JPH08265769A - Method and apparatus for adaptive quantization of image data - Google Patents

Abstract

PURPOSE: To improve picture quality by setting a quantization scale corresponding each component of flatness, texture and edge of picture. CONSTITUTION: A quantization part 16 quantizes a DCT coefficient and outputs quantization data. A quantization scale deciding part 14 generates a quantizing scale corresponding to area characteristics of an input picture and gives them to the quantization part 16. At this point, the area characteristics are defined and are divided into each characteristic of 'edge', in which each pixel value radically changes and has an irregular configuration, 'texture', in which a pixel value racially changes and has a regular configuration, and 'flatness', in which there is a slight change in a pixel value. The quantization scale deciding part 14 calculates from a maximum/minimum value of pixel value the first and the second speeds corresponding to the area characteristics with a specified method. The first speed changes from slow to fast in the order of 'edge', 'texture' and 'flatness', while the second speed alters to fast in a group of 'edge' and 'texture' and moderate in 'flatness'. The quantization scale is fixed to a value in proportion to a product of the first and the second speeds.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an adaptive quantization method for image data and an apparatus thereof, and more particularly to an adaptive quantization method and an apparatus used for a real-time encoder system for encoding a moving image.

[0002]

2. Description of the Related Art In a conventional MPEG encoder / decoder, in order to improve image quality or reduce cost, (1) motion estimation, (2) pre-processing / post-processing,
(3) There are some points to be improved in speed control and the like. The present inventors have already disclosed some novel inventions for the first two by patent applications, while the last item, the speed control in the encoding of image data, is still We are developing. Such speed control is one of the most important video encoding technologies.
Thought to be one.

During the MPEG-2 standardization project, a Test Model (TM) encoder was developed to provide a common platform for comparative experiments. The speed control algorithm forming a part thereof is called TM speed control and is based on the following three main steps.

Step 1: a step of allocating bits per image frame in order to estimate the number of bits available for coding per image frame. step2: To set the reference value of the quantization scale for each macroblock by the "virtual buffer",
The step of controlling the buffer in macro block units. Step 3: A step of performing adaptive quantization in macroblock units in order to adjust the quantization parameter due to spatial activity (change or complexity) in the macroblock.

Here, the adaptive quantization algorithm of step 3 will be described. First, the spatial activity (act) of the macroblock is calculated from the four luminance fields configured in the 8 × 8 sub-block using the original pixel value and the four luminance frames configured in the 8 × 8 sub-block. Calculate the measurement.

## EQU1 ## act = 1 + min (var_sblk) (1) where:

[Equation 2]

P _k is all pixel values (k = 1, 2, 3 ... 64) of the original 8 × 8 block. And N
The standardized activity indicated by _act is obtained as follows:

(Equation 3) Here, avg_act is the average value of the activities of the images immediately before the image to be encoded. In the first image,
avg_act = 400. Finally, we have the following formula:

## EQU00004 ## Q_step3 = Q_step2.times.N_act (4) Q_step2 is the step2 of the above TM algorithm.
This is the reference quantization parameter obtained in.

[0007]

In the prior art, the final quantization scale is Q_
It was obtained by multiplying step2 (reference quantization parameter obtained in step2) by N_act,
It has been pointed out that the TM step3 requires a large amount of calculation in order to calculate the variance based on the activity.

Further, as shown in equation (3), N_act
Is for adjusting the quantization depending on the activity in the macroblock, that is, the level of activity (generally, the flat component in the image area has low activity, while the edge component and the texture component have high activity). There is a problem that it is not possible to determine the optimum quantization scale by distinguishing the edge component and the texture component of the image.

SUMMARY OF THE INVENTION It is an object of the present invention to solve the above-mentioned conventional technique and to provide an adaptive quantization method of image data and an apparatus thereof capable of generating higher quality images at the same encoded data rate. To do. Another object of the present invention is to provide a quantization method with improved coding efficiency.

[0010]

In order to achieve the above object, an adaptive quantization method according to the present invention provides an area characteristic (component) within an image area composed of m × n size image data. Is obtained, and the image data is quantized by the quantization scale.

[0011]

In the adaptive quantization method according to the present invention, when calculating the quantization scale, (1) the texture and edge regions are classified and (2) the high contrast region and the low contrast region are classified according to the region characteristics. The adaptive algorithm enables high quality image generation.

[0012]

Embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing a configuration example of a coding apparatus to which the adaptive quantization method for image data of the present invention is applied. In the figure, 10 is an input image composed of 704 × 480 pixels, 12 is a DCT unit 12 for performing DCT (discrete cosine transform) processing on the input image, and 14 is a quantization scale for obtaining a quantization scale according to a region characteristic of the input image The deciding unit 16 quantizes the DCT count according to the instruction from the quantizing scale deciding unit 14, and quantizes the quantized data (only when the quantizing scale is different from the value transmitted immediately before,
Is newly transmitted).

Next, details of the quantization scale determination unit will be described. First, we define the adaptive quantisation factor denoted by N_rate.

