JPH0955664A - Coding method and coder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To facilitate signal processing and rate control by applying sequential signal processing to an input signal by a band split process, a quantization process, a coding process, a detection process and a quantization step control process. SOLUTION: The band split process applies band split to an input signal and provides an output as plural band split signals, the quantization process quantizes the band split signal received sequentially in the quantization step based on a quantization step control signal to provide an output of a quantization signal. The coding process encodes the quantization signal. The detection process in parallel with the processing above detects a generated code multi quantity corresponding to a preset band split signal. The quantization step control process provides an output of a quantization step control signal to control the quantization step of other split signals than the prescribed band split signal based on the detected generated code quantity. Then the output is easily rate-controlled (code quantity control).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a coding method and a coding apparatus, and more particularly to a coding method and a coding apparatus for performing image compression.

[0002]

2. Description of the Related Art Conventionally, in image compression, since a high compression rate is obtained, orthogonal transform coding has been used in various video equipment.

However, in image compression by orthogonal transform coding, there is a problem in terms of image quality because block distortion occurs in the restored image. In recent years, subband coding has been proposed to overcome this problem.

Here, the sub-band coding of the input image data using the wavelet transform for band division is shown in FIG.
2 to 14 will be described. The subband coding apparatus 50 is roughly classified into a band division unit 51, a quantizer 52, a quantization unit 53, and a coding unit 5 as shown in FIG.
4 and.

The band division unit 51 divides the input image data D _IN 'into each sub-band data _{DS B} ' by performing wavelet transformation for each field or each frame, and each divided sub-band data _{DS B} ' , Are input to the quantizer 53.

The quantizer 53 has a plurality of quantizers 52, obtains the standard deviation σ of each sub-band data _DSB ′ for each field or each frame, and calculates each subband data _DSB ′. _Are quantized by the given number of bits, and output data D _QSB 'is output.

The output data D _QSB 'is encoded by the encoder 54 and output as encoded data D _C '. In this case, the coding unit 54 uses a two-dimensional Huffman code or the like.

Input image data D in the band division unit 51
FIG. 13 shows the two-dimensional wavelet transform as the band division of _IN '. As shown in FIG. 13, in the second-dimensional wavelet transform, a process of performing one-dimensional subband division in the first direction and further performing one-dimensional subband division in the second direction is performed at the lowest frequency. Subband V ₁ (V ₂ ,
It can be realized by recursively applying to V ₃ ,.

In FIG. 13, symbols "H" and "G" are quadrature mirror filters (QMF) designed based on the wavelet theory, symbol "H" represents a low pass filter, and symbol "H". "G" represents a high pass filter.

In this case, the impulse responses of the low pass filter H and the high pass filter G are h
Assuming that (n) and g (n) are satisfied, g (n) = (-1) ^(1-n) h (1-n) is satisfied.

The symbol "↓ 2" represents 1/2 decimation, and the symbol "[] T" represents transposition of a matrix. The spatial frequency of the image is divided into 10 subbands by performing the above-described wavelet transform for three layers.

In FIG. 14, "W" and "V" indicate subbands, the subscript L indicates a low frequency band, the subscript H indicates a high frequency band, and the numbers indicate layers. For example, the subscript LH indicates a sub band having a low band in the vertical direction and a high band in the horizontal direction.

The image signal thus divided into subbands by the wavelet transform is adaptively quantized by the number of bits given to each subband, and subband coding is performed.

[0014]

In the above-mentioned conventional subband coding, although block distortion does not occur unlike in orthogonal transform coding, it is difficult to make the generated code amount constant (rate control). There was a problem.

Therefore, an object of the present invention is to provide an encoding method and an encoding device which are easy to process and easy to perform rate control (code amount control).

[0016]

In order to solve the above problems, the invention according to claim 1 provides a band division step of band-dividing an input signal and outputting a plurality of band-division signals, and a quantization step control signal. A quantization step of quantizing the band-divided signals sequentially input in the quantization step based on the quantization step and outputting a quantized signal; an encoding step of encoding the quantized signal; and a preset among the band-divided signals. A detecting step of detecting a generated code amount corresponding to a predetermined band-divided signal, and the quantization for controlling a quantization step of a band-divided signal other than the predetermined band-divided signal based on the detected generated code amount. And a quantization step control step of outputting a step control signal.

According to the first aspect of the invention, in the band dividing step, the input signal is band-divided and output as a plurality of band-divided signals. The quantization step quantizes the band-divided signals sequentially input in the quantization step based on the quantization step control signal and outputs the quantized signal.

The encoding step encodes the quantized signal.
In parallel with these, the detecting step detects the generated code amount corresponding to a preset predetermined band-divided signal in the band-divided signal, and the quantization step control step, the predetermined band based on the detected generated code amount. A quantization step control signal for controlling the quantization step of the band division signal other than the division signal is output.

