JPH0681308B2 - Quantization noise suppression method in interframe coding - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a highly efficient interframe coding method in which processing is performed for each block.

(2) Conventional technology A method of dividing an inter-frame difference signal into blocks of a predetermined size and encoding each divided block is an orthogonal method represented by a vector quantization method, a cosine transform, a Hadamard transform, or the like. There is a transform coding method.

Since these methods collectively code the pixels in the divided blocks, the influence of other areas in the block is affected even in the area where the pixel values in the block are zero. There is a common drawback that noise is generated. In order to explain this drawback, the vector quantization method will be described as an example.

In the vector quantization method, an image signal converted from analog to digital is divided into blocks, for example, every 8 lines × 8 pixels. When one pixel is represented by 8 bits, the information amount per block is 8 bits x 64 pixels and 512 bits. If this is transmitted as it is, the communication capacity is 512 bits x block.

On the other hand, in the vector quantization method, the amount of information per block is reduced by using the correlation between neighboring pixels of the image.
Significant bandwidth compression is performed.

In other words, in principle, there are ^{28 x 64} combinations of values taken by each pixel in the block. However, there is a very large bias in the occurrence probability of the combinations, and if this is used and the number of combinations is limited to 128, the amount of information per block becomes 7 bits, which is approximately 1/73. Band compression is possible.

In this way, the band compression rate is extremely high, but the quantization noise is generally large because the number of combinations per block is greatly limited. Since this quantization noise is determined by a combination of pixel values in a block, there is a drawback that even in a region where the pixel value is zero, the quantization noise is generated under the influence of other regions in the block.

As a method of suppressing such quantization noise, there is a method of performing non-linear processing on a prediction error signal containing quantization noise. This method is based on the S60-
2 "Examination of interframe coding method using vector quantization". In this method, a prediction error signal containing quantization noise was always subjected to a conversion process with nonlinear characteristics. For this reason, if a small prediction error signal is suppressed for the purpose of suppressing the quantization noise that occurs in a static area within a block that contains a large prediction error signal, such as a moving edge part, the edge part is not included if it is suppressed. The same process will be applied to the blocks,
There is a drawback in that the follow-up characteristics for a true minute level change are deteriorated and, as a result, the image quality is deteriorated as if a stain is stuck on the screen.

(3) Object of the Invention An object of the present invention is to suppress quantization noise in a frame including a moving edge portion without increasing sticking noise in a flat portion and improve S / N in a quantization noise suppression method in interframe coding. To provide.

(4) Configuration of the Invention (4-1) Difference between Features of Invention and Prior Art The present invention provides the characteristics of a non-linear circuit for suppressing quantization noise,
It is characterized in that it is adaptively controlled by the index information indicating the result of the quantization processing in block units or the optimum motion vector information detected by the motion compensation prediction. That is, the present invention differs from the prior art in the presence or absence of control of the nonlinear circuit.

(4-2) Embodiment FIG. 1 shows an embodiment of the present invention, in which 1 is an input terminal, 2 is a delay circuit, 3 is a subtraction circuit, 4 is an average value separation / normalization circuit, and 5 is block identification. Circuit, 6 is a vector quantization coding circuit, 7 is a code assignment circuit, 8 and 102 are buffer memories, 9 is a data output terminal, 1
0,104 is a vector quantization decoding circuit, 11,105 is a weighting / average value addition circuit, 12,106 is a non-linear circuit, 13,107 is an addition circuit, 1
4, 108 is a frame memory, 15, 109 are variable delay circuits, 16 is a motion vector detection circuit, 100 is a digital transmission line, 101 is a data input terminal, 103 is a code decoding circuit, and 110 is an output terminal.

The digitized video signal input from the input terminal 1 is delayed by the delay circuit 2 for a predetermined time and then subtracted from the prediction signal which is the output of the variable delay circuit 15 in the subtraction circuit 3, and the difference signal is averaged. It is supplied to the value separation / normalization circuit 4.

The average value separation / normalization circuit 4 has a predetermined size, for example, 4
The following processing is performed for each block having a size of line × 4 pixels.

