JPS59141887A - Estimating and coding device of dynamic picture signal - Google Patents

Abstract

PURPOSE:To suppress the generation of information without deteriorating picture quality by changing the roughness of quantizing characteristics between the static and dynamic parts of a picture. CONSTITUTION:The input dynamic picture signal is supplied to an optimum estimated function detecting circuit 12 as well as to a subtractor 10. The difference between said picture signal and the estimated signal supplied from a variable delay circuit 11 is supplied to a quantizing circuit 13. The output of the circuit 13 is supplied to a compressing/coding circuit 17 as well as to an adder circuit 14. Then the output added with the output of the circuit 11 is delivered to an interpolating circuit 15. The output of the circuit 15 is supplied to the circuits 11 and 12 via a frame memory 16. The circuit 12 selects an estimated function with which the error of estimation can be reduced from the input dynamic picture signal and the picture signal supplied from the memory 16 and supplies the function to circuits 11, 13 and 17, respectively. A coding control circuit monitors a buffer memory 18 storing the output of the circuit 17 to designate the quantizing characteristics as well as to indicate the execution and discontinuation of sampling as well as discontinuation of coding respectively.

Description

[Detailed description of the invention] The present invention relates to a predictive coding device for moving image signals.

If the information to be transmitted when digitally transmitting a moving image signal can be compressed, depending on the compression rate, it is possible to transmit multiple channels of moving image signals using a transmission path for one channel when no compression is used. As this compression method, interframe predictive coding is conventionally well known in the case of television signals. A feature of interframe predictive coding is that the coding efficiency for still images or single frames with little movement is extremely high, but as the movement increases, the efficiency drops rapidly. To compensate for this drawback, motion compensated interframe pre-1till coding has been devised.

This is to obtain the direction and speed of movement of the moving proboscis in one plane, that is, the motion vector, and perform interframe prediction corrected by this rX force vector. For example,
1978 Military Communication Society Technical Research Report Vol. 1.78.
There is a paper by Ninomiya entitled "Motion correction using χ (interframe coding)" (paper number IE78-6), which is described in No. 39.

In conventional predictive coding devices, regardless of whether motion compensation is applied, coding control modes such as quantization, subsampling, field repetition, or coding stop of the prediction error signal are speed matched with the transmission path. sufficiency of buffer memory for (how much of the total capacity is currently used)
It is normal to design the system so that it is selected adaptively according to

For example, when a video song is input and the prediction error becomes large and the amount of information to be transmitted increases as a result, the degree of sufficiency improves and the childization characteristic of (1) is made coarser. (2) repeating fields, (3) subsampling, (4+ coding stop), which naturally suppresses the generation of information.On the other hand, it includes only a small amount of movement depending on whether it is considered a still image or a still image. Usually, there is little information about the 1st group without pictures ((
A dense characteristic is also used in the quantization characteristic control in 1), but in the case of a still image containing a very complex pattern, a large amount of information may be generated due to thick sampling pulses during sampling. Of course, the same is true when there is a lot of noise (in this case, there is a lot of information), so the degree of sufficiency is high (and the same encoding control as in the case of moving images may be executed. Even though it is a picture, field repetition and 7' sampling are applied.If field repetition or shark sampling is applied, the amount of information generated will be roughly halved, so the buffer memory will naturally become less full. With field repetition, subsampling is canceled and the amount of generated information increases again.In other words, with field repetition, subsampling is applied and canceled frequently.As a result, image quality deterioration becomes very noticeable. This is because when field repetition uses subsampling, the resolution in the vertical or horizontal direction is halved, resulting in a decrease in resolution, that is, blurring, but if these are not used, no blurring occurs.Therefore, blurring occurs. According to the characteristics of the human eye, such frequent changes in the K11liIi negative value are very noticeable. The purpose is to reduce image quality deterioration in still parts.

