JPH07105948B2 - Video encoding method and encoding / decoding device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a technique of predictive coding using a correlation existing between screens of a moving image signal.

(Prior art and its problems) In compressing data of a moving image signal, for example, in the case of a television signal, interframe predictive coding that utilizes correlations existing between screens called frames is not effective. well known. In particular, this method realizes a very high compression rate for a quasi-still image containing little motion from a still image, and at the same time, a reproduced image of good quality can be obtained.

However, there is a weak point that the image quality is inevitably deteriorated when trying to maintain a high compression rate because the correlation between the screens is decreased when the motion is increased. In order to supplement this, highly accurate prediction is performed using motions within the screen, and a method called motion compensation has been applied that achieves high compression ratio and high quality images even for images containing large motions. This is explained, for example, by T. Ishiguro et al., "TV Band Widows Compression Transmission Bi-Motion Compensated Interframe Coding (Televisi).
on Bandwidth Compression Transmission by Motion-Co
mpensated Interframe Coding) ”(IEEE Communicatio
n Magazine, pp.24-30, November 1982). Even if this motion compensation is used, the effect is reduced when the motion is very fast, and it is quite possible that excessive information is generated as compared with the set transmission rate. In this case, in order to reduce the amount of information generated as much as possible, the quantization characteristic of the prediction error is changed from dense to coarse, or the number of pixels to be encoded is reduced, so-called subsampling or encoding. In addition to applying field thinning to reduce the number of fields to be used, encoding may be stopped.

In this case, degradation such as looking at the image through a dirty window called the dirty window effect due to coarse quantization, and pauses in motion that are more noticeable than the decrease in spatial resolution due to subsampling or field thinning occur. Will be. There is a method of applying orthogonal transformation to a prediction error due to motion compensation as one method for reducing the amount of generated information even when the motion is fast and motion compensation is not very effective. For example, JRJain
Papers such as "Displacement Measurement"
And it's application in interframe image coding (Displacement Measu
rement and Its application in Interframe Image Cod
ing) "(IEEE Trans.Commun., vol.COM-29, No.12, pp.17
99-1808, December 1981.). However, when considering low-speed transmission, most of the AC components of the transform coefficient after orthogonal transform must be ignored. At this time, usually, an operation of truncating a coefficient corresponding to a complicated pattern, that is, a coefficient corresponding to a high frequency by quantization is performed. As a result,
The part with many changes will be equivalently smoothed.

When the movement is very fast, this smoothing often causes no problem even if the moving portion of the image is slightly blurred. Originally, if you take an image with the current accumulation-type TV camera, a fast moving object is inevitably blurred. That is, the motion-compensated inter-frame prediction error for almost fast motion, which cannot sufficiently compensate for motion, originally has a strong intra-frame correlation and is suitable for orthogonal transformation, and the smoothing causes a slight increase in blurring, but the visual deterioration is small. However, the contour is less blurred and clearer for a slow motion where the effect of motion compensation is exerted, and the prediction error is often generated independently around the contour of the moving object.

That is, the intra-screen correlation of the prediction error signal itself is very low.
Originally, the orthogonal transform is effective for reducing the redundancy of the image signal because the correlation in the screen is high, and when it is low, the advantage of using the orthogonal transform is lost. If orthogonal transformation is used at this time, the transformation coefficients for high frequencies are truncated due to coarse quantization, resulting in blurring in the contour portion of the decoded image, which is visually noticeable as a large deterioration.

(Object of the Invention) The present invention has a very high correlation between screens. A still image-a fast decoded image that can reproduce a clear decoded image without blur even if the motion is not so fast and the correlation between screens is significantly reduced. Also, a coding / decoding device for a moving image signal having extremely high coding efficiency is realized.