That is, the average value per block is calculated for the output of the subtraction circuit 3, and the average value is subtracted from the value of each pixel included in the block. And for the result of subtracting the average value,
Calculate the standard deviation per block and normalize with the obtained standard deviation. Information on the standard deviation and the average value is supplied to the code allocation circuit 7 and the weighting / average value addition circuit 11.

The block identification circuit 5 monitors the output of the average value separation / normalization circuit 4 for each block, and when the values in the block are all smaller than a predetermined threshold value, this is set as an invalid block, and the other blocks are set as valid blocks. The identification information is supplied to the code allocation circuit 7.

Regarding this block identification, there is also a method of setting an invalid block when the average value / standard deviation is smaller than a predetermined threshold value.

The vector quantizing / encoding circuit 6 uses, for example, for a combination of values of 16 pixels (4 lines × 4 pixels) in a block,
256 combinations are prepared, and each combination is defined as one vector. For each combination of input signal values, the vector with the smallest evaluation value (eg cumulative value per block of the absolute difference between the input signal value and the quantized representative value) is detected as the optimum vector and the optimum The index information representing the vector is supplied to the code assignment circuit 7, the vector quantization decoding circuit 10 and the non-linear circuit 12.

The code assigning circuit 7 assigns a predetermined code to each input signal, outputs it to the buffer memory 8, and after matching the transmission speed there, sends it to the digital transmission line 100 via the data output terminal 9. To do.

The vector quantization decoding circuit 10 converts the index information supplied from the vector quantization coding circuit 6 into a signal representing a quantized representative value.

The quantized representative value is weighted in the weighting / average value adding circuit 11 according to the value of the standard deviation, and then the average value is added.

The result obtained by adding the average values is subjected to non-linear processing in the non-linear circuit 12.

The non-linear circuit 12 is composed of, for example, a ROM and converts it into an output value of a predetermined size according to the input value. Here, the index information output from the vector quantization coding circuit 6 or the output from the motion vector detection circuit 16 is output. It is controlled as will be described later using the optimum motion vector information.

The output of the non-linear circuit 12 is added to the output of the variable delay circuit 15 in the adder circuit 13 to become a locally decoded signal, which is stored in the frame memory 14 and becomes the predicted signal of the next frame.

The variable delay circuit 15 adjusts the delay amount of the output of the frame memory 14 for a predetermined time.

The motion vector detection circuit 16 divides the television signal supplied from the input terminal 1 into blocks, for example, every 8 lines × 16 pixels. Then, for each block, for example, from the periphery including the same place on the screen one frame before, a plurality of blocks having the same size as the block are extracted from the output of the variable delay circuit 15, and the evaluation function for the input block is extracted. , For example, the one having the smallest accumulated value per block of the pixel difference absolute value between the input block and the reference block is selected as the optimum block, and the optimum motion vector information representing the optimum block is
It is supplied to the code assignment circuit 7, the non-linear circuit 12, and the variable delay circuit 15.

The variable delay circuit 15 selects a TV signal for 8 lines × 16 pixels included in the optimum block specified by the optimum motion vector information from the output of the frame memory 14,
This is supplied to the subtraction circuit 3 and the addition circuit 13.

On the receiving side, the data input from the data input terminal 101 is temporarily stored in the buffer memory 102, converted into the decoding speed, and output.

The code decoding circuit 103 decodes the code of the output of the buffer memory 102 and converts the index information into a vector quantization decoding circuit.
The average value information and standard deviation information are supplied to the weighting / average value addition circuit 105 to the nonlinear circuit 106 and the nonlinear circuit 106, and the optimum motion vector information is supplied to the nonlinear circuit 106 and the variable delay circuit 109.

The vector quantizing / decoding circuit 104 and thereafter operate similarly to the transmitting side.

FIG. 2 is a diagram showing an example of characteristics of the non-linear circuit which is a feature of the present invention.

By suppressing the output value in the area where the input value of the nonlinear circuit is small, the static area in the block including the moving edge part is suppressed.
S / N is improved. Moreover, by suppressing the output value in the area where the input value is large, the overshoot and undershoot that occur at the moving edge part are reduced.

