JP3006324B2 - Subband image encoding apparatus and method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for encoding and storing images.

[0002]

2. Description of the Related Art When storing image information, encoding is generally performed to reduce the amount of data. At the time of this encoding, it is advantageous from the point of signal processing that the data amount of the encoded image information is constant regardless of the content of the image. Therefore, several techniques have been proposed for controlling the code amount of an image to be constant regardless of the content of the image.

As a conventional image code amount control technique, for example, the technique described in D-45, 1989, IEICE Fall, 1989 is known. Hereinafter, the configuration of the image coding apparatus described in the document will be described with reference to FIG. The input image 201 is divided into N × N blocks by a blocking circuit 202, and a discrete cosine transform circuit (shown by a DCT circuit in the figure) 2
At 03, discrete cosine transform is performed for each block. The transform coefficient subjected to the discrete cosine transform is temporarily stored in the memory circuit 204
And is quantized by the quantization circuit 205 by a plurality of different quantization step values, and is variable-length coded by the variable-length coding circuit 206. The quantization step measurement / estimation circuit 207 estimates a quantization step value that satisfies the set code amount from the code amounts based on the plurality of quantization step values, and performs quantization using the estimated quantization step value.
And, variable length coding is performed. Further, by performing estimation again using the estimated quantization step value, accurate estimation becomes possible.

[0004] As a similar code amount control technique, for example, a "rate adaptive DCT coding method based on the statistical properties of an image" (IEICE, BI Vol. J75-B-).
I, No. 5, pp. 353-361, 1992 5
Month). Hereinafter, the configuration of the image coding apparatus described in the document will be described with reference to FIG. Input image 2
11 is divided into N × N blocks by a blocking circuit 212 and is subjected to discrete cosine transform for each block by a DCT circuit 213. At the same time, the quantization step estimation circuit 214 calculates the activity of each block, and predicts the quantization step value of each block based on the activity. The activity is an index indicating the complexity of an image, such as the sum of squares of signals in each block, the sum of absolute values, and the like. The transform coefficient of each block is quantized by the quantization circuit 215 using the predicted quantization step value, and is variable-length coded by the variable length coding circuit 216. In this method, the quantization step estimation circuit 2
Since it is difficult to make the processing time in the DCT circuit 213 equal to the processing time in the DCT circuit 213, it is necessary to have a memory for storing the transform coefficients in the subsequent stage of the DCT circuit 213. Alternatively, it is necessary to have a memory for storing the input image and to perform the discrete cosine transform by the DCT circuit 213 after the estimation of the quantization step is completed.

[0005]

In the above two conventional image coding apparatuses, since discrete cosine transform is used, coding is performed independently for each block of the coding unit. Therefore, there is a problem that block distortion occurs and image quality deteriorates.

In both of the conventional systems, the relationship between the quantization step of a transform coefficient, which is a parameter affecting the code amount, and the code amount is checked in advance, and the code amount is controlled using this. In these methods, variable-length coding is used, and a control error in the code amount always occurs due to local fluctuation of an image. Therefore, when the code amount overflows, the code amount is forcibly included in the allocated code amount. In order to suppress this, there is a problem that the image quality deteriorates. In order to prevent this, it is necessary to give a sufficient margin to the target code amount, and as a result, there is a problem that the coding performance is deteriorated.

[0007] Also, the above-mentioned 1989 Dept.
In the technique described in No. 5, it is necessary to repeat the encoding by changing the parameters until the error becomes sufficiently small. For this reason,
It becomes difficult to control the encoding processing time.

Further, since the above two conventional image coding apparatuses perform variable-length coding, the coding processing time cannot be controlled, and an image is cut out and rotated without changing the code information. And the like, there is a problem that editing processing is difficult.

[0009] Therefore, the present invention provides a fixed encoding operation amount.
It is an object of the present invention to provide an image encoding device in which encoding processing time can be predicted or set in advance. It is another object of the present invention to provide an image coding apparatus which has a sufficiently high efficiency with a fixed code amount, and which can easily edit an image with the code information as it is.

