JPH02174388A - Picture encoder - Google Patents

Abstract

PURPOSE:To prevent the generation of new information by switching the code books with use of the estimated type-based signals which are originally required for a picture coding operation applying an estimated error. CONSTITUTION:A counter 11 decides a fast scene or a scene change as long as the count value 61 exceeds the threshold value since the intre-frame correlation is high and the intre-frame correlation is low. A switch 26 receives a selection signal 64 showing the result of decision of the counter 11 via an OR gate 14 and selects a static picture code book 28 when the signal 64 shows the fast scene or the scene change. While the switch 26 selects a dynamic picture code book 27 when the signal 64 does not show the fast scene nor the scene change. Thus the generation of new information is prevented.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image encoding device. Specifically, F7-
In order to distinguish between scenes and scene changes and prevent image deterioration associated with them, image quality is improved by vector quantizing the image signal using a pre-designed optimal dedicated code book. It is an object of the present invention to provide an image encoding device that can realize improvements.

[Prior Art] As an attempt to improve the image quality of digital images, in vector quantization, which is an image encoding method, the shape of the division of the quantized area and the bit allocation are not fixed, and the image is It is changed adaptively according to the situation (References: ``Method for detecting breathed regions in cosine transform differential encoding 2'' Toshiaki Watanabe, Fumio Sugiyama, Kenji Datake.

1986 IEICE General Conference, 1116)
.

Another method is to accumulate representative vectors, which represent the divided quantization regions and are the vectors that are finally output, and use the code book by sequentially communicating each block of the image signal. (Reference title: “Adaptive block size vector quantization using codebook self-generation,” Takaaki Izumioka, Hajime Suzuki, 1988 IEICE Spring National Conference, D-122).

Roughly speaking, some conventional methods that adaptively switch between intra-frame coding and inter-frame coding in response to changes in the screen involve giving the quantizer used two quantization characteristics with different quantization step widths. , in intraframe coding, which is selected when the screen is large, quantization is performed using a quantization characteristic with a smaller quantization step width, and in interframe coding, which is selected when the screen changes are small, quantization is performed using a quantization characteristic with a smaller quantization step width. There was also a method to improve image quality by quantizing using a quantization characteristic with a larger quantization step width (Japanese Patent Laid-Open No. 6
3-232691).

[Problems to be Solved by the Invention] However, in these conventional methods, additional information is generated when controlling the image in a way that adapts it to the situation of the image, and as a result, the amount of information is not reduced very much. Problems remained. Furthermore, since no special processing is performed for resolution changes in the first scene or scene change, there is a problem that the image quality deteriorates significantly.

Furthermore, in conventional image encoding devices, in order to deal with image quality deterioration at first scenes and scene changes, devices equipped with a function to determine whether it is a first scene or a scene change in advance have been realized. It had not been done.

[Means for Solving the Problems] The present invention has been made to solve such problems, and uses two types of prediction methods to subtract the predicted value of the image signal by 1, and For each constituent block, it is accurately determined whether it is a first scene or a scene change based on the signal indicating the selected prediction method with the minimum prediction error, and a video code book is created according to the determination result. We took a method of vector quantization by selectively using a code book for still images.

[Operation] By using such means, the presence or absence of a first scene or a scene change can be accurately determined, and the code book for moving images or the code book for still images can be switched and used based on the signals indicating these. However, since vector quantization is used, no additional information is generated, making it possible to prevent image quality deterioration during first scenes, scenes, and scene changes.

[Example] A circuit configuration of an embodiment of the present invention is shown in FIG. 1 and will be described.

In FIG. 1, an image signal @
50 consists of many frames, one frame is divided into many blocks, e.g. 32x32=1024, each block has a predetermined number of pixels, e.g. 8x8=6
4 pixels are included, and the image signal 50 is processed in block units.

The input image signal 50 is input to the motion vector detector 24, and the input image signal 50 is inputted to the motion vector detector 24, and the F of the previous frame from the frame memory 32 is
A motion vector for each block is determined from the fj image signal 72, and the motion vector signal 54 determined by )q is sent to the predictor 25 and the motion vector signal terminal 42. The predictor 25 uses two types of prediction methods to obtain a predicted value 55A of the next image signal 50, and in this embodiment, intra-frame prediction and inter-frame prediction are used. That is, intra-frame prediction is performed for each block of one frame from the image signal @72 of the previous frame from the frame memory 32, and motion compensation is used from the image signal @54 from the motion vector detector 24. Perform interframe prediction.

In this way, as a result of prediction performed using two types of prediction methods for each block of one frame, the value obtained by the prediction method that has the minimum prediction error is calculated as the predicted value 55 of that block.
It is output as A to the subtracter 21. At the same time, the predictor 25 outputs a prediction type signal 55B indicating the intra-frame prediction selected in prediction for each block of one frame to the counter 11 and the code book switching signal generator 13, and the receiving side A prediction type signal 55C indicating the selected prediction method is transmitted. Here, the counter 11 counts the number of blocks for which intra-frame prediction is selected for each frame, and the code book switching signal generator 13 indicates the selected prediction method for each block included in one frame. Outputs a discrimination signal.

