JPH0654975B2 - Predictive coding - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to digital transmission of an image signal, and particularly, to a highly efficient interframe predictive coding method that performs different quantization depending on a moving region of an input moving image signal. The present invention relates to band compression transmission technology.

(Prior art and its problems) Conventionally, as an interframe coding method that separates an input moving image signal into moving images and applies different quantization characteristics to each moving image region, Japanese Patent Application No. 59-169011 discloses "moving image". "Predictive coding apparatus for signals", there is known a predictive coding method that applies a quantization characteristic having a large dead zone (a range in which an output is zero for an input) with respect to a stationary area. However, in this method, the size of the dead zone changes abruptly due to the difference between the moving area and the still area, so that a tail-like screen stain remains after the object moves, and this area becomes the still area in particular. Since the dead zone is enlarged, there is a drawback that dirt remains forever. Further, since the normal selection of the quantization characteristic is made in units of a plurality of lines, in the still area after the scene change, a slight gradation difference may be noticeable at the boundary portion where the characteristic is changed.

(Object of the Invention) It is an object of the present invention to provide an encoding system that can obtain a high compression rate with little deterioration in image quality of a decoded image, particularly in a still region.

(Structure of the Invention) In order to achieve the above object, the present invention separates a moving region and a still region of a moving image signal from each other in a pixel included in the still region when performing inter-frame predictive coding of the moving image signal. Regarding prediction error, normally, coarse quantization is performed, but quantization is performed in order of coarse quantization, medium coarseness quantization, coarse quantization, and fine quantization at least over a plurality of screens at regular intervals. It was something that I was supposed to do.

(Detailed Description of Configuration) In the present invention, first, a moving image signal is separated into a moving region and a still region. Several methods for this separation are already known. For example, as described in Japanese Patent Application No. 59-194110 "Motion Image Separation Device", the screen is divided into blocks of a certain size, and the absolute value of the frame difference value of each pixel in the block is calculated within the block. There is a method of performing the addition and comparing the result of the addition with a predetermined threshold value to determine the motion of the block. Alternatively, in the method called the gradient method, from the brightness gradient in the frame and the frame difference value,
A motion vector for each pixel is obtained, and a motion region can be defined as a set of pixels whose motion vector is not zero. In the present invention, any of the dynamic and static separation methods may be used.

Next, a method will be described with reference to FIG. 1 in which coarse quantization is normally performed on pixels included in a still region and dense quantization is performed on a pixel included in a still region in an arbitrary frame. . The shaded portion in each frame in FIG. 1 is a portion determined to be a static region. Information that is generated by noise in the static area can be suppressed by performing coarse quantization on the static area in the shaded area of each frame. Also, a certain arbitrary frame, for example, the nth
By densely quantizing the entire static portion of the diagonal line of the frame, it is possible to remove the dirt on the screen at the boundary between the moving area and the still area and the still area after the scene change, which has been a problem in the past. Even if the quantization is roughened later, the stain on the screen does not occur in the still area, and the amount of generated information is reduced.

Next, referring to FIG. 2, a method of changing the quantization of the still region in the frame unit from coarse, medium, and fine will be described. Usually, coarse quantization is performed on the still region resulting from the motion / separation result, and medium roughness is performed on the still region of a certain frame k. Then, in frame m after coarse quantization is performed in the still region from the next frame for, for example, l frames, dense quantization is performed in the still region, and coarse quantization is performed in the still region from the next frame to n frames. Is performed, and the intermediate quantization is performed on the stationary area of the frame p. With coarse quantization, the screen stains that have occurred in the still parts due to scene changes etc. will remain, but by performing the quantization of medium coarseness, the image quality ) Can be improved. The image quality (dirt) can be further improved by performing denser quantization.
Once the medium roughness is quantized, the increase in information can be suppressed to a small amount even if the dense quantization is performed. On the contrary, when the coarse quantization is suddenly performed to the fine quantization, the amount of information generated increases excessively. Also, coarse quantization is used to smooth the increased information in the buffer memory during the period of 1 and n frames after the medium and dense quantization. That is, if the information is increased by finely quantizing, the generation of information is suppressed for a short time thereafter. In this way, the quantization characteristics are gradually made finer, and by repeating this at intervals, information generated by noise in the still area is suppressed, and the screen generated at the boundary between the moving area and the still area and at the still area. The dirt is removed and a high compression rate is obtained.

(Embodiment) An embodiment of the present invention will be described in detail with reference to FIG. Third
The figure is an embodiment of an encoder using the present invention. Signal line
The digitized moving image signal input from 101 is
It is supplied to the delay circuit 1 and the motion / static separation circuit 2. The motion determination signal calculated by the motion separation circuit 2 is supplied to the encoding control circuit 7 via the signal line 201. The moving image signal whose delay has been adjusted by the delay circuit 1 enters the subtractor 3. The subtractor 3 calculates a frame difference from the prediction signal delayed by one frame from the frame memory, and sends this to the quantization circuit 4 as a prediction error signal. When quantizing the prediction error signal, the quantizing circuit 4 uses the quantization switching signal from the encoding control circuit 7 to, for example, as shown in FIG. Dense quantization switching control is performed. The moving area is quantized, for example, based on the occupied state of the buffer memory 11, which will be described later, as in the conventional case. Next, the prediction error signal quantized by the quantization circuit 4 is supplied to the adder 5 and the code converter 8.