(Structure of the Invention) According to the present invention, in encoding a moving image, a unit that generates a prediction signal using correlation between screens, a unit that generates a prediction error from the prediction signal and the moving image, and the prediction error Means for obtaining a transform coefficient by orthogonally transforming each of the blocks into a block unit consisting of a plurality of pixels, a first quantizing means for quantizing the prediction error, a second quantizing means for quantizing the transform coefficient, The code lengths necessary for expressing the outputs of the first and second quantizing means are integrated in the block unit,
Means for generating a selection signal having a first level when the integration result corresponding to the first quantization means is smaller than the integration result of the second quantization means, and generating a selection signal having a second level otherwise. , A first selection for selecting the output of the first quantizing means when the selection signal indicates the first level and the output of the second quantizing means for the block unit when indicating the second level. Means, means for orthogonally transforming the output of the second quantizing means in block units, and when the selection signal indicates the first level, the output of the first quantizing means indicates the second level. Occasionally, the output of the orthogonal inverse transforming means is selected in response to the selection by the first selecting means, and a locally decoded signal is generated using the output of the second selecting means and the prediction signal. Means, the local decryption Approximately one screen time delayed by means for supplying to the generating means of the prediction signal of the moving image No.,
An encoding device for a moving image signal is obtained, which comprises at least an output of the first selecting means and a means for converting the selection signal.

Further, according to the present invention, a prediction error when a prediction is executed by using a correlation between selected screens in a block unit composed of a plurality of pixels or a transform obtained by orthogonally transforming the prediction error in the block unit. In decoding a moving image signal using a signal that has been code-converted by including at least information regarding the amplitude of the moving image of one of the coefficients and this selection signal, the sign inverse of the information regarding the amplitude and the selection signal is performed. Means for converting and outputting information on the amplitude and a selection signal; means for orthogonally inversely transforming the information on the amplitude in block units; information on the amplitude when the selection signal indicates the first level. , A means for selecting the output of the orthogonal inverse transform means in the block unit when the second level is indicated, and a means for generating a prediction signal using the correlation between the screens. A means for obtaining a decoded signal from the predicted signal and the output of the selecting means, and a means for delaying the decoded signal by about one screen time of the moving image and supplying the delayed signal to the means for generating the predicted signal. A decoding device for a moving image signal is obtained.

(Principle of the Invention) Since the effect of the present invention is maximized when motion compensation is applied, the following description will be given using motion compensation as an example of correlation between screens. However, the present invention is not limited to the application of this.

The principle of the present invention will be described with reference to FIGS. 1 and 2. FIG. 2 shows an example of a person's shoulder image that moves in the horizontal direction at a speed capable of motion compensation, and shows the occurrence state of a prediction error when motion-compensated interframe prediction is applied thereto. This is an example of the case where motion compensation was effective, but the non-zero prediction error (shaded lines in the figure) at this time occurs independently in the contour of the moving object, or at most linearly connected. There are many. When orthogonal transformation is applied to the prediction error in such a state, the obtained transform coefficient does not have a large specific coefficient, but rather many coefficients have a minute amplitude close to zero. And when the quantization is applied to this, most of the coefficients become zero. When the orthogonal transform is applied, the inverse transform must be performed later, but if most of the coefficients are zero or minute as described above, the value after the inverse transform (prediction error) is also almost zero. That is, the decoded image obviously includes distortion in the contour portion.
When orthogonal transformation is performed on a signal that is isolated or at most linearly connected and quantized, the original signal is almost lost, so in such a case, do not perform orthogonal transformation,
It is better to quantize the prediction error itself as shown in FIG. In this way, even if it occurs independently, if there is a certain amount of amplitude (called the dead zone of the quantization characteristic,
If the quantized output value is equal to or greater than the threshold value for the input value at which the value is zero), the error is not zero, so a contour without blur is reproduced at the time of decoding.

That is, if the motion compensation is effective, the orthogonal transformation is not applied, and if the effect decreases, the orthogonal transformation can be applied to avoid possible deterioration of the contour portion due to the orthogonal transformation. For evaluation of the effect of motion compensation, for example,
It can be easily understood by checking whether or not the value of the detected motion vector is at the limit of the motion compensation range. This is because in the normal motion vector detection, a kind of limiter operation for limiting a value within the nearest compensation range to a fast motion exceeding the compensation range is usually applied. For example, when the speed of movement of an object moving in the X direction is V and the compensation range is Vmax, even if V> Vmax, the detection result is Vmax. At this time, the actual speed is Vmax
It is indistinguishable from the case of, but has little effect on the evaluation of whether or not orthogonal transformation is applied. From the above, for example, when the detection result is close to the motion compensation range, the orthogonal transformation result may be selected, and when not, the prediction error may be selected. If the motion vector is transmitted, it is not necessary to transmit the signal representing this selection. Also, instead of using the motion vector as the evaluation criterion, a method of summing the code amounts necessary for expressing each quantization result in units of a plurality of pixels and selecting the smaller one may be considered. The prediction error is generally small while the motion is not very fast and motion compensation is effective. The prediction error increases as the speed becomes faster and the compensation becomes difficult, and as a result, the code amount required to represent it increases. Therefore, even if this code amount is used, the above-mentioned selection can be effectively performed. Note that when motion compensation is applied, the prediction error generally tends to be large for fast motions and small for slow motions. However, this is a general trend, and it is possible to deviate from this trend. As a countermeasure in this case, it is also effective to select the quantization result with the smaller code amount. That is, the reason why the orthogonal transform is applied to the inter-frame prediction error is, in other words, to reduce the code amount at the same time while maintaining good image quality. can do. In addition, if the size of the block when performing the motion compensation is set to be the same as the size of the block when performing the orthogonal transformation, the condition of the device is good. Particularly, in the case of orthogonal transformation, it is convenient to set the block size to 2 ⁿ lines × 2 ⁿ pixels, and thus the block size when performing motion compensation should also be set in this way.

(Example) Hereinafter, an example of the present invention will be described in detail with reference to the drawings.

The operation of the encoder will be described with reference to FIGS. Line 1000
The moving image signal input via the
And to the delay circuit 11. The motion vector detection circuit 10 detects the speed and direction (motion vector) of the motion in the image in units of a block (for example, N lines × M pixels) composed of a plurality of pixels, and this is detected via the line 2000 by the variable delay circuit 22. To the delay circuit 23 at the same time. The variable delay circuit 22 delays the locally decoded signal delayed by about one screen time supplied from the frame memory 21 in accordance with the speed and direction indicated by the motion vector, and uses this as a prediction signal as a prediction signal to the subtracter 12 via the line 2200. It is supplied to each delay circuit 29. Subtractor
12 is a motion vector detection circuit 10 supplied from the delay circuit 11.
A prediction error signal is generated from the motion picture signal delayed by the time required for detection and output of the motion vector in (1) and this prediction signal, and is supplied to the orthogonal transformer 13 and the delay circuit 16 via the line 1200. The orthogonal transformer 13 performs orthogonal transform of the prediction error by orthogonal transform represented by Hadamard transform or discrete cosine transform for each block suitable for transform. The size of the block at this time is made to match the block (N lines × M pixels) at the time of detecting the previous motion vector, and, for example, M
= N = 2 ⁿ (n is a positive integer) makes the processing easier. The output of the orthogonal transformer 13 is quantized by the quantizer 14, and the quantized output is supplied to the code transformer 26 and the orthogonal inverse transformer 15 via the line 1400. The orthogonal inverse transformer 15 performs an inverse transform corresponding to the orthogonal transform used in the orthogonal transformer 13,
The result is supplied to the switch 19. The prediction error supplied to the delay circuit 16 via the one-way line 1200 is delayed by the time required for calculation and output in the orthogonal transformer 13, and then the quantizer 17
Is quantized in. The output of the quantizer 17 is supplied to the delay circuit 18 and the code converter 26 via the line 1700. The delay circuit 18 converts the quantized prediction error, which is the input signal, into the orthogonal inverse transformer 15
Delayed by the time required for calculation and output of results in
Supply to switch 19. The switch 19 uses the motion vector delayed by the delay circuits 23 and 24 to select either the output of the quadrature inverse converter 15 or the delay circuit 18 for each block. The delay time of the delay circuit 23 is equal to that of the delay circuit 16 and that of the delay circuit 24 is equal to that of the delay circuit 18. The selection result of the switch 19 is essentially a prediction error, but the adder 20 generates the above-mentioned local decoded signal from this and the prediction signal supplied from the delay circuit 29, and stores approximately one screen time of the moving image. The frame memory 21 is supplied. For the output of this frame memory 21, the variable delay circuit 22
A delay time corresponding to the speed and direction indicated by the motion vector is given and output as a prediction signal. If the motion vector represents stillness, the total delay time in the delay circuits 16 and 18, the frame memory 21, and the variable delay circuit 22 is just one screen time. However, the calculator 12, the quantizer 17,
It is assumed that there is no delay in the switch 19 and the adder 20. The delay time of the delay circuit 29 is equal to the sum of the delay times of the delay circuits 16 and 18.

The transcoder 26 transcodes the motion vector supplied via the line 2300, and at the same time, the quantized prediction error supplied via the line 1700 or the quantized conversion result supplied via the line 1400. Is used to select either one of the two, and this is also code-converted. The operation of the code converter 26 will be described later. The code amount monitor signal generated in the code converter 26 is supplied to the coding control circuit 27 via the line 2627 and is used for determining the quantization characteristic to be used. The quantization characteristic selection instruction signal for each of the quantizers 14 and 17 is line 2
It is supplied via 714 and 2717, respectively. This is at the same time
It is also supplied to the code converter 26.

Here, the operation of the code converter 26 will be described in detail with reference to FIG.

The quantized prediction error supplied via the line 1700 is the Huffman code prepared by the encoder A261 from the occurrence probability of each error amplitude statistically obtained for the quantized prediction error. The code conversion is performed by a highly efficient unequal length code such as. Similarly, the quantized transform result supplied via the line 1400 is prepared in the encoder B262, and the probability of occurrence of each transform coefficient statistically obtained with respect to the transform result is expressed by a Huffman code. Code conversion is performed using a highly efficient unequal length code. Either of the outputs of the encoder A261 and the encoder B262 is selected in block units in the switch 266 using the motion vector supplied via the line 2300. line
The selection instruction signals of the quantization characteristics supplied via the 2717 and 2714 are respectively coded by the encoder C2
63, converted by encoder D264. Encoder C263, Encoder D2
Each output of 64 is supplied to the switch 267, and one of them is selected using the motion vector. The motion vector supplied via line 2300 is also transcoded. It is easy to construct a Huffman code suitable for the distribution of motion vectors, and the code converter E265 uses this to code-convert the motion vectors. The outputs of the switches 266 and 267 and the output of the encoder E265 are multiplexed in a predetermined order in a multiplexer 268 and supplied to a buffer memory 269 for matching the transmission speed of the transmission path 3000 or the writing speed of the recording medium. It The monitor signal indicating the degree of overflow (0 to 100%) in the buffer memory 269 is supplied to the encoding control circuit 27 via the line 2627.

Next, an example of selection by the switches 19, 266 and 267 will be described. Motion compensation has a great effect as long as it is within a compensable range, but for a fast motion exceeding the range, the detected motion vector has a value close to the compensation range. Therefore, for example, when a motion vector having a value close to the compensation range is given, if it is determined that the orthogonal transformation result is selected, the orthogonal transformation is performed for a relatively slow motion sufficiently within the compensation range. Can be avoided.

Next, the operation of the encoding device will be described with reference to FIGS. 5 and 6.