The present invention provides a means for determining a prediction function that reduces a prediction error from among a plurality of prediction functions, that is, an optimal prediction function, a means for generating a prediction signal based on the determined optimal prediction function, and an input video image. means for obtaining a prediction error from a signal and this prediction signal, a plate that can perform rough Doji conversion on this prediction error when the 'i & suitable prediction function matches a prediction function suitable for prediction of a stationary part; This is a predictive coding device for moving image signals that includes a small number of encoding means.As mentioned above, the deterioration in image quality is noticeable because the resolution decreases and the return to the original resolution frequently occurs. However, the problem is that a large amount of information is generated even though it is a still image. In other words, when a still image or a still part of an image is detected, it is good to prevent information from being generated here. In predictive coding that applies motion compensation, interframe prediction, which is usually the optimal prediction method for still images, is included in one of the many child ω1j methods. If the optimal prediction function is used for a part, then this part can be considered to be a stationary part.Predictive encoding is performed using this optimal prediction function, but the prediction error at that time is quantized. Sometimes, different childization characteristics are used for each of the stationary portion and the non-stationary portion.In other words, when the interframe prediction y141I indicating stationary is the optimal prediction function, the quantity and childization characteristics are Even if the generation of information is suppressed, image quality does not deteriorate.This is because the thickness of the sampling pulse is not thick enough to be visible on the screen. The width is 5 nanoseconds, and the distance between pixels on the TV monitor is 1i, lt (distance to adjacent pixels)
When 1 vIm, the variation in this distance due to thick is extremely small, ie, 1×5/143=o, o3s. -However, this effect is not necessarily small when it comes to information generation. Assuming that each pixel is represented by PCM (8 pixels per pixel) and that the level changes by 100/255 when the pixel is touched, the result is 1oo/255X0.0
A level change occurs due to the sick by 35=3.5/255. The inter-frame difference in the stationary part should originally be zero, but in this example it is 3.5/3 due to jitter.
As many as 255 level changes occur. 100 in the outline part
/ 255 There are many level fluctuations between adjacent pixels, so in complex images that consist mostly of outlines (when there is a chic, a large amount of information is generated even in static parts). (Accordingly, as in the present invention, if the quantization characteristics are adjusted in the static part and the inter-frame difference of 4/256 or less is set as zep, the generation of information can be suppressed without deteriorating the image quality.In the moving part, Quantization characteristic f: kan<
Dirty Win
It is best to avoid using coarse quantization characteristics as much as possible, as this causes a degradation similar to viewing the image through a dirty window, which is called dow.

In this way, changing the roughness of the quantization characteristics used for the static and moving parts of the image is a great idea! ! I quality'! il: It's about suppressing the generation of information without damaging it.

Implementation 1+11 according to the present invention with reference to the drawings below:
will be explained in detail.

FIG. 1 shows a block diagram of the encoding system according to the present invention.
The digitized input video image signal is supplied via line 1000 to calculator 10 and optimal prediction function detection circuit 12. The subtracter 10 takes the difference between the input signal and the predicted 1g signal supplied by the variable delay circuit 11, and supplies this difference, ie, the prediction error, to the quantization circuit 13 via bi 1o 13. Quantization circuit 13 performs quantization of prediction errors according to information representing the optimal prediction function supplied via line 1200 and encoding control information supplied via line 1900. The details of the operation of this quantization circuit 13 will be described later.

The quantized prediction error is sent to adder 1 via line 1300.
4 and is supplied to the compression encoding circuit 17. The adder circuit 14 adds this quantized prediction error and the prediction signal output from the variable delay circuit 11 to generate a locally decoded signal, which is supplied to the interpolation circuit 15 . Of these, 111.1 circuit 15 is line 1915
According to the interpolation control signal VC supplied from VC, pixels that are not encoded during subsampling are reproduced by interpolation, and when subsampling is not performed, the output image signal of the 7J11 calculator 14 is output as is. In the case of 2:1 subsampling, in which each pixel is subsampled, the interpolation method is as follows: The output of the interpolation circuit 15 is then supplied to the frame memory 16 where the signal can be processed by J frames or delayed by one field time as specified. The output of the memory 16 is supplied to the optimum prediction function detection circuit 12 and the variable delay circuit 11.The optimum prediction function detection circuit 2 detects the difference between the input moving image signal and the image No. 4g Eri prediction unit supplied from the frame memory 16. One prediction function that can be made small is supplied to the variable delay circuit 11, quantization circuit 13, and compression encoding circuit 17 via a wire 1200.

It is assumed that the detection of this optimal prediction function is performed in units of blocks each consisting of 11v1 pixels as described in Ninomiya's Rinbun mentioned above, but is not limited to this. The variable M delay circuit 11 receives the image 1#ll supplied from the frame memory 16 according to the supplied optimal prediction ω11 function.
A necessary delay is given to the signal K to generate a maximum prediction signal, and the signal is connected to the reducer 10 and the adder J4. The delay amount of the frame memory 16 is constant and only the delay amount of the 115T variable delay circuit 11 changes. However, for example, if the optimal prediction function matches the interframe prediction, the delay amount of the variable delay circuit 11 will change. It is set so that the sum with the delay amount is exactly equal to one frame time. In the case of field repetition, this is one field time. Compression encoding circuit 1
In addition to the encoded control information, 7 receives the optimal prediction function supplied through the spinner 1200 and the quantized prediction error supplied through the spinner 1300 using unequal length codes suitable for each. encode. As the unequal length code □, a Huffman code determined from each statistical distribution is suitable, but other unequal length codes may be used. The reseal control signal provided over line 1900 is also encoded. Compression encoding circuit 1
Since the output code of No. 7 changes over time, the speed is matched with the constant speed transmission line.

The data is stored in the file memory 18. The code whose speed has been matched in the buffer and memory 18 is output to the transmission line 2000. The encoding control circuit 19 constantly monitors the degree of sufficiency of the buffer memory 18 supplied via the line 1819, and outputs a control signal adapted to the degree of sufficiency every preset encoding control unit time. For example, when the degree of sufficiency is very low, a dense quantization characteristic is used, and as the degree of sufficiency increases, the quantization characteristic is coarsened. If the degree of sufficiency does not stop increasing, field repetition or subsampling is applied. to suppress the generation of information.In the worst case, the buffer memory 1
When 8 is about to overflow, that is, when the degree of sufficiency approaches 100, encoding is temporarily stopped. The encoding control circuit 19 specifies the quantization characteristic through the line 1900, instructs the execution of subsampling, and if necessary stops the encoding, and also instructs the field repetition through the line 44A 1916, and at the same time outputs the compression code. It is also supplied to the conversion circuit 17.