According to a second aspect of the present invention, in the first aspect of the present invention, the input signal is an image signal, the band division step performs band division of a plurality of layers, and the predetermined band division signal is The plurality of band-divided signals corresponding to the plurality of layers are configured to be band-divided signals having frequency components of a high-frequency component in the horizontal direction and a high-frequency component of the vertical method in the same hierarchy.

According to the second aspect of the invention, in addition to the operation of the first aspect of the invention, the band dividing step performs band division of a plurality of layers, and the predetermined band division signal corresponds to a plurality of layers. Since it is a band-divided signal that has frequency components of a high-frequency component in the horizontal direction and a high-frequency component of the vertical method in the same layer among a plurality of band-divided signals, a predetermined band-divided signal is correlated with other band-divided signals Since it has high performance, the quantization step corresponding to other band-divided signals can be set more accurately.

According to a third aspect of the present invention, in the second aspect of the invention, the quantization step control step encodes the predetermined band division signal of the nth layer (n: integer of 2 or more). The quantization step of the band-divided signal of the (n-1) th layer is set based on the code amount of.

According to the invention of claim 3, in addition to the operation of the invention of claim 2, the quantization step control step is
Since the quantization step of the band-divided signal of the (n-1) -th layer is set based on the code amount when the predetermined band-divided signal of the n-th layer (n: an integer of 2 or more) is encoded, It is possible to accurately set the quantization step corresponding to another band-divided signal by using the relationship having high correlation.

According to a fourth aspect of the present invention, in the third aspect of the invention, the quantization step control step sets the quantization step from the predetermined band division signal in the highest layer of the plurality of layers. It is configured to start and set the quantization step toward the band-divided signals of lower hierarchical layers.

According to the invention described in claim 4, in addition to the operation of the invention described in claim 3, the quantization step control step includes:
The quantization step is set from the predetermined band-divided signal in the highest layer of the multiple layers, and the quantization step is set toward the band-divided signal in the lower layers in sequence. The quantization step can be set from the band-divided signal having a large error, and the code amount can be made constant more easily.

According to a fifth aspect of the present invention, in the third aspect of the invention, the quantization step control step sets the quantization step from the predetermined band division signal in the lowest layer of the plurality of layers. It is configured to start and set the quantization step toward the band-divided signals of higher layers in sequence.

According to the invention of claim 5, in addition to the operation of the invention of claim 3, the quantization step control step is
The quantization step is set from the predetermined band-divided signal in the lowest layer of the multiple layers, and the quantization step is set toward the band-divided signal in the higher layers one by one, which affects the image quality during restoration. The quantization step can be set from a band-divided signal having a large value, and the code amount can be made constant while reducing the deterioration of image quality.

According to a sixth aspect of the present invention, a band dividing means for band-dividing an input signal and outputting the band-divided signal as a plurality of band-divided signals, and the band-divided signal sequentially inputted in a quantization step based on a quantization step control signal. A plurality of quantizing means for quantizing and outputting a quantized signal; an encoding means for encoding the quantized signal; and a generated code amount corresponding to a preset band-divided signal of the band-divided signals. And a quantization step control means for outputting the quantization step control signal for controlling the quantization step of a band division signal other than the predetermined band division signal based on the detected generated code amount. And, and is configured.

According to the sixth aspect of the present invention, the band division means divides the input signal into bands and outputs it to the quantization means as a plurality of band division signals. Each quantizing unit quantizes the band-divided signals sequentially input in the quantizing step based on the quantizing step control signal and outputs the quantized signal to the encoding unit, and the encoding unit encodes the quantized signal. To do.

In parallel with this, the detecting means detects the generated code amount corresponding to a preset predetermined band-divided signal in the band-divided signal, and the quantization step control means based on the detected generated code amount. And outputs a quantization step control signal for controlling the quantization step of the band division signal other than the predetermined band division signal to the quantization means.

According to a seventh aspect of the invention, in the invention according to the sixth aspect, the input signal is an image signal, the band division means performs band division of a plurality of layers, and the predetermined band division signal is The plurality of band-divided signals corresponding to the plurality of layers are configured to be band-divided signals having frequency components of a high-frequency component in the horizontal direction and a high-frequency component of the vertical method in the same hierarchy.

According to the invention of claim 7, in addition to the operation of the invention of claim 6, the band dividing means performs band division of a plurality of layers, and the predetermined band division signal corresponds to a plurality of layers. Since it is a band-divided signal that has frequency components of a high-frequency component in the horizontal direction and a high-frequency component of the vertical method in the same layer among a plurality of band-divided signals, a predetermined band-divided signal is correlated with other band-divided signals Since it has high performance, the quantization step corresponding to other band-divided signals can be set more accurately.

According to an eighth aspect of the invention, in the seventh aspect, the quantization step control means encodes the predetermined band division signal of the nth layer (n: integer of 2 or more). The quantization step of the band-divided signal of the (n-1) th layer is set based on the code amount of.

According to the invention of claim 8, in addition to the operation of the invention of claim 7, the quantizing step control means comprises:
Since the quantization step of the band-divided signal of the (n-1) -th layer is set based on the code amount when the predetermined band-divided signal of the n-th layer (n: an integer of 2 or more) is encoded, It is possible to accurately set the quantization step corresponding to another band-divided signal by using the relationship having high correlation.