[0010]

In order to achieve the above object, a subband image encoding apparatus according to the present invention comprises: an image storage means for storing an image signal; and an image storage means for storing an image signal in the image storage means. Frequency decomposing means for performing frequency decomposition and converting it into a band signal , and a code amount after quantization for each band signal frequency-decomposed by the frequency decomposing means has a constant value.
Quantization setting means for setting parameters in advance in so that,
A quantizing unit for quantizing a band signal frequency-decomposed by the frequency decomposing unit based on the parameter set by the quantization setting unit.

[0011]

The operation of the present invention will be described with a specific example with reference to the principle block diagram shown in FIG.

As shown in FIG. 1, an image coding apparatus according to the present invention comprises an image storage means 8 for storing a part of an input image signal 1, and a plurality of image signals stored by the image storage means 8. A first image area dividing means 2 for dividing the image area into a first image area, a frequency decomposing means 3 for decomposing the image area divided by the first image area dividing means into a predetermined two-dimensional frequency band, A second image area dividing means 4 for dividing the image into a plurality of second image areas for each frequency band decomposed by the frequency decomposing means 3, and a second image area divided by the second image area dividing means 4 And an quantizing means which is decomposed by the frequency decomposing means 3 and classified by the image area analyzing means 5 or adapted for each image area. Mel quantizer design means 6, is decomposed by the frequency decomposition unit 3, the classification by the image area analysis means 5 frequency bands or,
A quantization means for performing quantization for each image area.

In the present invention, the input image is divided into a plurality of first image areas by the first image area dividing means 2 to reduce the image size. The image storage area is reduced. Further, quantization matching with the local structure of the image can be performed.

Further, the input image is decomposed into a plurality of two-dimensional frequency bands by the frequency decomposing means 3, so that encoding can be performed in accordance with the frequency structure.

Further, the image decomposed for each of the plurality of two-dimensional frequency bands is divided into a plurality of second image regions by a second image region dividing unit 4 and classified by an image region analyzing unit 5. The quantizer is designed by the quantizer design means 6 for each class, so that the quantization conforming to the local structure of the image and the frequency structure can be performed.

Further, the image data is quantized by the quantization means 7 designed by the quantizer design means 6. The quantizing means 7 is set such that the quantized code amount takes a constant value for each predetermined first image area, and the code amount is fixed.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The features of the present invention will be specifically described below based on embodiments with reference to the drawings.

A first embodiment of the present invention will be described with reference to FIGS. 2, 3 and 4. The first embodiment is a method in which the entire input image is frequency-decomposed and quantization adapted to the frequency structure of the input image is performed.

In FIG. 2, reference numeral 21 denotes input image information;
3 is a frequency decomposition circuit for decomposing the input image information 21 into a plurality of two-dimensional frequency domains; 26 is a quantizer design circuit for designing a quantizer matched for each frequency band decomposed by the frequency decomposition circuit 23; 27 is a frequency decomposition circuit 23
A quantizer 29 performs quantization using the quantizer designed by the quantizer design circuit 26 for each frequency band decomposed by the above, and 29 is code information.

Next, the operation will be described. For example, the input image information 21 is input as raster information as shown in FIG. The memory circuit 28 stores the input image information 21 until the input image information 21 is frequency-decomposed by the frequency decomposition circuit 23. The image information stored in the memory circuit 28 is, for example, “Sub
band Coding of Images "(IE
EE Trans. on Acoustics, Sp
ech, and Signal Procedure
ng, Vol. ASSP-34, no. 5, Oct. 1
986), one-dimensional QMF (Quad
A separable 4-subband filter (Separabl) using a rattle mirror filter.
e 4-subband filter)
It is dimensionally frequency-decomposed. The image information frequency-decomposed by the separable 4-subband filter will be described with reference to FIG.

FIG. 4 is a diagram showing a two-dimensional frequency domain. As shown in FIG. 4, the image information frequency-decomposed by the separable 4-subband filter has frequency band 1
To four frequency bands 4 to 4.

Here, for example, the input image signal is represented by x (i,
j). Also, let the coefficient of the low-pass filter of the QMF be h ₀ (n) and the coefficient of the high-pass filter be h ₁ (n). For example, the signal of frequency band 1 shown in FIG.
_{Assuming that} (i, j), the signal of frequency band 2 is y ₁₀ (i, j), the signal of frequency band 3 is y ₀₁ (i, j), and the signal of frequency band 4 is y ₁₁ (i, j). , These signals

(Equation 1) Becomes Here, u and v are 0 or 1.

The image information frequency-decomposed by the frequency decomposition circuit 23 is sent to a quantizer design circuit 26, where a quantizer is designed. For example, in this embodiment, the quantizer 27 uses linear quantization. First, the quantizer design circuit 26 calculates the standard deviation σ _i of the signal of the frequency band i. At the same time, the maximum value U _i of the absolute value of the signal in the frequency band i is obtained. Assuming that the number of samples of the signal in the frequency band i is m _i and the code amount of the entire input image is B, the quantization step value S _i in the frequency band i is obtained as S _i = A · U _i / σ _i . Where A is

(Equation 2) Is a value that satisfies By fixing B to a predetermined value, the total code amount can be kept constant. Further, the quantizer 27 quantizes each frequency divided by the frequency decomposition circuit 23 using the quantizer designed by the quantizer design circuit 26 to obtain code information 29.

Next, a second embodiment of the present invention will be described with reference to FIGS. FIG. 5 is a block diagram showing the configuration of the second embodiment of the present invention. The configuration of the first embodiment shown in FIG.
A first blocking circuit 22 for dividing input image information into a plurality of first image areas is added. As will be described below, the first blocking circuit 22 reduces the capacity of the memory circuit 28 and designs a quantizer that matches local fluctuation of an image.

Next, the operation of the second embodiment will be described.

For example, the input image information 21 is input as raster information as shown in FIG. As shown in FIG. 6, when the input image is input, for example, with M lines (see FIG. 6A), the blocking circuit 22 converts the input image information into N pixels × the first image area. The image is divided into rectangles of M pixels (see FIG. 6B). Frequency decomposition circuit 23
Is the same as in the first embodiment. The image information frequency-decomposed by the frequency decomposition circuit 23 is sent to the quantizer design circuit 26, and the image in the first image area divided by the first blocking circuit 22 is compared with the first embodiment. Similarly, a quantizer is designed. At this time, by assigning the same code amount B to each first image area, fixed-length encoding can be realized for the entire input image and for each local first image area. Further, the quantizer 27 quantizes each frequency divided by the frequency decomposition circuit 23 using the quantizer designed by the quantizer design circuit 26 to obtain code information 29.

In the second embodiment, by providing the first blocking circuit 22, the quantizer 27 is matched to each frequency band and the local region of the image, and the coding efficiency is reduced. improves. The result of encoding is shown in FIG. The vertical axis in FIG. 7 represents the root mean square error between the original image and the image decoded using the code amount on the horizontal axis. It is shown that the coding efficiency of the second embodiment is improved with respect to the first embodiment.

In the second embodiment, since the image information is divided into blocks and frequency decomposition is performed for each block, it is not necessary to store the entire input image in the memory circuit 28. Therefore, compared to the first embodiment, the memory capacity can be reduced during frequency decomposition.

Next, a third embodiment of the present invention will be described. In the third embodiment, a method in which classification is introduced is adopted in the second embodiment. In the third embodiment, in addition to the processing of the second embodiment, by classifying each frequency band, quantization matching with the local property of the image is performed, and the coding efficiency is improved. .

The configuration of the third embodiment will be described with reference to FIG. The third embodiment differs from the second embodiment shown in FIG.
Blocking circuit 24 and image area analysis circuit 25
Is added.

Memory circuit 28, first blocking circuit 2
2. The operation of the frequency decomposition circuit 23 is the same as in the second embodiment. In the third embodiment, the second blocking circuit 24
Then, each frequency divided by the frequency decomposition circuit 23 is divided into a second image area, for example, an n × n area. The image area analysis circuit 25 classifies the second image areas divided by the second blocking circuit 24. For example, in the present embodiment, the classification is performed according to the magnitude of the variance of the signal in the second image area.