Furthermore, instructions are given to execute and stop interpolation via Makoto x91st-. Since the quantization circuit 13 usually has a plurality of quantization characteristics, a set of unequal length codes suitable for each characteristic is prepared in the compression encoding circuit 17, and the quantization characteristics are changed according to the specifications of the encoding control circuit 19. and that) N pairs of unequal lengths are selected.

Next, make 1 into a child! Section 21] Regarding the operation of 3, Section 2.3
This will be explained in detail with reference to the figures.

The quantization circuit 13 quantizes the prediction error supplied via the il'o 13 and outputs the result via the il'o 1300.

In this example, four types of Japanese and Japanese characteristics are used,
It is assumed that each conversion circuit is realized by l'1133 to l'136. These R>M circuits can be easily implemented using read-only memory (ROM). An example of the quantization characteristics is shown by a solid line in FIG. Return to Figure 2.

The conversion outputs of the four types of conversion circuits 133 to 136 are selected by the selection circuit 1.
37 selects only one and supplies it to the gate circuit 138.

Determination circuit 131 determines whether the information representing the optimal prediction function supplied via line 1200 matches the interframe prediction, and supplies the result to selection signal generator 132 . The selection signal generator 132 generates a selection signal for the selection circuit 137 and a gate signal for the gate signal 138 from the encoded control signal supplied via the line 1900, and supplies them via lines 3237 and 3'238, respectively. . The selection in the selection circuit 137 corresponds to the selection of quantization characteristics.

Next, a method of generating a gate signal will be explained.

In the case where the r*-f property is specified by the line 1900, the range of the input signal (referred to as the teno i'' zone) in which the output signal is zep changes depending on the judgment result in the judgment circuit 131. When the judgment result indicates inter-frame prediction, the prediction error at this time corresponds to a stationary part, so the dead zone is widened (but otherwise it is left as is, i.e. 133 to 13).
It is assumed that the currently selected variable circuit of No. 6 has the child characteristics shown by the solid line (dead zone is A) in FIG. If the prediction is indicated, the gate circuit 138 expands the dead zone to B (B>A). Otherwise, the Ted zone is not changed. In this way, the prediction error in the stationary part is・The zone will be widened.

Next, we will discuss the case where subsampling is designated by step 1900. In the case of subsampling, the dead zone can be set to infinity isometrically in order to make the quantization output zero for pixels that are thinned out and not coded. For pixels to be encoded, the dead zone is set to B for interframe pre-prediction [1, and set to A when it is not interframe prediction.

Therefore, in the case of 2=1 subsembling, the dead zone becomes infinite for every other pixel. In the case of a coding stop, the dead zone is chosen to be infinite during the stop period. Also, in the case of repeating to the field, the denoteboon is selected to infinity, and the frame memory 16
At the same time, the delay amount of the variable delay circuit 11 changes so that the sum with the frame memory J6 becomes exactly one field time.

On the receiving side, since the (-j1 encoding control signal is also transmitted, the ml of the unequal length code used can also be known, so that correct decoding can be performed.

In this embodiment, the case where information representing the optimal prediction function is transmitted has been explained as an example, but it is of course not necessary to transmit this information when detecting the optimal prediction function using only encoded pixels. It is. In this case, there is no need to supply the input moving image signal to the optimal prediction function detection circuit 12 shown in FIG.

As explained in detail above, by applying the present invention, a clear image can be provided for a still part even if it includes a crude pattern, which is extremely effective in practical use.

[Brief explanation of the drawing]

1 and 2 are 7' lock diagrams for explaining the predictive coding apparatus according to the present invention, and FIG. 3 is a diagram showing an example of quantization characteristics. In the figure, 10 is a subtracter, 11 is a variable delay circuit, 121
is an optimal prediction function detection circuit, 13 is a quantization circuit, and 14 is 7
10 calculator, 15 interpolation circuit, 16) L--A memo IJ
, 17 is a compression encoding circuit, 18 is a buffer memory, 19
is an encoding control circuit. Figure 2 Figure 3

Claims (1)

[Claims] Using a plurality of prediction functions, one prediction function that reduces the prediction error, that is, an optimal prediction function, is determined, and when performing predictive encoding of a video source signal based on this optimal prediction function, the optimal prediction function is means for determining, means for generating a prediction signal based on the determined optimal prediction function, means for obtaining a prediction error from the input moving image signal and the prediction signal, and the optimal prediction function is suitable for predicting a stationary portion. 1. A predictive coding device for a moving picture signal, comprising: a reduction means capable of performing rough doji conversion on the prediction error when the prediction error matches a predicted prediction function.