The counter 11 determines by the discriminator 12 whether the obtained count value 61 exceeds a set threshold value (for example, 50% of the total number of blocks). If the count value 61 exceeds the threshold, the intra-frame correlation is high and the inter-frame correlation is low, so it is determined that it is a first scene or a scene change, and if it is below the threshold, it is determined that the inter-frame correlation is high. Since the correlation is high, it is determined that this is not a first scene or scene◆change. The selection signal 64 indicating the judgment result is
The switch 26 receives the selection signal 6 via the R gate 14.
If 4 indicates a first scene or a scene change, the code book 28 for still images is selected, and if not, the code book 27 for moving images is selected.

If the discriminator 12 determines that it is not a first scene or a scene change, the code
If intra-frame prediction is selected for the block, the book switching signal generator 13 sends a selection signal 64 for selecting the still image code book 28 to the switch 26 via the OR gate 14.

On the other hand, the subtracter 21 receives the image signal 850 of the next frame.
A prediction error 51 is determined by the difference between the predicted value 55A for each block of one frame from the predictor 25, and the prediction error 51 is orthogonally transformed by the orthogonal transformer 22 and output to the quantizer 23. The quantizer 23 uses either the moving image code book 27 or the still image code book 28 selected by the switch 26 according to the selection signal @64 from the discriminator 12 via the OR gate 14. Then, the orthogonal transform coefficients 52 obtained by the orthogonal transformer 22 are vector quantized to obtain a quantized output signal @53.

Here, the code book 27 for moving images and the code book 28 for still images are designed in advance to be optimal for the prediction method (intraframe prediction and interframe prediction in this embodiment) used by the predictor 25. code books have been accumulated. The code book can be designed using vectors selected from a large number of test images, or generated from random numbers assuming a distribution (for example, Gaussian distribution) of the orthogonal transform coefficients 52 from the orthogonal transformer 22. Let the vector be the training sequence.B.

G. Algorithm (Reference: “An Algorithm for Vector Quantization Design”)
Y, Linde.

Buzo (A, Buzo), R.M. Gray (f?,) f, Gray), IE Transaction on Communications (I
EEE transaction on Commu
nications), Vol, Com-28,
1, pp. 84-95, January 1980).

For example, for still image code book 28,
Orthogonal transformation coefficient 5 for the difference between a uniform gray image and a person image
2 (average O1 variance 1) and design the vector as a training sequence. As for the moving image code book 27, a Gaussian distributed random number sequence is designed as a training sequence.

Video code book 27 designed in this way
Alternatively, from the still image code book 28, the signal converter 23 selects a representative vector having the shortest distance from the vector to be encoded, and outputs the number assigned thereto as the grouped output signal 53.

The quantized output signal 53 is transmitted to the receiving side via the quantized output signal terminal 41 together with the differential vector signal 54 from the differential vector detector 24 and the prediction type signal 55C from the predictor 25.

Further, the ω child output signal 53 is output from the ω Chishu output signal terminal 41.
19, is input to the inverse quantizer 29, and is input to the m child quantizer 2
After being inverse quantized using the code book 27 for moving pictures or the code book 28 for still pictures selectively used by Code 3, it is subjected to inverse orthogonal transformation by an inverse orthogonal transformer 3Q. The output and the predicted value 55A from the predictor 25 are combined by the adder 31, and the added value, which is the estimated value of the current image signal 5Q, is stored in the frame memory 32.

The above operations are performed sequentially for each frame of the input image signal 50, and the obtained quantized output signal 53 is transmitted to the receiving side together with the motion vector signal 54 and the prediction type signal 55G.

FIGS. 2A and 2B show the flow of control operations of the entire apparatus of the embodiment shown in FIG. 1.

In FIG. 2A, an image signal 50 is input from the input terminal 40 (FIG. 1) (3101), and the input image signal 50
The force vector is determined by the force vector detector 24 for each block (3102>, and the force vector signal 5 obtained by
4 is sent to the predictor 25. The predictor 25
From the image signal 72 of the previous frame from the memory 32, intra-frame prediction (
3103), performs inter-frame prediction using motion compensation from the motion vector signal 54 (S104), selects a prediction method with the minimum prediction error (3105), and uses the value obtained by the prediction method as the predicted value 55A of the block. At the same time, a prediction type signal 55B indicating the selected intra-frame prediction is output to the counter 11, and a prediction type signal 55C indicating the prediction method is transmitted to the receiving side.

The predicted value 55A obtained by the predictor 25 is input to the subtracter 21, and a prediction error, which is the difference between the predicted value 55A and the input image signal 50 of the next frame, is calculated for each block of one frame. (3106). The obtained prediction error 51 is orthogonally transformed by the orthogonal transformer 22 (S107).