According to a ninth aspect of the present invention, in the eighth aspect, the quantization step control means sets the quantization step from the predetermined band division signal in the highest layer of the plurality of layers. It is configured to start and set the quantization step toward the band-divided signals of lower hierarchical layers.

According to the invention of claim 9, in addition to the operation of the invention of claim 8, the quantization step control means is:
The quantization step is set from the predetermined band-divided signal in the highest layer of the multiple layers, and the quantization step is set toward the band-divided signal in the lower layers in sequence. The quantization step can be set from the band-divided signal having a large error, and the code amount can be made constant more easily.

According to a tenth aspect of the present invention, in the eighth aspect, the quantization step control means sets the quantization step from the predetermined band division signal in the lowest layer of the plurality of layers. It is configured to start and set the quantization step toward the band-divided signals of higher layers in sequence.

According to the invention of claim 10, claim 8 is provided.
In addition to the operation of the described invention, the quantization step control means starts the setting of the quantization step from a predetermined band division signal in the lowest layer among the plurality of layers, and sequentially proceeds to the band division signal in the higher layers. Since the quantization step is set, the quantization step can be set from the band-divided signal that has a greater influence on the image quality at the time of restoration, and the code amount can be made constant while reducing the deterioration of the image quality.

[0038]

Preferred embodiments of the present invention will now be described with reference to the drawings. In subband coding,
After subband analysis, quantization and variable length coding are performed for each subband.

At this time, it is empirically known that the relationship between the quantization step Q and the generated code amount B can be expressed as ln (B) ∝ln (Q) (1).

The meaning of this equation is that if the quantization step Q is made coarse, the code amount B will decrease, and if the quantization step Q is made fine, the code amount will increase. Furthermore, it is also shown that the quantization step Q and the code amount B are in a proportional relationship on the logarithmic graph.

Now, the result of the experiment based on the above consideration will be described. In the experiment, the target image "Mobi
As shown in FIG. 13, the spatial frequency of the image is divided into 10 sub-bands by using "le &Calendar" as shown in FIG.

Of the wavelet transform coefficients of the 10 subbands, the wavelet transform coefficients corresponding to 9 subbands except the subband W _{HH 1} are quantized by changing the quantization step, and the variable length is changed. The amount of code generated by encoding was examined.

As a result, the results when the predetermined reference quantization step is changed by 0.1 to 1.0 to 5.9 times are shown in FIGS. 1 to 3, the vertical axis represents the logarithm ln (B) of the generated code amount, and the horizontal axis represents the logarithm of the quantization step normalized based on a predetermined reference quantization step (Qdefault) for each subband. ln (Qsubband
) Represents.

For example, in FIG. 1, Qu of subband V ₃
If the value of antizer Scale is Q, the quantization step Q _V3 of subband V ₃ is

[0045]

[Equation 1] It is. The following can be seen from the above experimental results.

[0046] a) the slope of the graph of the sub-band W slope and sub-band W _LHj of the graph of _HHj are very similar. However, the subscript j represents a layer (hereinafter the same). b) The slopes of the graphs of the subband V ₃ , the subband W _LH3 , the subbands W _HL3 , and W _HL2 are almost constant regardless of the quantization step Q.

C) In the same layer j and normalized quantization step Q, the code amount of subband W _HLj is generally larger than the code amounts of other subbands. From these, it was found that the quantization step Q and the code amount B have a negative correlation regardless of the subband.

Therefore, it is considered that the equation (1) is approximated by a straight line using a coefficient a representing the inclination of the straight line and a coefficient b representing the intercept of the straight line. More specifically, it is approximated by ln (B) = − a × ln (Q) + b (4) or B = (e ^b / Q ^a ) (5).

The equation (4) is regarded as a linear regression model with 1.0 to 5.9 of the normalized quantization step Q as the applicable range, and the coefficient is calculated for each sub-band by using the linear regression analysis method. a and the coefficient b were obtained.

FIG. 4 shows the regression standard error V _e with respect to the obtained coefficients a and b and the code amount in the run length coding. Subband W _LHj belonging to the hierarchy j as shown in FIG. 4, sub-band W _HLJ, Arranging the order subbands W _HHj regression standard error V _e is _{small, W HLj <W LHj <W} HHj ... (6) Become.

[0051] The sub-band W _HLJ regression standard error V _e is small, larger variance of sub-band W _HLJ is, many wavelet transform coefficients having a value that is not zero even if coarse quantization step Q, runlength It shows that the coding efficiency is not so much improved.

On the other hand, subband W _LHj and subband W
The large regression standard error V _e of _HHj, the sub-band W
_{When the variance of LHj} and the subband _WHHj is small and the quantization step Q becomes coarse, there are many wavelet transform coefficients having a value of zero, which shows that the efficiency of run-length coding is improved.

Therefore, this time, conversely, let us consider obtaining the quantization step Q from the target code amount B. First,
Equation (5) is modified as follows.

B = e ^b / Q ^a (7)

[0055]

[Equation 2] If the variables are converted into a = 1 / a and b = 1 / b, then Q = e ^b / B ^a (9)

Therefore, if the coefficient a and the coefficient B can be determined for each subband, the quantization step Q can be obtained by giving the target code amount B, and the rate control can be realized. However, the coefficient a and the coefficient b for each sub-band obtained here are values that are valid only for the corresponding image.

Next, a method of determining the coefficient a and the coefficient b will be described. FIG. 5 shows the results of investigating the coefficient a and the coefficient b by regression analysis for various 50 types of images and examining the correlation between the coefficient a and the coefficient b.

According to this result, regardless of the type of image,
It can be seen that the correlation between the coefficient a and the coefficient b is high. That is,
The slope and intercept of the straight line representing the relationship between the code amount B and the quantization step Q are not independent but −a∝b ... (10), so if only one of the coefficient a and the coefficient b can be known, the purpose is The quantization step Q can be obtained from the code amount B.

By the way, in the case of performing rate control of one pass, assuming that there is no foreseeing information, in order to efficiently and accurately obtain the coefficient a or the coefficient b, the first half and the second half are as follows. It is conceivable to perform the processing in two stages separately.

1) The quantization step Q is fixed for a predetermined subband to be processed in the first half, and variable length coding is performed. 2) In the other subbands to be processed in the latter half, the quantization step Q is variable, the quantization step Q is set based on a predetermined parameter, and variable length coding is performed.

In this case, considering the one-pass rate control, the code amount after variable length coding in the processing of the first half subband is used as the above-mentioned predetermined parameter.

By the way, since the quantization step Q is fixed in the processing of the first half subband, in some cases, the rate over (the rate [code amount] rather than the predetermined rate [code amount] is exceeded by only the first half process. ] Is large).

In order to avoid this, the predetermined subband in the processing of the first half subband needs to satisfy the following two conditions. Condition 1) Even if the quantization step Q is rough, the sub-band is such that image quality deterioration is not so noticeable. That is, since the fixed quantization step is used, the coding amount is reduced as much as possible.

Condition 2) The detected code amount can be used for controlling the generated code amount of other subbands. In other words, it must be correlated with the amount of generated code in other subbands. Similarly, it is desirable that the predetermined subband in the processing of the latter subband satisfy the following condition.

Condition) It is possible to realize rate control with high accuracy by using the code amount of a predetermined subband obtained in the processing of the first half subband. When the variance σ for each subband is examined in the same layer j to satisfy the above conditions, σW _HLj > σW _LHj > σW _HHj .

Therefore, it can be expected that the subband W _HHj has a small code amount. Further, in the frequency domain, the frequency component of the subband W _HH is adjacent to the frequency component of the subband W _{LH and} the frequency component of the subband W _HL , and in the wavelet transform, band division is performed by the filter bank, so that they are adjacent to each other. There is a slight correlation between the frequency components of the existing subbands.

Therefore, it can be expected that the code amount of the subband W _HH can be used for rate control of the subband W _LH and the subband W _HL . From these viewpoints, the generated code amount of the subband W _HH , the subband W _LH , and the subband W
_The results of examining the correlation between the _HL coefficient a and the coefficient b are shown in FIGS. 6 to 8. Note that the correlation is 0.4 in FIGS.
The following items are not shown.

In this case, two subbands W
2 instead of encoding _LH and subband W _HL separately
The subband W _LH and the subband W _HL are collectively quantized, and the coefficient a and the coefficient b are obtained for the sum of the code amounts generated by the variable length coding. By thus handling the two subbands W _LH and W _HL together, it is possible to reduce the rate control parameters and avoid extreme deterioration in image quality.

From the above result, it is assumed that the number of layers is j, and the layer j
Two subbands W _LHj and subband W _{HLj in}
The coefficient a a _j, if the coefficient b and b _j in, 1) between the coefficient b _j and the code amount of the sub-band W _HHj than the correlation between the coefficient a _j and code quantity of the sub-band W _HHj Has a higher correlation.

2) Subband W _HLj and subband W
_The subband W is highly correlated with the coefficient b _{j for LHj.
It is} the code amount of the subband _{WHH (j + 1)} rather than the code amount of _HHj . That is, the code amount bitw of the subband W _{HH (j + 1)}
There is a relationship of b _j ∝bitw _{HH (j + 1)} ... (11) between _{HH (j + 1)} and the coefficient b _j for the sub-band W _HLj and the sub-band W _LHj . I can understand that.

A procedure for controlling the code generation amount (rate control) of the subband W _LHj and the subband W _HLj of the layer j obtained from the above results will be described below. Procedure 1) First, the subband W _{HH (j + 1)} is quantized by a fixed quantization step Q _CONST , and variable length coding is performed to obtain a code amount bitwHH _{(j + 1)} .

