JPS6032423A - Digital filter - Google Patents

Abstract

PURPOSE:To reduce the propagation of transmission error due to omission of fractions and rounding by rounding or discarding fractions produced by multiplication of filter coefficient, giving a delay to a value decided by quantizing error for a prescribed period and then adding the value to the next input signal. CONSTITUTION:A signal value X generated by a signal generator 1 is subject to subtraction with a forecast value (x) at a subtraction circuit 2 and a quantizer 3 applies quantization in response to the value (X-x) and allocates a code word. A representative is set (4) based thereupon, added (5) to the forecast value (x), forecast values M, N delayed through a 1-picture element delay circuit 6 and a 1-line delay circuit 7 are obtained and the values produce the forecast value (x) through an adder circuit 8 and a 1/2-multiple circuit 9. In this case, a fraction (m) discarded by the 1/2-multiple circuit 9 is subject to 1-picture element delay by a delay circuit 10, the value is doubled into 2m by a double circuit 11, the value is added to the output (M+N) of the circuit 8 and the forecast value is obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a recursive digital filter, and particularly to a configuration of a digital filter suitable for preventing propagation of quantization errors that occur when the feedback coefficient is a non-integer.

[Background of the invention]

As a method for predictively encoding a two-dimensional code such as an image, there are known a one-dimensional encoding method illustrated in FIG. 1(a) and a two-dimensional encoding method illustrated in FIG. 1(b). That is, one-dimensional encoding is run: f
, the pixel to be encoded is predicted using (a combination of) the immediately preceding pixels in the line, and the difference (prediction error) between it and the actual value is encoded. Two-dimensional encoding also uses pixels of the immediately preceding scanning line to make predictions and encodes prediction errors.

Due to the difference in pixel arrangement used for these predictions, both encodings have the following characteristics. That is, since one-dimensional encoding performs prediction using only pixels within a scanning line, prediction errors are generally large. However, when a transmission error occurs, its influence is limited to within the line, so the deterioration in image quality due to this is small. On the contrary, in two-dimensional encoding, the prediction error is one order of magnitude smaller, but when a transmission error occurs, the influence propagates between scanning lines, resulting in a large deterioration in image quality.

Therefore, the effects of transmission errors can be roughly divided into two categories: μm and μm. The first problem occurs even when the total number of the predictive encoders and four coders that perform predictive encoding and decoding is infinitely large. It is impossible to prevent this effect, and the effect becomes smaller depending on the distance of propagation.

The second reason is that the above calculation IT/I If is finite. In other words, since this is finite, the predicted value is truncated or rounded, but because the signal value of one pixel used for prediction differs between the encoder and decoder due to transmission errors, truncation or rounding is not necessary. Results may vary.

The propagation of transmission errors due to these different results is
In principle, it continues indefinitely. The above describes a picture-like predictive coding circuit, but the above coding circuit constitutes a recursive digital filter, and errors due to "rounding" occur in arithmetic processing even in general recursive digital filters. do.

[Purpose of the invention]

An object of the present invention is to realize a means for reducing the propagation of transmission errors due to the above-mentioned truncation and rounding in the finite total E)i of 14 degrees of a recursive digital filter.

[Summary of the Invention] In order to achieve the above object, the present invention provides a recursive digital filter in which the output signal of the filter is multiplied by a coefficient and added to the input signal. It is characterized by adding the value determined from the conversion error to the next input signal after delaying it for a certain period of time.

[Embodiments of the invention]

Hereinafter, the effects of the present invention will be explained using figures. That is, using the pixels M and N shown in FIG. 1(b), the predicted value X of the pixel
An example will be explained in which the prediction is made using equation (1).

However, X, M, N are integers, [ ] is rounded down to an integer, m
@represents the rounded down fraction at the previous pixel M.

FIG. 2 shows an example of how the transmission error propagates when the circled pixel is mistaken as +5 due to a transmission error when all original signals are 0.

However, the information in parentheses is the l[! Shows the rounded down fraction in 11 elements.

For comparison, FIG. 3 also shows how transmission errors propagate when rounding is performed using conventional rounding as in equation (2). If you want to truncate instead of rounding, change the above +5 to -5.
, a result similar to that shown in FIG. 3 is obtained.

Comparing FIG. 2 and FIG. 3, it can be seen that the former converges to the original signal (all 0s) within a finite range, but the latter does not converge, and rather the range expands.

Next, a block diagram of an encoder for embodying the above embodiment is shown in FIG. In the figure, the percentages of the present invention are shown within the dotted lines, and the remaining parts are shown in, for example, [Digital Signal Processing of Images], published by Nikkan Kogyo Shimbun, P147. See Figure 963 for details.

A signal value X generated by a signal generator (for example, an A/D converter) 1 is subtracted from a predicted value X by a subtraction circuit 2 . The quantizer 3 performs quantization according to the jilIK of (X-Xl) and assigns the code @. Based on this code word,
The representative value setting circuit 4 sets a representative value, and the adder 5 calculates the sum with the predicted value X described above. The addition result is delayed through a 1-pixel delay circuit 6 and a 1-line delay circuit 7 to become predicted values M and N. These values M and N are passed through an adder circuit 8 and a 1/2 times circuit 9 to generate a predicted value X.

