JPH04127689A - Predictive encoding system for moving image - Google Patents

Abstract

PURPOSE:To improve picture quality by improving efficiency for encoding a moving image by storing local decoded images over several frames in the past and referring to the most suitable part in those images as a predictive value. CONSTITUTION:A frame memory stores decoded images 16 over several frames in the past. The frame part most suitable for prediction is searched from those images, a moving amount 31 is detected and defined as a three-dimensional motion vector 32 together with a frame number corresponding to time. Then, the part shown by the vector is extracted from the local decoded images 16 so as to produce a predictive image 12. In this case, since the local decoded images 16 over the several frames in the past are stored, probability is improved for containing the suitable one as the predictive image 12 in the stored images. Thus, the efficiency for encoding the moving image is improved, and the picture quality is improved as well.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a video predictive coding method that aims to reduce redundancy of video information by compensating for the movement of a subject and predicting the image. It relates to a method for improving image quality and encoding efficiency.

[Conventional technology]

Conventionally, for motion compensation, the transmitting side stores locally decoded images for one frame in the past and compares them with the input image to estimate motion components, that is, motion vectors, and use these to make motion-compensated predictions. Is going. This is described in, for example, Japanese Patent Laid-Open No. 56-102179.

It also has a means to store one frame's worth of background images called background memory, which stores the static background and other parts (
Attempts are also being made to improve the efficiency of predictive coding by detecting people (such as people) and making predictions accordingly (
For example, “Adaptive predictive coding method using background prediction J,1
988 Image Coding Symposium 7-4).

[Problem to be solved by the invention]

However, when only one frame's worth of locally decoded images is used, even if the background hidden behind a subject such as a person reappears, it is impossible to predict the reappeared part, resulting in a problem that efficiency decreases. there were.

In schemes with background memory, when the background reappears, the accumulated background images can be used to improve efficiency. However, this method cannot be used if the background is also moving. Furthermore, it is difficult to distinguish between a background and a subject such as a person, and image quality is inevitably degraded due to misjudgment.

[Means to solve the problem]

In order to overcome the above-mentioned problems, the present invention has means for accumulating locally decoded images over the past several frames or more and referring to the most appropriate part from among these images as a predicted value.

[Effect]

By accumulating locally decoded images of past several frames or more, the probability that an appropriate predicted image will be included in the accumulated images increases. For example, the effect becomes noticeable when a part hidden in a shadow reappears or when an image changes periodically, such as with flicker.

〔Example〕

An embodiment of the present invention will be described below using the drawings.

An embodiment of a moving picture predictive encoding device according to the present invention is shown in FIG. The differences from conventional encoding devices are as follows:
Part of frame memory 107 and interframe prediction unit 106
is part of. First, the operating principle will be briefly explained.

1. The prediction method switching unit 101 selects, from the input image 11 and the predicted image 13, interframe prediction 13, which predicts using the correlation between frames, and intraframe prediction 14, which predicts only within the frame.

When intra-frame prediction 14 is selected, the input image 11 is
CT (orthogonal transform) quantization or vector quantization (V
Q) Send to department 103. In addition, when interframe prediction 13 is selected, the difference between predicted image 12 and input image 11 is calculated by subtracter 1
02 and sent to the DCT quantization or vector quantization unit 103. It is also possible to omit this switching unit and perform only interframe prediction.

2. DCT quantization or vector quantization section 103:
Redundancy in image information is suppressed using spatial correlation of images. The compressed image signal is then efficiently encoded and output by the entropy encoder 104. Further, the compressed image signal is subjected to inverse DCT quantization or inverse vector quantization, and added to the predicted image 12 by an adder 106 to generate a locally decoded image 16.

3. The locally decoded image 16 is stored in the frame memory 107 for several frames or more in association with its time information.

4. The interframe prediction unit 108 uses the frame memory 107
A motion vector is detected from the input image 11 and the past locally decoded image 16 stored in the past, and a predicted image 12 is generated from the past locally decoded image 16.

FIG. 2 shows the principle of conventional motion compensation prediction. In the conventional method, the amount of motion 21 is determined from the locally decoded image 16 stored in the frame memory and the input image.

Then, using the motion amount 21 as a two-dimensional motion vector 22, a predicted image 12 is created by extracting a portion indicated by the vector from the locally decoded image 16.

FIG. 3 shows the principle of motion compensation prediction in an embodiment of the present invention. In this method, past locally decoded images are stored in the frame memory over several frames or more. Among them, the part of the frame most suitable for prediction is found, the amount of motion 31 is detected in the same manner as in the conventional method, and a three-dimensional motion vector 32 is obtained by combining it with a frame number corresponding to time. Then, a predicted image 12 is created by extracting a portion indicated by a vector from the locally decoded image 16.

Note that in the above method, locally decoded images in the frame memory are stored as they are, but it is also possible to reduce the memory capacity required for storage by processing such as thinning.

Further, although the above method mentions DCT and vector quantization, it is obvious that this part can be replaced with other redundancy compression methods such as DPCM.

〔Effect of the invention〕

As described above, by applying the present invention, the encoding efficiency of moving images is increased and the image quality is also improved, so the effects of implementing the present invention are extremely large.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention, FIG. 2 is a diagram explaining the principle of motion compensation prediction according to the prior art, and FIG.
The figure is an explanatory diagram of the principle of motion compensation prediction according to an embodiment of the present invention. 11... Input image to the device, 12... Predicted image as a result of motion compensation prediction, 13... Inter-frame predicted signal, 14... Intra-frame predicted signal, 15... Encoded output code, 16... locally decoded image, 101...
Prediction method switching unit, 102...subtractor, 103DCT
Quantization section or vector quantization (VQ) section, 104.
- Entropy encoder, 105... Inverse DCT inverse quantization or inverse vector quantization unit, 107... Frame memory, 108... Inter-frame prediction unit, 21... Motion amount of liquid deep body, 22...・Two-dimensional motion vector, 31...
Amount of movement of the subject, 32...3D ¥, 21 ¥1 (return) ζme 1! A prelude to the strength! 1st 3rd Ktn-Support tfyljz34s

Claims (1)

[Claims] 1. In a method that improves encoding efficiency by detecting motion vectors representing the amount and direction of motion of a subject from a moving image signal and obtaining a predicted image that compensates for these motion vectors, 1. A moving picture predictive coding method, characterized by having means for accumulating information related to two pictures, that is, two or more frames. 2. In the moving picture predictive coding method according to claim 1, the moving picture is characterized by having means for detecting an optimal motion vector by referring to two or more past frames of pictures. Image predictive coding method.