Procedure 2) From the equation (11), the code amount bitw
_{HH (j + 1)} determine the coefficients b _j from (10) coefficient from Equation b _j
The coefficient a _j is calculated from Step 3) The code generation amount of the subband W _LHj and the subband W _HLj is a desired value bitw _LHj · w _HLj from the equation (9).
The quantization step Q normalized so that

Procedure 4) Subband W according to equation (2)
Quantization step Q corresponding to _LHj and subband _WHLj
Find w _LHj and quantization step Qw _HLj . Procedure 5) Subband W _LHj and subband W using quantization step Qw _LHj and quantization step Qw _HLj
HLj is quantized and variable-length coded.

In this case, for example, in the wavelet transform of three layers, there is no subband W _HH4. Therefore, when the layer j = 3, the subband V ₃ is used instead of the subband W _HH4 .

By performing the processing in the above-described manner in this way, the target code amount bitw _LHj for each layer j
・ While w _HL j can be obtained, rate control can be easily realized.

As a result, according to the present embodiment, the generated code amount can be kept substantially constant regardless of the type of image signal.

[0077]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, preferred embodiments of the present invention will be described with reference to FIGS.
This will be described with reference to FIG. First Embodiment FIG. 9 shows a block diagram of a subband coding device.

In the following description, for simplification of the description, each subband when the wavelet transform of three layers is performed is specified by the number in [] of FIG. For example, subband W _LH1 is called subband [7], subband W _HL2 is called subband [5], and subband V ₃ is called subband [0].

Along with this, for example, coefficient a ₁₂ and coefficient b
Reference numeral ₁₂ represents the coefficient a and the coefficient b in the case of collectively rate controlling the subband [1] and the subband [2]. In addition, the subband [9] is not the target of encoding in this embodiment.

As shown in FIG. 9, the sub-band coding apparatus 10 performs wavelet conversion on the input image signal G _IN in field units or frame units to divide it into sub-bands, and blocks each sub-band. Output as a plurality of sub-band signals S _SB (n) (n: 0 to 8, hereinafter the same), and a sub-band signal S _SB (n for each sub-band divided by the band dividing unit 11. ) _Is quantized and output as a quantized output signal S _SQ (n) 9
The number of quantizers Q _{0 to} Q ₈ and the quantized output signals S _SQ (n) input by the quantizers Q _{0 to} Q ₈ are coded and output as coded signals S _C. Coded amount detection signal S _bit (n) indicating the generated code amount for each band
Is output to the step number control unit 12 described later.
And a step number controller for outputting a quantization step control signal S _QC (n) for controlling the quantization step in each of the quantizers Q _{0 to} Q ₈ based on the encoding amount detection signal S _bit (n) It is composed of 12.

In the above configuration, quantizers Q ₀ , Q ₃
, And the quantization step in Q ₆ is fixed, and the quantization steps in the quantizers Q ₁ and Q ₂ are commonly controlled by the quantization step control signal S _QC (1, 2). Also, the quantization steps in the quantizers Q ₄ and Q ₅ are
The quantization step control signals S _QC (4, 5) are commonly controlled, and the quantization steps in the quantizers Q ₇ and Q ₈ are commonly controlled by the quantization step control signals S _QC (7, 8). Controlled by.

Next, a processing flowchart of the first embodiment is shown in FIG. First, subband [0], subband [3], and subband [6] are quantized by a fixed quantization step, and variable length coding is performed (step S1).

More specifically, when the input image signal G _IN is input to the band division unit 11, the band division unit 11 performs wavelet transform on a field unit basis or a frame unit basis and subbands [0] to [8]. ] And corresponding quantizers Q _{0 to} Q ₈ as a plurality of subband signals S _SB (n)
Output to Of these, the subband signals S _SB (0), S
_SB (3) and S _SB (6) are quantized by fixed quantization steps defined respectively, and quantized output signal S
Encoding unit 1 as _SQ (0), S _SQ (3) and S _SQ (6)
3 is output. Then, the encoding unit 13 encodes the input quantized output signals S _SQ (0), S _SQ (3), and S _SQ (6), and generates the result of encoding each quantized output signal. The code amount is the code amount detection signal S _bit (0), S
Step number control unit 1 as _bit (3) and S _bit (6)
Output to 2.

Next, subband [0] and subband [3]
And the coding amount detection signals S _bit (0), S corresponding to the generated code amount generated as a result of the coding of the subband [6]
_{Based on bit} (3) and S _bit (6), the step number control unit 12 causes the coefficient a ₁₂ and the coefficient b ₁₂ , the coefficient a _45, and the coefficient b.
₄₅ , coefficient a ₇₈ and coefficient b ₇₈ are calculated (step S
2). Here, since the function of the coefficients and the generated code amount is the aforementioned relationship, the encoding amount detection signal S _{bi t} (0) subband [1] and [2], encoded amount detection signal S _bit ( 3)
To the subbands [4] and [5], the coefficients a and b of the subbands [7] and [8], that is, the generated code amount can be predicted from the coding amount detection signal S _bit (6).

In this case, the coefficients a and b are calculated based on an empirical formula obtained by performing regression analysis beforehand using a plurality of completely different images (input image signal G _IN ).

Subsequently, the roughness variable SCALE is calculated and the code amount is assigned (step S3). Here, the remaining subbands ([1] and [2], [4] and [5] or [7] are obtained by subtracting the generated code amount of each subband in step S1 from the preset total code amount. ] And [8]) are assigned.

More specifically, the coarseness variable SCALE times for the quantization steps Q _n ^default preset for the subbands [1] and [2], [4] and [5] and [7] and [8]. Roughly, and using the set quantization step, the code amount is trial-calculated for each subband [1] and [2], [4] and [5] or [7] and [8], and the total code amount is calculated. Until the value is within the preset code amount, the roughness variable SCALE is changed (roughened), and the iterative calculation is performed while roughening the quantization step. From the calculation results, the subbands [1] and [2], [ 4] and [5], [7] and [8] are determined.

Next, based on the result of step S3, the quantization step control signal S _QC (7, 7) for the subband [7] and the subband [8] is output from the step number controller 12
8) is output to the corresponding quantizers Q ₇ and Q ₈ for quantization, and the coding unit 13 performs variable length coding (step S4).

Next, subbands [0], [3],
Based on the generated code amounts of [6], [7] and [8], the quantization steps of subbands [1], [2], [4] and [5] are obtained again, and similarly, 4] and [5] are quantized and variable length coding is performed (step S
5). Then, the subbands [0], [3], [6],
Based on the generated code amounts of [7] and [8], [4] and [5], the quantization steps of subbands [1] and [2] are obtained again, and the subbands [1] and [2] are quantized. Quantization and variable length coding (step S6) are performed, and the process ends. Here, the reason why the quantization step is calculated again in steps S5 and S6 is to absorb the error between the generated code amount based on the actually set quantization step and the previously predicted generated code amount (steps S3 and S5). Is.

As a result, according to the first embodiment, the code amount of the obtained encoded signal S _C can be kept substantially constant regardless of the type of image signal. Second Embodiment In the above first embodiment, the rate control is performed from the high frequency side (subbands [7] and [8] side), but in the second embodiment, the rate control is performed from the low frequency side ( Subband [1], [2] side).

Next, FIG. 11 shows a processing flowchart of the second embodiment. First, similarly to step S1 of the first embodiment, subbands [0], [3] and [6] are quantized and variable length coded (step S11).

Then, the coded amount detection signal S _bit corresponding to the generated code amount of the subbands [0], [3] and [6]
(0), S _bit (3) and S _bit (6) are output to the step number control unit 12.

Then, similarly to step S2 of the first embodiment,
The step control unit 12 receives the input encoding amount detection signal S
Coefficient a ₁₂ and coefficient b ₁₂ , coefficient a ₄₅ and coefficient b ₄₅ , coefficient a based on _bit (0), S _bit (3) and S _bit (6)
₇₈ and the coefficient b ₇₈ are calculated (step S1)
2).

Subsequently, the coding amount detection signals S _bit (0), S _bit (3) and S are calculated from the total allowable code amount.
_The amount obtained by subtracting the generated code amount corresponding to _bit (6) is used as the assigned code amount to other subbands in step S of the first embodiment.
The roughness variable SCALE is changed in the same manner as in 3, and the quantization steps of the subbands [1] and [2], [4] and [5], [7] and [8] are selected.

Based on the result of step S13, subbands [1] and [2] are quantized by the selected quantization step, and the coding unit 13 performs variable length coding (step S14). Next, based on the results of steps S11 and S14, the quantization steps of subbands [4], [5], [7], and [8] are recalculated as in step S13. The allocated code amount at this time is calculated from the allowable total generated code amount to subbands [0], [3],
The generated code amount of [6], [1] and [2] is subtracted. By the quantization step recalculated in this way, the subbands [4] and [5] are quantized and variable length coding is performed. (Step S15). Next, step S1
1, the quantization step of subbands [7] and [8] is recalculated based on the results of S14 and S15, and the quantization and variable length coding of subbands [7] and [8] are performed (step S16). ).

Next, it is determined whether or not the total code amount actually obtained is within the preset code amount, that is, whether or not the rate is over (step S17). In this case, it is determined whether or not the rate is over, as compared with the case of the first embodiment, when the prediction error code amount of the rate control of the subbands [7] and [8] is the subband [1. ] And [2] are larger than the error of the rate control, and the probability of rate over is high.

Therefore, when the rate is over, the subband quantized signal S _SQ (7) corresponding to the subband [7] is not used for encoding (step S18).

As a result, according to the second embodiment, the code amount of the obtained encoded signal S _C can be kept substantially constant regardless of the type of image signal. Here, in step S18, the subband that is not used for encoding is the subband [7], but the subband [7] is not limited to this.
And [8] may be subbands not used for encoding.

Further, as compared with the case of the first embodiment, the code amount is preferentially assigned to the sub-bands on the low frequency side (sub-band [1], sub-band [2]) which have a great influence on the image quality. Therefore, the image at the time of restoration is of higher quality.