The code-converted moving image signal supplied from the transmission path 3000 or the recording medium is code-inverted in the code inversion converter 50, and the prediction error, the conversion result, and the motion vector are respectively represented by the line 500.
It is output via 1,500,2,5003. Details of the sign inverse converter 50 will be described later.

A quadrature inverse transformer 51 for the signal supplied via line 5002.
Performs an inverse transform corresponding to the orthogonal transform used in the encoder. At this time, it does not matter if the prediction error which has not been orthogonally transformed originally is mixed in the signal supplied via this line 5002. It suffices that at least this signal includes a conversion result obtained by correctly performing code inverse conversion. Similarly lines
The signal supplied via 5001 only needs to include at least a prediction error that has been code-inverted correctly.

The signal supplied via the line 5001 is delayed by the delay circuit 52 by the delay time in the quadrature inverse converter 51 and supplied to the switch 53. The selection of the switch 53 is controlled depending on whether or not the value of the motion vector indicates the value just within the compensation range, as in the encoding device. That is, the output of the quadrature inverse converter 51 is selected if the value is close to the limit.

The motion vector supplied via the line 5003 is used for selection by the switch 53 and generation of a prediction signal in the variable delay circuit 55 after being delayed by the delay circuit 57 for the same time as the delay circuit 52. The switch 53 selects one of the two inputs by the above-mentioned method using the motion vector and supplies it to the adder 54.
The adder 54 decodes the moving image signal using the prediction error output from the switch 53 and the prediction signal supplied from the variable delay circuit 55, and outputs it from the decoding device via the line 5000. At this time, it is also supplied to the frame memory 56 capable of storing approximately one screen of the moving image signal. The variable delay circuit 55 uses the output of the frame memory 56 to generate a prediction signal according to the motion vector. The variable delay circuit 55 may have the same configuration as the variable delay circuit 22 in the encoding device.

Next, the operation of the code inverse conversion circuit 50 will be described in detail with reference to FIG.

The code-converted signal supplied via the transmission line 3000 is first separated into a motion vector, a quantization characteristic selection instruction signal, and a motion vector in a separation circuit 500, and a delay circuit 506 and a line are respectively connected via a line 5056. Decoder C503 and decoder D504 are supplied via 5005, and decoder E505 is supplied via a line 5055.
The decoder E505 is a motion vector code inverse converter, and the encoder E2
Corresponds to 65. The sign-inverted motion vector is output via line 5003. The quantization characteristic selection instruction signal which has been code-converted is code-inverted by the decoder C503 and the decoder D504, but the decoder C503 is code-converted by the decoder C503 for the block whose prediction error is code-converted. For each block, the decoder D504 performs the correct code inverse conversion, but otherwise executes the erroneous code inverse conversion. The result is the decoder A501, via lines 5531, 5542.
It is supplied to the decoder B502 and is used to specify the quantization characteristic to be used at the time of code inverse conversion of the code-converted prediction error or orthogonal transform result supplied from the delay circuit 506. The delay circuit 506 delays the input by the time required for the code inverse conversion and output of the decoder C503 and the decoder D504 of the quantization characteristic selection instruction signal. Decoder A501, Decoder B5
In 02, the code inverse conversion corresponding to the specified quantization characteristic is executed, and the result is output via the lines 5001 and 5002, respectively. The code inversion at this time is the correct code inversion for the decoder A501 for the block in which the selection instruction signal supplied via the line 5531 is correctly code-inverted, and only when the prediction error is code-inverted. For the decoder B502, on the other hand, for the decoder B502, the correct sign inverse transform is performed only for the block in which the selection signal supplied via the line 5542 is correctly sign inverse transformed, and only when the orthogonal transform result is sign transformed. To be executed. Otherwise, the wrong code reverse conversion is executed. The switch 53 of the decoding device selects the code inverse conversion result of either the prediction error or the orthogonal conversion according to the first or second level indicated by the selection signal, but the delay circuits 506, 52 and 57 select between the signals. Since the phases are matched, it is possible to always select a signal whose sign is inversely converted. Signals that are not selected need not be properly sign-inverted.