Now, let the variance of the signal in the second image area be s
And Assuming that the number of classes is K, K + 1 threshold values T ₀ to
T _K is provided (where T ₀ = 0 and T _K is a sufficiently large value), and if T _p−1 ≦ s <T _p , the class of the second image area is classified as p.

The image areas classified by the image area analysis circuit 25 are sent to a quantizer design circuit 26, where a quantizer is designed. For example, in the third embodiment, the first embodiment,
As in the second embodiment, the quantizer design circuit 26 uses linear quantization. In the following description, as in the second embodiment, the design of the quantizer is independently performed in the first image area. First, the quantizer design circuit 26 calculates the standard deviation σ _ij of the signal of the frequency band i and the class j. At the same time, the maximum value U _ij of the absolute value of the signal of the frequency band i and the class j is obtained. Assuming that the predetermined code amount is B, the frequency band i, and the number of samples of the signal of class j is m _ij , the quantization step value S _ij of frequency band i and class j is S _ij = A
U _ij / σ _ij . Where A is

(Equation 3) Is a value that satisfies As in the second embodiment, by assigning the same code amount B to each first image area,
Fixed-length coding can be realized for the entire input image and for each local first block. Further, the quantizer 27 quantizes each frequency divided by the frequency decomposition circuit 23 using the quantizer designed by the quantizer design circuit 26 to obtain code information 29.

In the third embodiment, since the image area analysis circuit 25 is provided, the quantizer is matched for each class, that is, since it is matched for each second image area, the coding efficiency is improved. I do. The result of encoding is shown in FIG. The vertical axis in FIG. 9 represents the root mean square error between the original image and the image decoded with the code amount on the horizontal axis. For the second embodiment,
It is shown that the coding efficiency of the third embodiment is improved.

Next, a fourth embodiment will be described.

The second embodiment described above, or the third embodiment
In the embodiment, in the first blocking circuit 22, the frequency decomposition is independently performed for each of the divided image areas.
As shown in FIG. 10, it is conceivable to divide (N + a) pixels × (M + a) pixels into blocks. This is the fourth embodiment. The circuit configuration itself is the same as that of the second embodiment shown in FIG. 5 or the third embodiment shown in FIG. 8, but the method of blocking processing in the first blocking circuit 22 is different. That is, as shown in FIG. 11A, when dividing, the halftone dot area in FIG. 10 is overlapped and divided. Here, a is the number of taps of the filter used for frequency decomposition. After blocking as described above, frequency decomposition is performed in the frequency decomposition circuit 23, and the quantizer design and quantization after the frequency decomposition are performed on a portion of N pixels × M pixels without a central mesh. (See FIG. 11B). In the fourth embodiment, information of adjacent blocks is also referred to at the time of encoding, so that block distortion due to division into the first image area can be reduced.

With this configuration, the coding efficiency is improved. FIG. 12 shows the improvement results. The vertical axis in FIG. 12 represents the root mean square error between the original image and the image decoded with the code amount on the horizontal axis. That is, since FIG. 12 is a diagram comparing the same code amount, it can be seen that the code amount is smaller when there is overlap in order to obtain the same image quality.

In the first and second embodiments described above, the first image area is an N × M rectangle, but is not limited to a rectangle.

In all of the above embodiments, in the frequency decomposition circuit 23, the input image area is decomposed into four frequency bands as shown in FIG. 4, but each frequency band is further decomposed. Is also good. For example, as shown in FIG. 13, each frequency band may be further decomposed into four parts to make 16 parts. Further, it may be further decomposed into four and decomposed into 4 ⁿ . Further, only a specific frequency band may be decomposed. For example, as shown in FIG. 14, only the lowest frequency band may be decomposed again. Further, the number of times of decomposition and the frequency band are not limited to the methods described above.

Further, in the third embodiment, the second image area is rectangular, but it is obvious that the present invention is not limited to this.