Next, the features of the present invention within the dotted lines will be explained. First, the fraction m rounded down in the above 172x circuit is the delay circuit 10
is delayed by one pixel, and the value is further doubled by the doubling circuit 11 to become 2m. This value is added to the output (M+N) of the adder circuit 8 by the adder circuit 12, and the value is added to the output (M+N) of the adder circuit 8 using the formula (1
).

The configuration of the decoder is shown in FIG. Blocks with the same numbers as in FIG. 4 have the same functions. Particularly, since the scope of the present invention is common to FIGS. 4 and 5, explanation thereof will be omitted.

It is clear that the following also falls within the scope of the present invention.

(1) In the above embodiment, the prediction coefficients shown in FIG. 1(b) are used, but any other prediction coefficients or prediction pixels may be used. - As an example, 1 shown in Figure 1(a)
The case where the present invention is applied to dimension coding and the case using conventional rounding are explained in Section 6. It can be seen that the predicted music does not converge to 0 with the method shown in FIG. 7, but converges to 0 with the present invention.

(2) The delay circuit 10 is not limited to one pixel; for example, the delay circuit 10 may be delayed by half a pixel, which is the same as the delay circuit 6 which delays the pre-411] value in the direction of the scanning line. Of course, the effect is exactly the same even if the above fraction is added to the signal input to the delay circuit 6 instead of the delay circuit 10.

(3) The above fraction m may be reset to 0 or the like at every predetermined period, such as every scanning line or the beginning of a field.

(4) In order to further reduce the propagation of transmission error 1,
Encoding and decoding may be performed by temporarily treating the scanning line immediately above as completely white or completely black for each field or frame.

In this case, a special control key code may be transmitted in order to synchronize the encoder and decoder.

(5) The above description has been made with reference to image signals, but any signal can be similarly considered. Further, an example has been described in which the difference between each input signal and the predicted value is predictively encoded, but it goes without saying that this predictive encoding is not necessary in a general digital filter.

(6) In the above explanation, the fraction was returned to that number, but
A value determined from this (for example, a value multiplied by 172) may be fed back.

〔Effect of the invention〕

As explained above, in one-dimensional encoding, even if calculations are made using the same number of digits, the predicted value converges to 0 in the present invention, whereas it does not converge in the conventional method. Furthermore, in 42-dimensional encoding, a delay of one scanning line is required, but in the present invention, only the result of truncating fractions needs to be delayed. can also be made smaller. Therefore, it has a great practical effect.

[Brief explanation of drawings]

Figure 1 is an explanatory diagram of one-dimensional encoding and two-dimensional encoding, and Figure 2.
Figure 6 and Figure 3. FIG. 7 is a diagram explaining transmission error propagation in the present invention and the conventional method, respectively, and FIGS. 4 and 5 are block diagrams of a two-dimensional encoder and a decoder, respectively, explaining an embodiment of the present invention. . DESCRIPTION OF SYMBOLS 1... Signal generator, 2... Subtraction circuit, 3... Quantizer, 4... Representative value setting circuit, 5, 8.12... Calculation circuit, 6.10...・Delay (1 pixel) circuit, 7...
Delay (1 run ¥i1 figure 000 θ θ 0 θ Oθ θ −(0)(0)(
0)(0n(θn(0)(0)(θ)(θ)<0)-θ
θ 0 ρ■Z/I θq−−− (θ) (θ) (0n(ρn(θ)(-E)(+2(
θ) (J") (ara θ θ ozzzzrrao- (0) (θ)
) (Zuten (responsibility) θ θ l l Z I I I
Q σ −−−(0) (Oon (of(+n(0) (
+2c sun (0) (responsibility (su) crystal 0 crystal j seat 2. (, crystal:) <'!+ '-σ ρ l l l l θ θ
θ θ (0) (+H's (θ) (θ) (ρ) (+) (t)
(U) (Responsibility) --θ θ l l l θ θ θ θ
θ −1−(0)(su) (0) (1)ノ(0)(
) (+) (zu1) (nano<4)Oθ 11 ρ ρ
θ ρ ρ ′−−−(0λ(4:)(θ)(θ)(f
) (Mawari t-kH woman) (-1 Banano θ 000 θ ρ
ρ t ρ ρ (0n (seeH-k)<-k) (÷) (shi] (+1 (±
) (illusion (responsibility) −−−θ θ θ θ ρ ρ ρ
ρ θ ρ ---(11)(0)(θJ(1)n(θ
(tz) (0 (Q) (θ) (θ2θ θ 0 ρ
θ θ 0 θ 00-0000■3 Z I I +
- θ θ θ 3332222-- 00 Z333 3 333-- 0 1 233 33333-- 1 2333 33 333-- No 4 Figure

Claims (1)

[Claims] 1. In a recursive digital filter that multiplies the output signal of the filter by a non-integer coefficient and recursively adds it to the input signal, quantization error due to rounding or truncating the fraction caused by multiplying by the above non-integer After a certain period of delay, apply the value determined from to the next input signal:
A digital filter characterized by: