (112) by a linear sum. The subtraction processing unit 114
The difference between the input image F _k (113) of the k-th frame and the predicted image P _k (112) _obtained by the linear sum is calculated to obtain the prediction error e _k.
(115) is obtained. The prediction error e _k (115) is subjected to the redundancy reduction, quantization, and entropy coding as necessary in the prediction error coding processing unit 116, and the prediction error coded data 118 is output. In parallel with this process,
The coefficient a _kn (105) and the coefficient b _kn (106) are entropy coded in the coefficient coding processing unit 117, and coefficient coded data 119 is output. The prediction error coded data 118 and the coefficient coded data 119 are transmitted to the decoding side.

The above is the process of predictively coding the input image 113 of the k-th frame. The variable k is changed within the range of n <k <m, and the same process is repeated.

Next, the decoding unit (FIG. 2 (2)) will be described. In FIG. 2B, it is assumed that the reference frame used for prediction on the decoding side has already been decoded.

Now, a process of decoding the prediction error coded data 118 and the coefficient coded data 119 to reproduce the image of the k-th frame will be described.

The prediction error encoded data 118 is a prediction error recovery data.
Error in the signal processing unit 213 _k Decrypt to (215)
Is done. The coefficient coded data 119 corresponds to the coefficient decoding processing unit 21.
4, the coefficient a_kn (205) and coefficient b_kn 
It is decrypted to (206). On the other hand, the frame memory 20
1. The frame memory 202 restores the nth frame
Issue image C_n (203), the decoded image C of the m-th frame_m 
(204) is accumulated. The multiplication processing unit 207 is the nth
Decoded image C of frame_n The coefficient a in (203)_kn (20
5) multiplied by a coefficient and decoded image 20 of the nth frame
Ask for 9. Similarly, the multiplication processing unit 208 determines that the m-th frame
Decoded image C_m The coefficient b in (204)_kn (206)
The decoded image 210 of the m-th frame multiplied by the coefficient
You. The addition processing unit 211 adds the coefficient to the n-th frame.
Decoded image of m-th frame multiplied by the decoded image 209
210 is added, and the predicted image P is obtained by the linear sum._k (212)
Is calculated. The addition processing unit 216 calculates the predicted image P_k (21
Prediction error e in 2)_k (215) is added to the decoded image 217.
Get.

The above is the process of decoding the image of the k-th frame, the variable k is changed within the range of n <k <m, and the same process is repeated.

[0024]

As described above, according to the present invention, a method similar to the method of generating a fade-in / fade-out mixed image is introduced into the process of generating a predicted image to predict a fade-in / fade-out mixed image. And as a result, the coding efficiency is significantly improved.

[Brief description of the drawings]

FIG. 1 is a principle diagram of a moving image interframe encoding / decoding method according to the present invention.

FIG. 2 is an encoding unit ((1) in the same figure) and a decoding unit ((2) in the same figure) according to the interframe encoding / decoding method for a moving image of the present invention.
It is a block diagram of.

[Explanation of symbols]

101, 102 Frame memory 103 Decoded image of nth frame C _n 104 Decoded image of mth frame C _m 105 Coefficient a _kn 106 Coefficient b _kn 107, 108 Multiplication processing unit 109 Decoded image of the nth frame 110 Coefficient multiplied Decoded image of m-th frame subjected to multiplication 111 Addition processing unit 112 Predicted image by linear sum 113 Input image P _k 114 Subtraction processing unit F _k 115 Prediction error e _k 116 Prediction error coding processing unit 117 Coefficient coding processing unit 118 Prediction Error coded data 119 Coefficient coded data 201, 202 Frame memory 203 Decoded image C _n 204 of mth frame C _m 205 Coefficient a _kn 206 Coefficient b _kn 207, 208 Multiplication processing unit 209 Multiply coefficient Decoded image of the n-th frame 210 Decoding the m-th frame by multiplying the coefficient Image 211 addition unit 212 predicted by the linear sum image P _k 213 the prediction error decoding unit 214 the coefficient decoding unit 215 decodes the prediction error e _k 216 addition unit 217 decodes the image

Claims (2)

[Claims] 1. An inter-frame encoding / decoding method for a moving image, wherein on the encoding side, a decoded image of an n-th frame and an m-th (n <
m) by the linear sum of the decoded images of the frame k-th (n <k <
m) generates a predicted image of the frame, encodes the prediction error of the kth frame, and sends out with the coefficient used for the linear sum. At the decoding side, for the decoded image of the nth frame and the decoded image of the mth frame, , Decoding the kth frame by generating a prediction image of the kth (n <k <m) frame using the decoded coefficient for the linear sum, decoding the prediction error, and adding to this prediction image An interframe coding / decoding method for moving images, characterized by: 2. The nth frame and the mth (n <m) respectively
First decoded image of the frame is stored, a second frame memory, a first multiplication unit for multiplying the coefficients a _kn in the decoded picture of the n-th frame, the coefficients b _kn in the decoded image of the m-th frame A second multiplication unit that multiplies, a decoded image of the nth frame multiplied by a coefficient a _kn, and a decoded image of the mth frame multiplied by a coefficient b _kn are added to generate a predicted image by a linear sum. An addition processing unit and a k-th (n <k <
m) obtaining the difference between the input image of the frame and the predicted image,
A first subtraction processing unit for obtaining a prediction error, a prediction error coding processing unit for coding the prediction error and outputting prediction error coded data, and a coding process for the coefficients a _kn , b _kn , An encoding unit that includes a coefficient encoding processing unit that outputs coefficient encoded data, a prediction error decoding processing unit that decodes the prediction error encoded data, and a coefficient decoding processing unit that decodes the coefficient encoded data, respectively. Third and fourth frame memories in which the decoded images of the nth frame and the mth frame are accumulated, and a third multiplication unit that multiplies the decoded image of the nth frame by the decoded coefficient a _kn , A fourth multiplication unit that multiplies the decoded image of the m-th frame by the decoded coefficient b _kn ;
The decoded image of the frame multiplied by the coefficient a _kn and the nth
A second addition processing unit that adds a product of the decoded image of the frame and a coefficient b _kn to generate a prediction image by a linear sum, and a prediction error decoded by the prediction error decoding processing unit on the prediction image. And an inter-frame encoding / decoding apparatus for a moving image, which has a decoding unit including a third addition processing unit that obtains a decoded image by adding.