However, such non-linear processing is controlled by the index information and may be applied only to the block including the edge portion. Also, it has a plurality of non-linear characteristics, and these are adaptively switched by the index information. Is also possible.

It is also possible to perform non-linear processing or to perform adaptive control according to the optimum motion vector information. Of course, it is obvious that adaptive control of a plurality of nonlinear characteristics is possible. It is also possible to control using both the optimum motion vector information and the index information.

In the above description, an example of vector-quantizing an inter-frame difference signal using motion compensation prediction has been described, but the present invention is not limited to this method. For example, an inter-frame difference signal using motion compensation / background prediction may be used. On the other hand, it is obvious that the method can be applied to a method of performing vector quantization after performing transform coding such as cosine transform.

In the above explanation, the nonlinear circuit
Although the example inserted in the decoding loop was described, it controls the noise reduction characteristics of the noise reducer (sometimes called a post filter) installed outside the loop to suppress noise in the reproduced TV signal. The method can also be applied.

Further, in the above description, the case where the means for suppressing the quantization noise is provided for the inter-frame difference signal to improve the characteristics in the frame direction has been described.
For example, by performing spatial processing on a signal in the television signal format corresponding to the input stage or the output stage of the frame memory in FIG. 1, it is generated due to quantization noise and other encoding processing. It is obvious that the present invention includes a method of controlling the spatial filter by index information or the like when removing noise.

(5) Effects of the Invention As described above, the nonlinear circuit provided for suppressing the quantization noise with respect to the television signal containing the quantization noise,
By performing adaptive control using the index information and the optimum motion vector information, there is an advantage that the S / N for the stationary area in the block including the moving edge portion can be improved without increasing the sticking noise in the flat portion. is there.

Further, according to the present invention, since the adaptive control is performed using the index information and the optimum motion vector information transmitted for encoding, there is an advantage that no special additional information is required.

[Brief description of drawings]

FIG. 1 is a block diagram of one embodiment of the present invention, and FIG. 2 is a diagram showing an example of characteristics of a non-linear circuit. In FIG. 1, 1 ... Input terminal, 2 ... Delay circuit, 3 ... Subtraction circuit, 4 ... Average value separation / normalization circuit, 5 ... Block identification circuit, 6
… Vector quantization coding circuit, 7… Code assignment circuit, 8, 102…
Buffer memory, 9 ... Data output terminal, 10, 104 ... Vector quantization decoding circuit, 11, 105 ... Weighted average value addition circuit, 1
2, 106 ... Non-linear circuit, 13, 107 Adder circuit, 14, 108 ... Frame memory, 15, 109 ... Variable delay circuit, 16 ... Motion vector detection circuit, 100 ... Digital transmission line, 101 ... Data input terminal, 103 ...
Code decoding circuit, 110 ... Output terminal

Claims (2)

[Claims] 1. A prediction signal is constructed by using a signal one field or more before a digitized input television signal,
The prediction error signal for the input signal is coded by block-based quantization and transmitted, and the receiving side adds the transmitted coded data to the prediction signal constructed using the signal one field or more before. In an interframe coding method that includes a quantization noise suppressing unit that reproduces a signal and suppresses the quantization noise of a reproduced image, the quantization noise suppressing unit is controlled by identification information indicating the result of the processing in block units. , A quantization noise suppression method in interframe coding, characterized in that the characteristics of the quantization noise suppression means are changed. 2. A digital input TV signal is subjected to motion compensation / background prediction using a signal of one field or more before, a prediction signal for the input signal is constructed, and a prediction error signal is encoded. At the receiving side, the received coded data is regenerated by adding the transmitted coded data to the prediction signal composed of the signal one field or more before, and the quantization noise is suppressed to suppress the quantization noise of the reproduced image. In the inter-frame coding method including the suppression means, the optimum prediction vector information detected by motion compensation prediction
A quantizing noise suppressing method in interframe coding, characterized in that the quantizing noise suppressing means is controlled to change the characteristics of the quantizing noise suppressing means.