Next, another embodiment will be described with reference to FIGS. 7 to 10.

This embodiment is another example of the signal generation method for selecting one of the quantized prediction error and the quantized orthogonal transformation result. Same as the example. Therefore, only different points will be described.

In FIG. 7, the quantized prediction error that is the output of the quantizer 17 is sent to the evaluation circuit 28 and the delay circuit 171 to the quantizer 1
The result of the orthogonal transformation, which is the output of 4, is supplied to the evaluation circuit 28 and the delay circuit 141, respectively. The evaluation circuit 28 uses the length of the unequal length code typified by the Huffman code used in the code converter 26 for both inputs, uses the code amount per block as the evaluation scale, and the quantizer. If the output of 17 has a smaller code amount, the first level selection signal is output, and if not, the second level is output, and the operation of the switch 19 is controlled via the line 2800. . For example, when the intra-block sum of the code amount required to represent the prediction error supplied from the quantizer 17 is smaller, it is supplied to the switch 19 through the delay circuit 171, the line 1700 and the delay circuit 18. Select the prediction error. This evaluation circuit 28
The output of is also supplied to the transcoder 26 at the same time for transmission to the decoding device as a selection signal. Each of the delay circuits 141 and 171 can delay the input signal by the time required for evaluation by the evaluation circuit 28, and is output via the lines 1400 and 1700, respectively. These two signals are the same as those in the previous embodiment.

The quantization characteristic selection instruction signal is delayed by the time required for the above evaluation. That is, the selection instruction signals supplied via the lines 2714 and 2717 are delayed by the delay circuits 142 and 172.
Respectively, and are supplied to the code converter 26 via the lines 1420 and 1720.

Others are the same as in the previous embodiment.

Next, the operation of the code converter 26 will be explained with reference to FIG.

The output of the evaluation circuit 28 supplied via line 2800 is the encoder F2
At the same time as being supplied to 70, it is also supplied to the switches 266 and 267. The output of the evaluation circuit 28 is a 2-level signal of 0 or 1, that is, a selection signal equivalent to a simple ON / OFF signal. Therefore, when all the switches used in this embodiment are designated to be ON / OFF, a pair of two inputs is provided. One shall be selected. For example, when the evaluation result of the quantized prediction error is smaller, when the output = 0, that is, the first level = 0, the switches are all configured to select the one relating to the quantized prediction error. I shall.

That is, the evaluation result, that is, the selection signal is essentially a binary signal, so that the encoder F270 efficiently encodes it by using a run-length encoding method or the like. The coding result is multiplexed in the multiplexer 268 with each output of the other switches 266 and 267 and the output of the encoder E265 in a predetermined order. The following is the same.

Next, the decoding device will be described with reference to FIGS. 9 and 10. In FIG. 10, a separation circuit 500 separates a motion vector, a quantization characteristic selection instruction signal, a selection signal, and a motion vector, and the selection signal is transmitted via a line 5057 to a decoder.
It is supplied to F507, where run-length decoding is performed. The result is output again via line 5004 as the selection signal.

In FIG. 9, this selection signal is
It is delayed by the same time as the delay circuit 57, is supplied to the switch 53, and is used for selection instead of the motion vector in the previous embodiment.

The other points are the same as those in the above-described embodiment, and thus the description thereof is omitted.

(Effect of the Invention) When the present invention is used, a clear moving image without blur can be reproduced for a motion which is relatively fast and has a relatively high correlation between screens, and the correlation between screens is greatly reduced. For motion, orthogonal transformation is applied to the inter-frame prediction error to cause some blurring, but smooth motion can be reproduced by greatly reducing the amount of generated information, and good image quality can be provided for all motion speeds. Therefore, the effect is extremely large.