[0100]

As described above, according to the first aspect of the present invention, the quantizing step quantizes the band division signals sequentially input in the quantizing step based on the quantizing step control signal. The signal is output, the encoding step encodes the quantized signal, and in parallel with these, the detection step detects the generated code amount corresponding to a preset predetermined band division signal of the band division signal, Since the quantization step control process outputs a quantization step control signal for controlling the quantization step of a band-divided signal other than the predetermined band-divided signal based on the detected code amount, rate control (code amount control) is performed. Easy to do. Furthermore, processing can be performed in one pass, and it is easy to implement by hardware.

According to the invention described in claim 2, in addition to the effect of the invention described in claim 1, the band division step performs band division of a plurality of layers, and the predetermined band division signal corresponds to a plurality of layers. Since it is a band-divided signal that has frequency components of a high-frequency component in the horizontal direction and a high-frequency component of the vertical method in the same layer among a plurality of band-divided signals, a predetermined band-divided signal is correlated with other band-divided signals. The rate of quantization is high, the quantization step corresponding to other band division signals can be set more accurately, and more reliable rate control can be performed.

According to the invention of claim 3, in addition to the effect of the invention of claim 2, the quantization step control step is
Since the quantization step of the band-divided signal of the (n-1) -th layer is set based on the code amount when the predetermined band-divided signal of the n-th layer (n: an integer of 2 or more) is encoded, The quantization step corresponding to another band-divided signal can be accurately set by using the relationship with high correlation, and the rate control can be performed more easily.

According to the invention of claim 4, in addition to the effect of the invention of claim 3, the quantization step control step sets the quantization step from a predetermined band division signal in the highest layer of a plurality of layers. Since it starts, the quantization step is set toward the band-divided signal of the lower hierarchy in sequence,
The quantization step can be set from the band-divided signal with the largest error in quantization and encoding, and the code amount can be made constant more easily.

According to a fifth aspect of the invention, in addition to the effect of the third aspect of the invention, the quantization step control step sets the quantization step from the predetermined band division signal in the lowest layer of a plurality of layers. Since the quantization step is set toward the band-divided signal of the higher hierarchy in sequence, the quantization step can be set from the band-divided signal that has a greater effect on the image quality at the time of restoration, and the deterioration of the image quality is reduced. Meanwhile, the code amount can be made constant.

According to the invention described in claim 6, each quantizing means quantizes the band-divided signals sequentially inputted in the quantizing step based on the quantizing step control signal and outputs the quantized signal to the coding means. Then, the encoding means encodes the quantized signal, and in parallel with this, the detecting means detects the generated code amount corresponding to a preset predetermined band-divided signal in the band-divided signal, and performs the quantization step control. The means outputs a quantization step control signal for controlling the quantization step of the band-divided signals other than the predetermined band-divided signal based on the detected code amount to the quantizing means, so that the rate control (code amount control) Can be done easily. Furthermore, the processing can be performed in one pass, and it is easy to realize with hardware.

According to the invention of claim 7, in addition to the effect of the invention of claim 6, the band dividing means performs band division of a plurality of layers, and the predetermined band division signal corresponds to a plurality of layers. Since it is a band-divided signal that has frequency components of a high-frequency component in the horizontal direction and a high-frequency component of the vertical method in the same layer among a plurality of band-divided signals, a predetermined band-divided signal is correlated with other band-divided signals. The rate of quantization is high, the quantization step corresponding to other band division signals can be set more accurately, and more reliable rate control can be performed.

According to the invention of claim 8, in addition to the effect of the invention of claim 7, the quantization step control means is:
Based on the code amount when the predetermined band-divided signal of the n-th layer (n: integer of 2 or more) is encoded, the (n-1) th
Since the quantization step of the band division signal of the hierarchy is set,
The quantization step corresponding to another band-divided signal can be accurately set by using the relationship having higher correlation, and the rate control can be performed more easily.

According to the invention of claim 9, in addition to the effect of the invention of claim 8, the quantization step control means is:
The quantization step is set from the predetermined band-divided signal in the highest layer of the multiple layers, and the quantization step is set toward the band-divided signal in the lower layers in sequence. The quantization step can be set from the band-divided signal having a large error, and the code amount can be made constant more easily.

According to the invention of claim 10, claim 8 is provided.
In addition to the effect of the invention described, the quantization step control means starts setting the quantization step from a predetermined band division signal in the lowest layer among the plurality of layers, and sequentially proceeds toward the band division signal in higher layers. Since the quantization step is set, the quantization step can be set from the band-divided signal that has a greater influence on the image quality at the time of restoration, and the code amount can be made constant while reducing the deterioration of the image quality.

[Brief description of drawings]

1 is a diagram illustrating the relationship between the sub-band _{V 3, W LH3, W HL3} , the code amount of W _HH3 and normalized quantization step.

FIG. 2 is a diagram illustrating the relationship between the code amount of subbands W _LH2 , W _HL2 , and W _HH2 and the normalized quantization step.

