JPH06268989A - Sub-band encoding device - Google Patents

Abstract

PURPOSE:To perform efficient encoding and to reduce the scale of hardware. CONSTITUTION:In the sub-band encoding system of pictures, it is normal to make the ratio of a sub-sampling cycle and the sampling cycle of input data smaller than the tap number of a band dividing filter so as to make quantization distortion inconspicuous. As a result, the absolute values of respective band components become large at the edges of a screen and efficient data compression is interfered. Thus, endless scanning for passing through entire picture elements inside a frame or a field only once is performed, the edges of the screen are equivalently eliminated and sub-band encoding is performed for data strings obtained as the result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a subband coding apparatus suitable for coding image data for image transmission and recording.

[0002]

2. Description of the Related Art An example of the configuration of a conventional one-dimensional information subband encoding system is shown in FIG.
This configuration is a basic element, low-pass filter (LPF)
The input data string is divided into many band components by repeatedly using a pair of a high pass filter and a high pass filter (HPF).

As shown in FIG. 1, the LPF output is repeatedly divided, and the HPF output is transmitted as it is. On the receiving side, the band component is synthesized by the synthesis filter bank shown on the right side of FIG. 1 to restore the data.

In subband coding of two-dimensional information such as an image, band division is performed independently in the horizontal and vertical directions of the image. When the division is performed in multiple stages in each direction, it is necessary to divide the horizontal direction and the vertical direction alternately one by one. That is, after the division in the horizontal direction, the division in the vertical direction is performed, the division in the horizontal direction is performed in the low region in the vertical direction, and this division is repeated. Therefore, a configuration example of the sub-band division filter bank of two-dimensional information is as shown in FIG.

[0005]

The low-pass filter (LPF) and the high-pass filter (HPF), which are the basic elements of the configuration shown in FIG. 2, have the number of taps (the number of delay elements) when configuring a digital filter. ) Is two or more digital filters. Here, the tap means an outlet from each joint to which a large number of 1-sample delay elements are connected.

Since the output of each filter is subjected to 2-to-1 sub-sampling as shown in FIG. 2, when the number of taps of the filter is 2, it is related to the value of each sample which is continuous in time in the sub-sampler output. The input data do not overlap each other. Sub-sampling means sampling at a sampling frequency equal to or lower than the Nyquist frequency, and is also called sub-Nyquist sampling.

Here, in order to perform data compression, when each subsampler output is quantized, that is, output values are rounded and transmitted and then decoded on the receiving side, distortion due to quantization appears in the decoded output as it is. . This distortion is visually noticeable when projected on a TV monitor or the like.

In order to make this distortion inconspicuous, the number of taps of the digital filter may be increased to overlap the input data relating to the subsampler output value. When performing 2-to-1 sub-sampling, the number of taps of the filter needs to be 4 or more.

However, another problem arises here.
For example, it is assumed that the number of filter taps (length of coefficient string) is 4 and the filter is configured in multiple stages. The length of the coefficient sequence when the first stage and the second stage are considered as one filter is 10. Further, the length of the coefficient string when the first through third stages are regarded as one filter is 22. in this way,
The length of the coefficient string increases sharply as the number of stages increases.

More specifically, since the number of taps of the filter is 4, the coefficients are A ₀ , B ₀ , C ₀ and D _0.
Then, the transfer function of the first stage filter is

[0011]

[Formula 1] A ₀ Z ⁰ + B ₀ Z ^-1 + C ₀ Z ^-2 + D ₀ Z ^-3 (b) The second-stage filter handles data that has been subjected to 2-to-1 sub-sampling, so A ₁ , B ₁ , C _1, and D ₁ are used as coefficients.

[0012]

[Equation 2] A ₁ Z ⁰ + B ₁ Z ^-2 + C ₁ Z ^-4 + D ₁ Z ^-6 (b). Therefore, when the first and second stages are combined into one filter, (i) and (b) are multiplied to obtain the following.

[0013]

[Equation 3] [A ₀ Z ⁰ + B ₀ Z ^-1 + C ₀ Z ^-2 + D ₀ Z ^-3 ] [A ₁ Z ⁰ + B ₁ Z ^-2 + C ₁ Z ^-4 + D ₁ Z ^-6 ] = A ₀ A It becomes ₁ Z ⁰ ＋ ・ ・ ・ ＋ D ₀ D ₁ Z ^-9 (C). Furthermore, the filters of the third stage are A ₂ , B ₂ , C
₂ and D ₂ as coefficients,

[0014]

[Formula 4] A ₂ Z ⁰ + B ₂ Z ^-4 + C ₂ Z ^-8 + D ₂ Z ^-12 (d). If you consider the 1st stage to the 3rd stage as one filter, multiply (c) and (d)

[0015]

[Formula 5] [A ₀ A ₁ Z ⁰ ＋ ・ ・ ・ ＋ D ₀ D ₁ Z ^-9 ] [A ₂ Z ⁰ ＋ B ₂ Z ^-4 ＋ C ₂ Z ^-8 ＋ D ₂ Z ^-12 ] ＝ A ₀ A ₁ A ₂ Z ⁰ + ... + D ₀ D ₁ D ₂ Z ^-21 Therefore, the length of the coefficient sequence is 4, 10, 22 ,.
It increases in the order of. When the coefficient sequences overlap each other, a part of the coefficient sequences will extend off the screen at the edge of the screen. Considering that the input data is zero outside the screen, a large level change occurs at the screen edge. The fact that there is a stepwise (rapid) change in the level means that all frequency components exist from the low frequency band near the DC component to the upper limit of the band. There is a problem that good data compression cannot be performed.