[Brief description of drawings]

1 and 2 are diagrams for explaining the principle of the present invention, FIGS. 3 and 4 are diagrams for explaining an embodiment of an encoding apparatus according to the present invention, and FIGS. 5 and 6 are for the present invention. 3 is a diagram for explaining a decoding device according to FIG. FIGS. 7 and 8 are diagrams for explaining another embodiment of the encoding device, and FIGS. 9 and 10 are diagrams for explaining another embodiment of the decoding device. In the figure, 10 is a motion vector detection circuit, 11, 16, 18, 23, 24, 29 are delay circuits, 12 is a subtractor, 13 is an orthogonal transformer, 14 is a quantizer, 15 is an orthogonal inverse transformer, 17 Is a quantizer, 19 and 25 are switches, 20 is an adder, 21 is a frame memory, 22 is a variable delay circuit, 26 is a code converter, 27 is an encoding control circuit, 28 is an evaluation circuit, 3000 is a transmission line or A recording medium, 50 is a code inverse converter, 51 is an orthogonal inverse converter, 52 and 57 are delay circuits, 53 is a switch, 54 is an adder, 55 is a variable delay circuit, and 56 is a frame memory.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI technical display location 9071-5L G06F 15/62

Claims (3)

[Claims] 1. A prediction error obtained by prediction using a correlation between screens or a transform coefficient obtained by executing an orthogonal transform on the prediction error is selected and encoded in a unit of a plurality of pixels. On the other hand, when the motion in the screen is fast, in the moving picture coding method for selecting the transform coefficient, when the code amount for the prediction error counted in units of the plurality of pixels is smaller, the prediction error is selected. A method for encoding a moving image that is a feature. 2. When encoding a moving picture, a means for generating a prediction signal using correlation between screens, a means for generating a prediction error from the prediction signal and the moving picture, and a plurality of predetermined prediction errors. Means for obtaining transform coefficients by orthogonally transforming in block units of pixels, first quantizing the prediction error
Quantizing means, second quantizing means for quantizing the transform coefficient, and lengths of codes necessary for expressing the outputs of the first and second quantizing means are integrated in the block unit. , First
Means for generating a selection signal having a first level when the integration result corresponding to the quantizing means is smaller than the integration result of the second quantizing means, and generating a selection signal having a second level otherwise. First selecting means for selecting the output of the first quantizing means when the selection signal indicates the first level and selecting the output of the second quantizing means for the block unit when indicating the second level, Means for orthogonally inversely transforming the output of the second quantizing means in block units, the output of the first quantizing means when the selection signal indicates a first level, and the second quantizing means when the selection signal indicates a second level. Second selecting means for selecting the output of the orthogonal inverse transforming means in response to the selection by the first selecting means, means for generating a locally decoded signal using the output of the second selecting means and the predicted signal, The locally decoded signal A moving image comprising: a unit that delays the moving image by approximately one screen time and supplies the predicted signal to the generation unit; and a unit that converts at least the output of the first selection unit and the selection signal. Image coding device. 3. A prediction error when a prediction is executed by using a correlation between screens selected in a block unit made up of a plurality of pixels, or a transform coefficient obtained by orthogonally transforming the prediction error in the block unit. In performing decoding of a moving image using a signal that has been code-converted by including at least information about the amplitude of one of the moving images and this selection signal, the information about the amplitude and the sign-inverse conversion are performed on the selection signal, respectively. For outputting the information and the selection signal, the means for orthogonally inversely transforming the information for the amplitude in the block unit, the information for the amplitude when the selection signal indicates the first level, and the second level for the selection signal. Sometimes the output of the orthogonal inverse transforming means is selected in block units, the prediction signal is generated using the correlation between the screens, the prediction Means for supplying No. and means for obtaining a decoded signal from said selecting means output, delaying the decoded signal by approximately one frame time of the moving image to the generation means of the prediction signal,
An apparatus for decoding a moving image, comprising: