JP2561855B2 - Encoder - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an encoding device, and more specifically to an image encoding device that encodes image information.

(Prior Art) Various encoding methods have been proposed in order to reduce the number of transmission bits when digitally transmitting information such as images and sounds. One of them is the correlation between adjacent sample values. Predictive coding (hereinafter,
There is a method called DPCM). In this encoding, the predicted value of the next sample point is obtained using the coded sample value, and the difference between this predicted value and the actual sample value is quantized and coded.

On the other hand, when the image signal is DPCM-encoded, the spectrum of the coding noise that is generated is generally white, but it is difficult to detect it visually by frequency shaping it into a spectrum that matches human visual characteristics. be able to.
As a method for this, a noise shaping filter that feeds back the quantization noise to the DPCM prediction difference value has been proposed.

A conventional image coding apparatus will be specifically described with reference to FIG. The sampled value X _i of the input terminal 10 is applied to the subtractor 12,
The subtractor 12 subtracts the predicted value from the predictor 14. Subtractor 12
The difference value output from is converted into an absolute value by the absolute value circuit 16, and is compared with the threshold value Th by the comparator 18. The comparison result of the comparator 18 controls the switch 20 as described later. Subtractor 22
Subtracts the correction value by the noise shaping filter 24 from the output of the subtractor 12. The output of the subtractor 22 is amplitude-limited by the limiter 26 and applied to the terminal a of the switch 20. The output e _i of the subtractor 12 is applied to the terminal b of the switch 20. Input terminal 10
The input values X _i, when the prediction value output of the predictor 14 and _i, the switch 20 is | X _i - _i | <is connected to the terminal a side in the case of Th, the terminal b otherwise Connected.

The difference value e _i ′ selected by the switch 20 is quantized by the quantizer 28, and the quantized output is encoded by the encoder 33 and output from the output terminal 30 to the communication path or the like. Quantizer 28
The output of is also applied to the inverse quantizer 32 for decoding, where it is decoded into the difference value Q (e _i ′). The adder 34 outputs the predicted value output from the predictor 14 to the output Q (e _i ′) of the inverse quantizer 32.
_The sampled value is restored by adding _i and applied to the predictor 14. The predictor 14 predicts the value of the next pixel from the restored sample value and outputs it as the predicted value _i . This predicted value _i is
It is applied to subtractor 12 and adder 34 in the next differential encoding cycle.

The difference value e _i ′ selected by the switch 20 and the inverse quantizer 32
The difference value Q (e _i ′) decoded in (4) is applied to the subtractor 36, and the subtractor 36 applies the difference value q _{i of the} two as quantization noise to the noise shaping filter 24. This quantization noise q _i has its spectrum shaped by the noise shaping filter 24, and the subtractor
Applied to 22. That is, it is fed back to the predicted difference value.

Here, when the transfer function of the noise shaping filter 24 is H, the spectrum N of the quantization (encoding) noise is represented by the following expression by this configuration.

N = | 1−H | ² Q However, Q represents the spectrum of the quantization noise q _i . That is, by changing the transfer function H, it is possible to realize coding noise adapted to human visual characteristics.

The noise shaping filter 24 in the one-dimensional case is, for example, the sixth
As shown in the figure, it comprises a multiplier 38 that multiplies the input signal by k, and a D flip-flop 40 that delays the output of the multiplier 38 by one pixel.

However, when the characteristic of the quantizer 28 is set to a non-linear characteristic as shown in Table 1 in order to improve the image quality, the transfer function H of the noise shaping filter 24 changes depending on the difference value. Since the quantization noise of becomes a large value and is fed back to the prediction difference value, the image quality is rather deteriorated. Further, there is a problem that the noise shaping filter 24 increases the entropy of the quantized representative value, which increases the amount of information.

(However, the negative level is symmetrical with the positive level.) In order to solve these problems, conventionally, in consideration of human visual characteristics with respect to signal difference, as shown in FIG.
When the prediction error | X _i − _i | exceeds a certain threshold Th, the switch 20 is connected to the terminal b side so that the noise shaping filter 24 is not used. However, even when such a measure is taken, there are still the following problems. When the quantization characteristics of the quantizer 28 are the characteristics shown in Table 1 and the previous value predictive coding using the previous value as the predicted value is performed, the relationship between the input value and the decoded value is Th = 5, k = 1 /
It becomes 2 as shown in FIG. R in FIG. 7 (a) indicates the range of the predicted value ± Th, and it is determined that there is correction at point A and B
The points are out of the range of the predicted value ± Th, and it is determined that there is no correction, and the respective code values are shown in FIG. 7B.