[0016]

In order to solve the above-mentioned problems, the present invention equivalently eliminates the screen edge by performing endless scanning which passes through all the pixels in one frame or one field only once. It is characterized in that sub-band division coding is performed on the obtained data string.

In the data string obtained by such endless scanning, since the stepwise level change does not occur, the above problem does not occur. Further, according to the present invention, since the two-dimensional information is made one-dimensional by the endless scanning,
The filter bank for one-dimensional information shown in FIG. 1 or FIG. 3 (only the dividing unit) can be used.

Next, a method of constructing an endless scanning pattern which passes through all the pixels in the screen only once will be described. "one"
2 open circuit patterns based on the U-shape or the U-shape
Connect them face to face to form an endless pattern. The open circuit pattern is divided into a basic circuit open pattern and a composite circuit open pattern.

An example of the basic open circuit pattern is shown in FIG. Vertical,
The lower side of the rectangle, in which pixels are arranged side by side, is an open circuit. The composite open circuit pattern is a combination of a plurality of basic open circuit patterns, some examples of which are shown in FIG. Similarly, the lower side of the rectangle is open.

By replacing each point on the endless pattern thus completed with an open circuit pattern, a larger endless pattern can be constructed. The replacement with the open circuit pattern can be repeated any number of times, and the endless pattern can be enlarged without limit.

As can be seen from the examples of the patterns shown in FIGS. 4 and 5, it is not possible to form an open circuit pattern having an odd number of pixels in the horizontal direction and an even number of pixels in the vertical direction. Therefore, it is not possible to construct an endless pattern having an odd number of pixels in both the horizontal and vertical directions. However, in the moving image and still image systems that have been practically used so far, there is no example in which the number of pixels is odd in both horizontal and vertical, and it is considered that there will be no future. It can be said that there is no problem in practical use.

Next, the component television system (CCIR) of 525 scanning lines and 60 fields per second is used.
A specific configuration method will be described with reference to Rep 601).
In this system, the number of effective pixels in one frame is 720 horizontally and 480 vertically. We decompose this number into the following product form.

[0023]

[Formula 6] 720 × 480 = 3 × 2 × 15 ² × 16 ² The final shape of the product should be (2 ^P ) ² as much as possible, and the value of P should be large. First, an endless pattern having a size of 3 × 2 is first constructed by a combination of a 2 × 2 open circuit pattern and a 2 × 1 open circuit pattern, as shown in FIG. Next, set each point of this endless pattern to 15 ²
Replace with the open circuit pattern.

The open pattern of 15 ² is shown in FIG. Each point of the 3 × 2 × 15 ² sized endless pattern is further replaced by 16 ² open circuit pattern. The 16 ² open circuit pattern is obtained by combining four 4 × 4 open circuit patterns shown in FIG. As a result, an endless pattern having a size of 3 × 2 × 15 ² × 16 ² is obtained.

[0025]

EXAMPLE One example of the present invention will be described with reference to FIG.

A video signal is applied to the input terminal 1. A / D
The converter 2 converts the input into digital data. The frame memory or field memory 3 converts the raster scan into endless scan. The memory read address for converting to endless scanning is created in advance by the address generator 4 so as to have the endless pattern as shown in the previous section, and the pattern is repeated for each frame or field.

Next, the division filter bank 5 divides into a large number of subband components. The division filter bank 5 is
For example, it has a left half portion of FIG. 1 or a structure as shown in FIG. Although FIG. 1 and FIG. 3 show the case where the number of division filter stages is two, the number of stages is arbitrary.

Reference numeral 6 is a quantizer. The quantizer 6 is a circuit for rounding the filter output, that is, processing for reducing the precision (number of digits) of the data to a necessary minimum. Needless to say, this quantization is performed to compress the data. Quantization is performed for each divided filter so that the accuracy can be set for each subband.

The quantized filter outputs of the respective stages are integrated by the multiplexer 7 and arranged in time series. As a method of arranging data, first, data in the latter stage is arranged first, in one stage is arranged in time order, and some outputs obtained simultaneously in one stage are arranged in ascending order of bandwidth. Sometimes. The reason for arranging the data in the latter stage first is that the total amount of data is smaller than that in the former stage, and the power is large and the importance is high.