Further, in each of the above embodiments, the filter used for frequency decomposition is QMF. However, any filter generally used for subband coding can be applied, and is not limited to QMF. Furthermore, in the above embodiment, the one-dimensional filter is applied horizontally and vertically to perform two-dimensional frequency decomposition. However, the same effect can be obtained by using a two-dimensional sub-band filter.

In each of the above embodiments, simple linear quantization is used. However, the quantization method is not limited to this.
Similar effects can be obtained regardless of the quantization method.

In each of the above embodiments, the quantization step value is adaptively given to all frequency bands. However, a quantization method is determined in advance for a specific frequency band. May be. For example, since the lowest frequency band has a large effect on vision, 8-bit information may be given to the band.

In each of the above embodiments, a quantizer is designed for each input image area. However, a quantizer may be designed in advance using a typical image. In this case, it is possible to omit the image area analyzing means in FIG. 1, and there is an effect that hardware can be simplified.

In each of the above embodiments, the encoding is performed only by quantization. However, after quantization, entropy encoding is performed. Also, DPCM is performed for each frequency band.
By performing conventional encoding such as (differential pulse code modulation), DCT (discrete cosine transform), and BTC (block truncation encoding), it is possible to suppress the redundancy remaining in each frequency band. is there. In this case, the coding efficiency is further improved.

[0046]

As described above, according to the present invention, (1) it is not necessary to use entropy coding in which the code amount changes, so that the amount of calculation in the coding is reduced by a block circuit, a frequency decomposition circuit, image area analysis circuit, quantizer design circuits, and, Oite to, respectively that of the quantization circuit, can always take a constant value. Therefore, the encoding processing time can be set or predicted in advance.

(2) Since the quantizer can be designed by the quantizer design circuit so that the total code amount is constant, the code amount can be controlled to be constant.

(3) The input image is decomposed into two-dimensional frequencies,
Since quantization can be performed for each of the frequency bands, sufficient coding efficiency can be improved even by coding using only the quantization shown in the present invention.

(4) In the present invention, since the input image information is merely divided into bands by a filter and quantized by sub-band coding, the position information of the input image information is held in the code information. . Therefore, it is easy to edit the image without changing the code information.

[Brief description of the drawings]

FIG. 1 is a basic block diagram for explaining the operation of the present invention.

FIG. 2 is a block diagram showing a first embodiment of the present invention.

FIG. 3 is an explanatory diagram of input image information.

FIG. 4 is an explanatory diagram showing a decomposition form in a frequency decomposition circuit.

FIG. 5 is a block diagram showing a second embodiment of the present invention.

FIG. 6 is an explanatory diagram of the operation of the blocking circuit.

FIG. 7 is a graph showing a result of matching a quantizer in a local region.

FIG. 8 is a block diagram showing a third embodiment of the present invention.

FIG. 9 shows a result of matching a quantizer for each class.

FIG. 10 is an explanatory diagram showing blocks in a second embodiment of the present invention.

FIG. 11 is an explanatory diagram showing a state in which adjacent blocks overlap.

FIG. 12 is a graph showing the effect when overlapping and blocking are performed.

FIG. 13 is an explanatory diagram showing another decomposition mode in the frequency decomposition circuit.

FIG. 14 is an explanatory diagram showing still another decomposition mode in the frequency decomposition circuit.

FIG. 15 is a block diagram illustrating a configuration example of a conventional encoding device that performs code amount control.

FIG. 16 is a block diagram illustrating another configuration example of a conventional encoding device that performs code amount control.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Input image signal, 2, 4 ... Image area division means, 3 ... Frequency decomposition means, 5 ... Image area analysis means, 6 ... Quantizer design means, 7 ... Quantization means, 8 ... Image storage means, 21 ... Input image information, 22: Blocking circuit, 23: Frequency decomposition circuit, 24: Blocking circuit, 25: Image area analysis circuit, 26: Quantizer design circuit, 27: Quantization circuit, 28
... memory circuit, 29 ... code information, 201 ... input image information, 202 ... blocking circuit, 203 ... DCT circuit, 2
04: memory circuit, 205: quantization circuit, 206: variable length coding circuit, 207: quantization step measurement / estimation circuit, 211: input image information, 212: blocking circuit,
213 DCT circuit, 214 quantization step estimation circuit, 215 quantization circuit, 216 variable length encoding circuit