FIG. 3 is a diagram for explaining the relationship between the code amounts of subbands W _LH1 and W _HL1 and the normalized quantization step.

FIG. 4 is an explanatory diagram of a relationship between coefficients a and b and regression standard error V _e .

FIG. 5 is a diagram illustrating a correlation between coefficient a and coefficient b.

FIG. 6 is a diagram illustrating a correlation between a code generation amount of a subband and a coefficient a.

FIG. 7 is a diagram for explaining the correlation between the generated code amount of a subband and the coefficient b.

FIG. 8 is a diagram showing the amount of generated code of subbands and coefficients aj and b
It is a figure explaining the correlation of j.

FIG. 9 is a block diagram showing a schematic configuration of a subband encoding device.

FIG. 10 is a processing flowchart of the first embodiment.

FIG. 11 is a processing flowchart of the second embodiment.

FIG. 12 is a block diagram showing a schematic configuration of a conventional subband encoding device.

FIG. 13 is a diagram illustrating a two-dimensional wavelet transform in a band division unit.

FIG. 14 is an explanatory diagram of a spatial frequency division state of an image when wavelet transformation is performed in three layers.

[Explanation of symbols]

10 ... sub-band coding apparatus 11 ... band dividing section 12 ... step number control unit 13 ... encoding unit _{Q 0, Q 1, Q 2} , Q 3, Q 4, Q 5, Q 6, Q 7, Q
₈ ... Quantizer G _IN ... input image signal S _SB (n) ... sub-band signal S _SQ (n) ... quantized output signal S _C ... coded signal S _bit (n) ... coded amount detection signal S _QC ... Quantization step control signal

Claims (10)

[Claims] 1. A band division step of band-dividing an input signal and outputting it as a plurality of band-divided signals, and quantizing and quantizing the band-divided signals sequentially inputted in a quantization step based on a quantization step control signal. A quantization step of outputting a signal; an encoding step of encoding the quantized signal; a detection step of detecting a generated code amount corresponding to a preset predetermined band division signal among the band division signals; A quantization step control step of outputting the quantization step control signal for controlling a quantization step of a band division signal other than the predetermined band division signal based on the detected generated code amount, Encoding method. 2. The encoding method according to claim 1, wherein the input signal is an image signal, the band division step performs band division of a plurality of layers, and the predetermined band division signal is divided into a plurality of layers. An encoding method, which is a band-divided signal having frequency components of a high-frequency component in a horizontal direction and a high-frequency component of a vertical method in the same layer among a plurality of corresponding band-divided signals. 3. The encoding method according to claim 2, wherein the quantization step control step determines an amount of code when the predetermined band division signal of the nth layer (n: integer of 2 or more) is encoded. An encoding method characterized by setting a quantization step of a band-divided signal of the (n-1) th layer based on the above. 4. The encoding method according to claim 3, wherein the quantization step control step starts setting the quantization step from the predetermined band division signal in the highest layer of the plurality of layers, and sequentially. An encoding method, wherein the quantization step is set toward a band-divided signal of a lower hierarchy. 5. The encoding method according to claim 3, wherein the quantization step control step starts setting the quantization step from the predetermined band division signal in the lowest layer of the plurality of layers, and sequentially. A coding method characterized in that a quantization step is set toward a band-divided signal in a higher hierarchy. 6. Band division means for band-dividing an input signal and outputting it as a plurality of band-divided signals, and quantizing and quantizing the band-divided signals sequentially inputted in a quantization step based on a quantization step control signal. A plurality of quantizing means for outputting a signal, an encoding means for encoding the quantized signal, and a detecting means for detecting a generated code amount corresponding to a preset predetermined band division signal of the band division signals A quantization step control means for outputting the quantization step control signal for controlling a quantization step of a band division signal other than the predetermined band division signal on the basis of the detected generated code amount, An encoding device characterized by. 7. The encoding device according to claim 6, wherein the input signal is an image signal, the band division unit performs band division of a plurality of layers, and the predetermined band division signal is divided into a plurality of layers. An encoding device, which is a band-divided signal having frequency components of a high-frequency component in a horizontal direction and a high-frequency component of a vertical method in the same layer among a plurality of corresponding band-divided signals. 8. The encoding device according to claim 7, wherein the quantization step control unit determines a code amount when the predetermined band-divided signal of the n-th layer (n: integer of 2 or more) is encoded. An encoding apparatus which sets a quantization step of a band-divided signal of the (n-1) th layer based on the above. 9. The encoding device according to claim 8, wherein the quantization step control means starts setting the quantization step from the predetermined band division signal in the highest layer of the plurality of layers, and sequentially. An encoding apparatus, wherein the quantization step is set toward a band-divided signal of a lower hierarchy. 10. The encoding device according to claim 8, wherein the quantization step control means starts setting the quantization step from the predetermined band division signal in the lowest layer of the plurality of layers, and sequentially. An encoding apparatus, wherein a quantization step is set toward a band-divided signal in a higher hierarchy.