The adder 5 adds the quantized prediction error signal and the prediction signal from the frame memory 6 and supplies the result to the frame memory 6 as a local decoded signal. The code converter 8 performs variable length coding and compresses the prediction error signal supplied from the quantization circuit 4. When performing variable-length coding, the code converter 8 performs variable-length coding corresponding to which quantization characteristic is selected by the quantization switching signal from the coding control circuit 7, and outputs the variable-length code. Output to the multiplexing circuit. The code converter 9 encodes which quantization characteristic is selected by the quantization switching signal from the encoding control circuit 7 as a mode code and outputs it to the multiplexing circuit 10. The multiplexing circuit 10 multiplexes the mode code and the variable length code and outputs them to the buffer memory 11. The buffer memory 11 matches the encoding speed and the transmission path speed so that the transmission speed of the transmission path 1101 is kept constant. Also, a buffer occupancy indicating the occupied state of the buffer memory 11 is supplied to the encoding control circuit 7 via 1102. The encoding control circuit 7
As shown in FIG. 4, the quantization switch 701, the timing control circuit 702, and the timer 703 are included.
The timer-703 supplies a timing signal to the timing control circuit 702 at an arbitrary cycle. Timing control circuit 7
Reference numeral 02 converts the timing signal from the timer 703 into, for example, two kinds of timing signals indicating the timings of performing the medium, fine and fine quantization. If a timing signal is given from the timer 703 to the timing control circuit 702, a timing signal for quantizing, for example, is output, and when the next timing signal is given from the timer 703, dense quantization is performed. Output timing signal. As described above, the timing signal from the timer 703 alternately outputs the timing signal for performing the medium quantization. Further, the timing control circuit 702 monitors the amount of generated information by the buffer occupancy supplied from the buffer memory 11 via the signal line 1102, and when there is a large amount of information, a timing signal for performing medium or dense quantization. Do not put out. At this time, preparation is made to output a timing signal for medium or dense quantization, and when the next timing is given from the timer 703 and the amount of information generated is small, the prepared medium or dense is prepared. , And outputs a timing signal for performing quantization to the quantization switching unit 701. The quantization switching unit 701 sends a quantization switching signal according to the signal indicating the timing for performing the medium or dense quantization supplied from the timing control circuit 702 and the motion determination signal supplied from the motion / separation separation circuit 2 via the signal line 201. Output. When the timing signal is not received from the timing control circuit 702 to the quantization switching unit 701, the quantization switching signal for selecting coarse quantization is output to the portion determined to be the still region by the motion determination signal. Alternatively, when a timing signal for performing medium or dense quantization is received, a quantizer switching signal for performing medium or dense quantization is quantized at a portion determined as a stationary region by the motion determination signal. Supply to 4. Also, such timing control is performed by the signal line 1102.
It is also possible to supply the buffer occupancy to the timer 703 and stop the operation of the timer 703 when the buffer occupancy becomes large. In this way, the selection of the quantization characteristic for the moving region by the motion determination signal and the buffer occupancy, and the selection of the quantization characteristic for the still region, where the characteristic is gradually finely divided by the unit of the coarse, medium, and dense regions. Thus, the quantization switching is adaptively performed.

Next, the decoder will be described with reference to FIG. The buffer memory 12 matches the compression-coded signal sent from the encoder through the transmission line 1201 at a constant speed with the decoding speed, and outputs it to the code inverse converters 13 and 14. The code inverse converter 13 extracts only the mode code from the multiplexed compressed code and decodes it. The decoded signal (representing the quantization characteristic) is supplied to the code inverse converter 14 by 1301 and used for controlling the quantization switching. The sign inverse converter 14 is
Decoding of the variable-length code excluding the mode code in the multiplexed compression code is performed based on this quantization characteristic, and the decoded prediction error signal is output to the adder 15. Adder 15
Is added to the decoded prediction error signal from the code inverse converter 14 and the prediction signal from the frame memory 16 to supply the decoded signal to the frame memory 16, and is also output from the decoder via the line 1501.

(Effects of the Invention) As described in detail above, while coarsely quantizing a still region to suppress information generated by noise in the still region, the image becomes finer and denser in the still region on a frame-by-frame basis. By performing the quantization, it is possible to remove the stain on the screen in the still area, which has been a problem in the past, by adding a small amount of information, and an interframe coding method with a high compression rate is realized. Thus, when the present invention is put to practical use, its effect is extremely large.

[Brief description of drawings]

FIG. 1 is a diagram for explaining the principle of the present invention, FIG. 3 is a diagram for explaining an embodiment of an encoder according to the present invention, and FIG. 4 is a configuration example of an encoding control circuit. 5 and 5 are diagrams for explaining an embodiment of the decoder according to the present invention. In the figure, 1 ... Delay circuit, 2 ... Motion / separation circuit 3 ... Subtractor, 4 ... Quantization circuit, 5 ... Adder, 6 ... Frame memory, 7 ... Encoding control circuit, 8 ... Code Converter 9 …… Code converter, 10 …… Multiplexing circuit 11 …… Buffer memory 12 …… Buffer memory 13 …… Sign inverse converter, 14 …… Sign inverse converter 15 …… Adder, 16 …… Frame Shows each memory.

Claims (1)

[Claims] 1. In inter-frame predictive coding of a moving picture signal, a moving area and a still area of the moving picture signal are separated,
For the prediction error for pixels included in the still region,
It is characterized by performing coarse quantization, coarse quantization, medium coarseness quantization, coarse quantization, and fine quantization in order at least over a plurality of screens at regular intervals, while normally performing coarse quantization. Predictive coding method.