The output data of the multiplexer 7 is input to the entropy encoder 8 and is entropy encoded. Entropy coding is coding in which the length of an output code is changed according to the occurrence probability of an input data value. That is, the data amount is statistically reduced by assigning a short code to data with a high occurrence probability and a long code to data with a low occurrence probability. The probability of occurrence of each data value is measured, and a conversion table is created based on that.

Reference numeral 9 is a check data generating and adding circuit for performing error correction on the receiving side. An error correction method suitable for each is adopted depending on whether the error occurring in the transmission path is mainly a random error or a burst error.

Reference numeral 10 is a digital modulation circuit. The normal transmission line including the recording system is at both ends of the transmission frequency band,
That is, the transmission characteristics in the low frequency region and high frequency region close to the DC component are not good. Therefore, a circuit for converting a code into a form in which there are few frequency domain components with poor characteristics is a digital modulation circuit.

Reference numeral 11 is a transmission line or recording system, for example, a VTR. The code data transmitted or recorded / reproduced by the transmission line or VTR 11 is given to the digital demodulation circuit 12. The digital demodulation circuit 12 is a circuit that operates in the opposite manner to the digital modulation circuit 10.
An error correction circuit 13 detects and corrects an error caused by passing through the transmission line or the VTR 11.

Reference numeral 14 denotes an entropy decoder, which is a circuit that operates in the opposite manner to the entropy encoder 8. The demultiplexer 15 is a circuit that operates in reverse to the multiplexer 7, and is a circuit that parallelizes serial data into data for each subband.

Reference numeral 16 denotes an inverse quantizer, which is a circuit for restoring the digit number of the data reduced by the quantizer 6.

Reference numeral 17 denotes a synthesis filter bank, which has a shape like the right half of FIG. 1, for example.

A scanning conversion memory 18 converts endless scanning into raster scanning.

Reference numeral 19 is an address generator for the scan conversion memory 18.

Reference numeral 20 denotes a D / A converter, which has an output terminal 21.
Get the decoded output.

Summary An example of a two-dimensional information sub-band division filter bank according to the prior art is shown in FIG. On the other hand, an example of subband division in which the number of divisions is the same as that of the example of FIG. 2 in one embodiment of the present invention is as shown in FIG.

To implement the method of the present invention, a raster scan to endless scan conversion memory is required at the input of the arrangement of FIG. Comparing the configuration of FIG. 3 with the addition of the scan conversion memory and the configuration of FIG. 2, the number of filters is smaller by four in the embodiment of the method of the present invention. Also,
The conventional method requires five scan conversion memories, whereas one embodiment according to the method of the present invention requires only one scan conversion memory. Obviously the hardware scale will be smaller according to one embodiment of the invention.

It has been described above that the image data sequence obtained as a result of scanning by the endless pattern is subband-encoded, but the image data sequence obtained here has a large correlation between consecutive samples. Since it has, efficient coding can be performed even when predictive coding such as DPCM is performed.

[0043]

As described above, according to the present invention, not only efficient encoding can be performed, but also the scale of hardware can be reduced.

[Brief description of drawings]

FIG. 1 is a diagram illustrating a configuration example of a subband encoding system for one-dimensional information.

FIG. 2 is a diagram showing a configuration example of a subband division filter bank of two-dimensional information.

FIG. 3 is a diagram showing a configuration example of a subband division filter bank of one-dimensional information.

FIG. 4 is a diagram showing an example of a basic opening pattern.

FIG. 5 is a diagram showing an example of a composite open circuit pattern.

FIG. 6 is a specific structure of an endless pattern (720 × 4).
It is a figure showing (in the case of 80).

FIG. 7 is a block diagram showing an embodiment of the present invention.

[Explanation of symbols]

 1 Input Terminal 2 A / D Converter 3 Scan Conversion Memory 4 Address Generator 5 Divided Filter Bank 6 Quantizer 7 Multiplexer 8 Entropy Encoder 9 Check Data Generation and Addition Circuit 10 Digital Modulation Circuit 11 Transmission Line or Recording System (for example, VTR) 12 Digital demodulation circuit 13 Error correction circuit 14 Entropy decoder 15 Demultiplexer 16 Inverse quantizer 17 Synthesis filter bank 18 Scan conversion memory 19 Address generator 20 D / A converter 21 Output terminal 50 2: 1 subsampler 60 adder 70 scan conversion memory 80 4 to 1 subsampler HPF high pass filter LPF low pass filter

Claims (1)

[Claims] 1. A rectangular or square pixel that passes through all the pixels forming a rectangle or square whose length is an integral multiple of the sampling interval of a sampled image signal (hereinafter, referred to as pixel) once Any one of an endless pattern scan created by facing two open circuit patterns configured by using an open circuit having one side as an open circuit, and an endless pattern scan in which each pixel forming the endless pattern is replaced by the open circuit pattern. A subband coding apparatus comprising: a unit configured to scan in one scanning mode; and a unit configured to convert an image data string obtained as a result of the scanning into band component data by a band division filter.