The counter 11 counts the number of blocks for which intra-frame prediction is selected 810B>, and the discriminator determines whether the count value 61 necessarily exceeds a set threshold value.
Determine based on 2. If it exceeds the threshold, a discrimination signal indicating a first scene or a scene change is output (3109). If it is below the threshold, the code book switching signal generator 13 outputs a discrimination signal indicating this for the block for which intra-frame prediction has been selected (S110), and the OR gate 14 outputs a discrimination signal indicating this from the discriminator 12. The discrimination signal is turned off (51
11, Figure 2B). The switch 26 is switched according to the discrimination signal 64 that is the output (5112), and! Senseiki 23
vector quantizes each block (5113).

Therefore, it is asked whether vector quantization has been completed for all blocks of one frame (3114), and if it has not been completed (3114N), the operations from step 5111 are repeated. If completed (S114Y), the quantization output signal @53 is transmitted to the receiving side together with the motion vector signal 54 and the prediction type signal @55C (S115>, and at the same time, the quantization output signal 53 is transmitted by the inverse quantizer 29. The inverse orthogonal transformer 30 performs inverse orthogonal transform (S117), and the adder 31 adds the output and the predicted value 55A from the predictor 25, and the added value is stored in the frame memory. 32, step 51
Repeat the work from 01.

In the above, intra-frame side prediction and inter-frame prediction have been explained as examples of two types of prediction methods for obtaining the predicted value 55A by the predictor 25, but the same function can be achieved by using the Haku type prediction method. Can be done. In that case, vJ@ designed as the best one for each prediction method
A code book for images and a code book for still images are used.

Note that the object of the present invention can be achieved even if the orthogonal transformer 22 and the inverse orthogonal transformer 30 in FIG. 1 are omitted.

Furthermore, the counting of the number of blocks for which intra-frame prediction has been selected by the counter 11 and the determination of each block by the code book switching signal generator 13 are performed not for all blocks included in one frame, but for a specific preset block. The process may be performed only for a range, for example, blocks forming the center of the screen.

[Effect of the invention] As is clear from the above explanation, according to the present invention,
Vector quantization of image signals using a dedicated code book that is optimally designed to efficiently and accurately judge changes in the image situation when there is a first scene or a scene change, and to deal with such changes. , it is possible to prevent deterioration of image quality in the first scene or scene change. Moreover, since the code book switching control is performed using the prediction type signal originally required for image encoding using prediction errors, no new information is generated. Therefore, the effects of the present invention are extremely large.

[Brief explanation of the drawing]

FIG. 1 is a circuit configuration diagram of an embodiment of the present invention, and FIGS. 2A and 2B are flowcharts showing the flow of operation of the circuit shown in FIG. 1. 11...Counter 12...Discriminator 21...
- Subtractor 22... Orthogonal transformer 23... Quantizer 24... Motion vector detector 25...
Predictor 26...Switcher 27...Moving image code book 28...Still image code book 29...Inverse quantizer 30...Inverse orthogonal transformer 3
1... Adder 40... Input terminal 41... Quantization output signal terminal 42... Motion vector signal terminal 43... Prediction type signal terminal 50... Image signal 51... Prediction error 52・
・Orthogonal transform coefficient 53 ・Quantized output signal 54 ・
...Motion vector signal 55A...Predicted value 558.55
C... Prediction type signal 61... Count value 64... Selection signal 72...
...Image signal. 32...Frame memory

Claims (1)

[Claims] Motion vector detection for obtaining a motion vector of each block by inputting an image signal including many frames and the image output of the previous frame, and dividing each frame into a plurality of blocks. means (24) receives the output from the motion vector detection means and the image output of the previous frame, and calculates the current frame of the input image signal using an intra-frame prediction method and an inter-frame prediction method. a prediction means (25) for obtaining a predicted value; a counting means (11) for receiving a prediction type signal indicating a prediction method selected for each block in the prediction means and counting the number of prediction type signals; , determine whether or not the count value of the counting means has reached a predetermined value, and apply the selected prediction method to the selected prediction method from code books optimally configured for the intra-frame prediction method and the inter-frame prediction method, respectively. a discriminating means (12) for generating a discriminating signal for discriminating the optimum code book; and a discriminating signal for discriminating the optimum code book for each block in response to the discriminating signal from the discriminating means. a code book switching signal generating means (13) for generating a code book switching signal; an OR gate means (14) for ORing the output from the discriminating means and the output from the code book switching signal generating means; switching means (26) for receiving a discrimination signal from the OR gate means and switching to a code book most suitable for the selected prediction method; and the intra-frame prediction method and the inter-frame prediction that are switched and used by the switching means. Code book means (2) storing the optimal code book for each method
7, 28); subtraction means (21) for obtaining a prediction error between the predicted value obtained by the prediction means and the input image signal; quantization means (23) for quantizing using the code book selected according to the above; and dequantization means (29) for dequantizing the output from the quantization means; an addition means (31) for adding the output from the dequantization means and the predicted value obtained by the prediction means; An image encoding device comprising frame memory means (32) for sending the image output of the previous frame to the prediction means.