That is, when the above algorithm is used, if the predicted value _i and the input value X _i happen to match within the threshold value Th even in the edge portion, the correction value by the noise shaping filter 24 is fed back. In the case of non-linear quantization, this value becomes considerably large, which disturbs the image quality. Further, when the prediction error due to noise or the like exceeds | X _i − _i | threshold Th, the correction value by the noise shaping filter 24 is not fed back. Therefore, when the threshold Th is small and the noise level of the input value is high,
Whether or not the correction is performed will change depending on the noise level.
The above image disturbance is caused by the edge business (C in Fig. 7 (b),
It appears as D) and causes a significant deterioration in image quality.

(Object of the Invention) It is an object of the present invention to provide an encoding device in which such signal deterioration does not occur.

(Structure of the Invention) A first encoding device according to the present invention is a predictive encoding device that encodes an output of a quantizer that quantizes a value corresponding to a prediction error between an input signal and its predicted value signal. A noise calculating means for calculating quantization noise using the input and output of the quantizer, a noise shaping filter for frequency shaping the quantization noise, and the frequency shaped quantization noise for the input signal. Corresponding to the prediction error or the prediction error, a subtracter provided on the input side of the quantizer so as to subtract from the limiter, and a limiter arranged between the output of the subtractor and the input of the quantizer. Comparing means for detecting a first state and a second state respectively corresponding to whether the value exceeds a predetermined threshold value, and the quantum means when the comparing means detects the second state. Noise to the noise shaping filter Then, when the first state is detected, a switch controlled by the comparison means is configured to perform an operation of blocking the supply of the quantization noise to the noise shaping filter.

A second encoding device according to the present invention is a predictive encoding device that encodes an output of a quantizer that quantizes a value corresponding to a prediction error between an input signal and a prediction value signal thereof. Which is obtained at the output of the noise calculating means for calculating the quantization noise using the input and output of the converter, the non-linear circuit for limiting the level of the quantization noise below a predetermined threshold value, and the output of the non-linear circuit A noise shaping filter for frequency-shaping the level-limited quantization noise, a subtractor provided on the input side of the quantizer so as to subtract the frequency-shaped quantization noise from the input signal, and the subtraction And a limiter disposed between the output of the quantizer and the input of the quantizer.

A third encoding device according to the present invention is a noise calculating means for calculating the quantization noise of the encoder by using the input and the output of the encoder, and a threshold value for presetting the level of the quantization noise. A non-linear circuit for limiting the value to a value or less, a noise shaping filter for frequency-shaping the level-limited quantization noise obtained at the output of the non-linear circuit, and subtracting the frequency-shaped quantization noise from the input signal And a subtractor provided on the input side of the encoder.

(Operation) Since the quantization noise control means controls the input amount to the noise shaping filter such as prohibiting the quantization noise having a large value from inputting to the noise shaping filter, In this case, it is possible to effectively prevent a large image quality disturbance from occurring in the edge portion.

Embodiments Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows a configuration block diagram of an embodiment of the present invention. The same components as those in FIG. 6 are designated by the same reference numerals.
The switch 42 determines the quantization noise q _i according to the comparison result of the comparator 18.
Alternatively, 0 data is selected and applied to a noise shaping filter 44 similar to the noise shaping filter 24. That is, input value X _i , predicted value
_{For i} and the threshold value <Th, when | X _i − _i | Th, connect to the terminal a side to select the quantization noise q _i ; otherwise, connect to the terminal b side and set 0 data. Select. The noise shaping filter 44 calculates a correction value similarly to the noise shaping filter 24, and supplies it to the subtractor 46 at the next clock (DPCM cycle). The subtractor 46 subtracts the correction value from the noise shaping filter 44 according to the input value X _i of the input terminal 10, and the subtracted value is amplitude-limited by the limiter 48. The subtractor 50 subtracts the predicted value _i output by the predictor 14 from the output of the limiter 48. The output e _i ′ of the subtractor 50 is applied to the quantizer 28 and the subtractor 52. The quantizer 28 performs the quantization as described above, and the subtractor 52 uses the decoded difference value Q (e _i ′) from the inverse quantizer 32 and the output e _i ′ from the subtractor 50 to quantize the noise q. Calculate and output _i .