Continuation of front page (56) References JP-A-4-181887 (JP, A) W. Woods and SD O'Neil, "Subband Coding of Images", IEEE Transactions on Acoustic's, Speed and Signal Processing, 1986 Oct. , Vol. ASSP-34, no. 5, pp. 1278- 1288 ・ Hirohisa Nyozawa. Hiroshi Watanabe, "Frequency-domain realization of motion-compensated interframe prediction in equal-division subband coding," IEICE Technical Report, November 22, 1991, Vol. 91, No. 336, pp. 9-16 (IE91-82) (58) Fields surveyed (Int. Cl. ⁷ , DB name) H04N ^7/ 24-7/68 H04N 1/41-1/419 JICST file (JOIS)

Claims (5)

(57) [Claims] An image storage means for storing an image signal and an image stored in the image storage means are stored in an adjacent image storage means.
In a manner in which the image areas overlap, the
Image blocking means for dividing, the field of the divided image areas by the image blocking means
Frequency decomposition means for performing frequency decomposition on the image and converting it to a band signal ;
And the divided image area excluding the overlap
Parameters that can be quantized to a certain amount based on
A quantization setting unit for setting a meter; and a band signal frequency-decomposed by the frequency decomposition unit based on the parameters set by the quantization setting unit, wherein the divided image excluding the overlap is removed.
A subband image encoding apparatus, comprising: a quantization unit that performs quantization on an image corresponding to an image area . 2. An image storage means for storing an image signal; The image stored in the image storage unit is replaced with an adjacent image.
In a manner in which the image areas overlap, the
Image blocking means for dividing; The image of the image area divided by the image blocking means
Frequency to perform frequency decomposition on image and convert to band signal
Disassembly means; By the frequency decomposition means based on predetermined parameters
A frequency-resolved band signal that overlaps
For those corresponding to the above divided image area
And a quantization means for performing quantization.
Subband image encoding device. 3. An image storage means for storing an image signal, and an image stored in the image storage means is stored in an adjacent image storage means.
In a manner in which the image areas overlap, the
Image blocking means for dividing, the field of the divided image areas by the image blocking means
Frequency decomposition means for performing frequency decomposition on the image and converting it to a band signal ;
And the divided image area excluding the overlap
Band signal blocking means for dividing a signal corresponding to a band into a plurality of areas; band signal analyzing means for classifying band signals divided by the band signal blocking means; and the band signal analyzing means On the basis of the band signals classified according to the following, quantization setting means for setting a parameter to be quantized by a fixed amount for each region divided by the band signal blocking means , and a parameter set by the quantization setting means And a quantizing means for performing quantization on the band signal divided by the band signal blocking means based on the band signal. 4. An image storage means for storing an image signal, The image stored in the image storage unit is replaced with an adjacent image.
In a manner in which the image areas overlap, the
Image blocking means for dividing; The image of the image area divided by the image blocking means
Frequency to perform frequency decomposition on image and convert to band signal
Disassembly means; With the band signal frequency-decomposed by the frequency decomposition means
And the divided image area excluding the overlap
Band signals that are divided into multiple areas for those corresponding to the area
No. blocking means, Band signal divided by the band signal blocking means
Band signal analysis means for classifying against The band signal blocking means based on a predetermined parameter;
Quantizer that performs quantization on band signals divided by stages
Subband image code, comprising:
Encryption device. 5. An image storing step of storing an image signal.
When, The image stored in the image storing step is replaced with an adjacent image.
In a manner in which the image areas overlap, the
An image blocking step for dividing; The image area divided by the image blocking step
Frequency resolution for the image of
A wave number decomposition step; The frequency decomposition step based on a predetermined parameter;
This is a frequency-resolved band signal that
Corresponding to the above divided image area excluding the For things
And a quantization step for performing quantization.
Subband image encoding method.