For convenience of explanation, if the noise shaping filter 44 has the same configuration as the noise shaping filter 24 of FIG.
The corrected difference value e _i ′ output by is as follows.

| X _i − _i | <Th, e _i ′ = X _i − _i −kq _i-1 | X _i − _i | ≧ Th, e _i ′ = X _i − _i = e _i Therefore, the flat part (| X _i − _i | <Th and | X _i-1 − _i-1 | Th), the same operation as in the conventional example of FIG. 6 is performed. On the other hand, in the edge portion, the quantization noise is not selected by the switch 42 and 0 data is input to the noise shaping filter 44, so that it is not fed back to the difference value. When the characteristic of the quantizer 28 is non-linear, considerably large quantization noise occurs at the edge portion where the prediction error increases, but in the present embodiment,
Since the quantization error is determined based on the magnitude of the prediction error in the calculated portion, large quantization noise is not actually fed back, and the interference in the conventional example as shown in FIG. 7 does not occur.
In the present embodiment, the quantization noise is fed back at the edge portion next to the flat portion below the threshold value Th, but the feedback amount in this case is extremely small (at most ± 1 level), so there is no problem at all. .

FIG. 2 shows a block diagram of the configuration of the second embodiment of the present invention. The present embodiment differs from the embodiment of FIG. 1 in that the input to the absolute value circuit 16 is not the prediction error e _i but the corrected prediction error e _i ′. That is, the output e _i ′ of the subtractor 50
Is applied to the absolute value circuit 16. In this case, the reason why the quantization noise q _i is fed back by the noise shaping filter 44 by the switch 42 is | e _i ′ | = | X _i − _i −kq _i-1 | <Th, that is, −Th + kq _i-1 < X _i − _i <Th + kq _i−1 . Therefore, in order to obtain the same characteristics as in the conventional example in the flat portion, the threshold value Th may be Th + | q _i-1 |. On the other hand,
With respect to the edge portion, the range of the quantization noise is determined by the corrected prediction error e _i ′, so that large quantization noise is not fed back, and the disturbance as in the conventional example does not occur.

Further, even if the input to the absolute value circuit 16 is the quantization noise q _i as shown by the broken line in FIG. 2, the threshold value Th at the comparator 18 is set correspondingly as shown by the broken line in FIG. By doing so, the same effect can be obtained.

FIG. 3 shows a configuration block diagram of a third embodiment of the present invention. The difference from the configurations of FIGS. 1 and 2 is that the quantization noise q _i (output of the subtractor 52) is non-linearly converted by the non-linear circuit 54 and supplied to the noise shaping filter 44. The characteristic of the non-linear circuit 54 is, for example, as shown in FIG. In any of the cases of a, b, and c, the output is limited by n _max , so that the noise shaping filter processing can be performed without causing the above-mentioned interference at the edge portion. Further, by setting the slope of the portion equal to or less than the threshold Th to k, the multiplier 38 in the noise shaping filter 44 can be omitted. In addition, the characteristic of the quantizer 28 is set to y
= Kx (y: output, x: input), and by adaptively switching the value of k with the DPCM code or the difference value e _i ′, the characteristics of the noise shaping filter can be changed dynamically, making it more visual. Coding noise suitable for the characteristics can be realized.

In any of the above embodiments, the decoding device on the receiving side is
A DPCM decoding device will do.

FIG. 5 is a block diagram showing the configuration of an embodiment when the present invention is applied to an encoding device other than the DPCM system. The signal input to the input terminal 60 is applied to the subtractor 62, the correction value output from the noise shaping filter 64 is subtracted, and the signal is supplied to the encoder 66. The encoder 66 performs encoding according to a predetermined encoding algorithm, and outputs it from the output terminal 68 to a communication path or the like. The coded code output from the encoder 66 is also applied to the decoder 70 and decoded into the original signal. The subtractor 72 calculates the difference between the output of the subtractor 62 and the decoded value output by the decoder 70, that is, the coding noise. The output of the subtractor 72 is supplied to the noise shaping filter 64 via the non-linear circuit 74. The non-linear circuit 74 is
For example, the input / output characteristics are as shown in FIG. The noise shaping filter 64 calculates a correction value for correcting the coding noise into a spectrum that matches human visual characteristics,
It is applied to the subtractor 62.

The noise shaping filter 64 is a filter for frequency-shaping the spectrum of the coding noise into a spectrum suitable for human visual characteristics. Therefore, the encoder 66 uses scale / index coding, vector quantization (coding). It is obvious that the present invention can be applied to any of the above cases.

In this embodiment, as indicated by the non-linear circuit 74, of the coding noise, the one with a large value is not fed back, so that a noise shaping filter is designed which does not cause deterioration and has a very small increase in entropy. It will be easier. Also, the configuration of the receiving side decoder corresponding to this embodiment may be a known configuration corresponding to each encoding.

(Effects of the Invention) As can be easily understood from the above description, according to the present invention, when quantization (or coding) noise is large, feedback is not performed through the noise shaping filter, so that the edge No major image quality disturbance occurs in the department. Combined with shaping into coding noise suitable for human visual characteristics, the image quality can be greatly improved.

[Brief description of drawings]

FIG. 1 is a block diagram of the configuration of the first embodiment of the present invention, and FIG.
FIG. 4 is a block diagram of the configuration of the second embodiment of the present invention, FIG. 3 is a block diagram of the configuration of the third embodiment of the present invention, and FIG. 4 is a characteristic of a non-linear circuit used in the third embodiment of the present invention. FIG. 5 is a block diagram of the configuration of the fourth embodiment of the present invention, FIG. 6 is a block diagram of the conventional example, and FIG. 7 is a relationship between the input value and the decoded value in the case of the apparatus of FIG. FIG. 10,60 …… Input terminal, 12,22,36,46,50,52,62,72 …… Subtractor, 14 …… Predictor, 16 …… Absolute value circuit, 18 …… Comparison circuit, 20,42 ...... Switch, 24,44,64 …… Noise shaping filter, 26,48 …… Limiter, 28 …… Quantizer, 30,68 …… Output terminal, 32 …… Inverse quantizer, 33,66 …… Encoder, 34 ...
... Adder, 38 ... Multiplier, 40 ... D flip-flop, 54, 74 ... Non-linear circuit, 70 ... Decoder.

Claims (3)

(57) [Claims] 1. A predictive coding apparatus for coding an output of a quantizer for quantizing a value corresponding to a prediction error between an input signal and a predicted value signal thereof, wherein the input and output of the quantizer are: Noise calculating means for calculating quantization noise using, a noise shaping filter for frequency shaping the quantization noise, and an input side of the quantizer for subtracting the frequency shaped quantization noise from the input signal. A limiter arranged between the output of the subtractor and the input of the quantizer, and the prediction error or a value corresponding to the prediction error exceeds a predetermined threshold value. Comparing means for detecting a first state and a second state respectively corresponding to whether or not, and the quantizing noise is supplied to the noise shaping filter when the comparing means detects the second state. When the first condition is detected Encoding apparatus and a switch controlled by said comparing means to an operation to block the supply to the noise shaping filter of the quantization noise. 2. A predictive coding device for coding an output of a quantizer for quantizing a value corresponding to a prediction error between an input signal and its predicted value signal, wherein the input and output of the quantizer are Noise calculating means for calculating the quantization noise by using a non-linear circuit for limiting the level of the quantization noise to a predetermined threshold value or less, and the level-limited quantum obtained at the output of the non-linear circuit. Noise shaping filter for frequency shaping the quantization noise, a subtractor provided on the input side of the quantizer so as to subtract the frequency shaped quantization noise from the input signal, an output of the subtractor and the quantum And a limiter disposed between the input of the coder and the encoder. 3. Noise calculating means for calculating the quantization noise of the encoder by using the input and the output of the encoder, and a non-limiter for limiting the level of the quantization noise to a threshold value or lower. A linear circuit, a noise shaping filter for frequency-shaping the level-limited quantization noise obtained at the output of the nonlinear circuit, and a encoder for subtracting the frequency-shaped quantization noise from the input signal. An encoding device having a subtracter provided on the input side.