## EQU00005 ## Q_step3 = Q_step2.times.N_rate (5) Q_step2 is the reference quantization parameter of the TM described in the conventional art. That is, the difference between the expressions (4) and (5) is N_rate and N_act, and the feature of the present embodiment lies in the way of determining the N_rate. Therefore, in the present embodiment, step1 and step2 of the above TM
Will be described in common, and a method of obtaining improved step3 will be described.

Since the conventional TM step 3 is an activity measure based on fluctuation, it is difficult to distinguish the texture from the edge region. In order to perform high image quality encoding with an optimum quantization scale, it is desirable that two areas can be identified. At the same time, since the transmission capacity of encoded data is limited, it is necessary to optimize the allocation of the encoded bit amount according to the image area characteristics.

The quantization method of this embodiment uses a region identifier, and the region identifier determines a quantization scale depending on the edge component, the texture component and the flat component in the macroblock region.

Here, the "edge" is a partial area where the pixel value changes abruptly, and the shape is irregular, and the "texture" is a partial area where the pixel value changes abruptly. Of which the regularity is recognized, and "flat" means a partial region in which the change of the pixel value is small.

In this embodiment, the area identifier has two types of descriptors, an activity descriptor and a contrast descriptor. These descriptors are obtained as follows.

First, the dynamic range of the pixel value of each original block is calculated, and the maximum value (m
ax_range) and the minimum value (min_range)
Find e). There are two alternative definitions of sub-block shape. 2 (a) and 2 (b) respectively,
There are 4 × 4 blocks and 16 × 4 blocks. When giving priority to accuracy, it is desirable to use the former sub-block, but on the other hand, the cost is high. On the other hand, if the latter sub-block is used, the cost will be lower, but the accuracy will be reduced. When 4 × 4 subblocks are used, 16 subblock operations are required for one macroblock (16 × 16), and in the case of 16 × 4 subblocks, 4 P-size subblocks are required for a macroblock. , Q size sub-blocks 4 times, a total of 8 calculations are required. The points in the sub-blocks are the pixels to be calculated.

In both cases, the difference between the fields is assumed to be small, and the pixel to be calculated exists only in the first field (the odd field in the example of FIG. 2). There are some adaptation methods using the above-mentioned two descriptors, but in this embodiment, the following concept is adopted from them. (1) The edge region requires a finer quantization scale than the texture or the flat region. (2) The low contrast region requires a finer quantization scale than the high contrast region. Thus, only the maximum range and the minimum range obtained by the above calculation are used to determine the activity descriptor and the contrast descriptor, and the velocity 1 and the velocity 2 are derived.

Finally, N_rate is defined as in equation (6).

(Equation 6)

Table 1 shows how speed 1 and speed 2 change as a general tendency depending on the characteristics in the image area. As can be seen from the table, the velocity 1 determined by the activity descriptor tends to become smaller in the order of the flat component, the texture component, and the edge component in the macroblock region, while the velocity determined by the contrast descriptor. 2 is small in the flat component and “medium to large” in the texture component and the edge component, respectively.
Is. This means that even if the present embodiment is a little rougher in the quantization scale of the texture area than the flat area and the edge area, that is, the image quality is deteriorated, the reproduced image is in the human visual sense, This is because it is assumed that there are not many problems. Therefore, the value of velocity 1 is smaller in the edge component than in the texture component.

[0022]

[Table 1]

Thus, the N_rate of the equation (5) for determining the quantization scale is obtained by the definition of the equation (6), and the small value of the velocity 1 means that the quantization scale is small (fine). Which, on the other hand, means that the bit budget for coding is large.
In other words, the quantizer scale is proportional to the product of rate 1 and rate 2 and inversely proportional to the bit budget.

As described above, the coding is performed according to the edge area and the flat area in the image area where high-precision coding is required, and the texture area in which the deterioration of the image quality accuracy is relatively tolerable. By adaptively adjusting the speed, that is, the allocation of the encoding bit amount, it is possible to reproduce a higher quality image with the same encoded data.

In the present embodiment, the quantization scale is adjusted by the above (6), but the present invention is not limited to this, and the adaptation method based on the combination of two area identifiers (speed 1 and speed 2) is appropriately changed. Note that this can improve the image quality. Therefore, it is possible to investigate the human visual information processing process and find an optimal adaptation method in place of (6) above.

Next, a second embodiment of the present invention will be described. When the quantization scale is variously changed according to the region characteristic of the macroblock as in the above embodiment, it is necessary to transmit the quantization scale coded together with the quantized data to the decoder side. However, it has been found that changing the quantization scale too often adversely affects the coding efficiency, especially when the data transmission rate is low. That is, in practice, encoding the quantization scale requires 5 bits, and this value is approximately D.
This corresponds to 0.5-1 CT coefficient. Therefore, when the transmission speed is low, the coding efficiency is better if the DCT count is preferentially transmitted rather than the quantization scale having a relatively small difference.

Therefore, in this embodiment, when the quantization scale obtained by the equation (5) is considerably different from the one transmitted immediately before, the quantization scale is encoded and transmitted. The conditions for encoding the quantization scale are shown below.

## EQU00007 ## R = Q_step3 / quantization scale transmitted immediately before (if R <thr1 or thr3 <R, the quantization scale is encoded and transmitted)

If Q_step3 is a ratio with the quantization scale transmitted immediately before and does not exceed the range between the threshold values thr1 and thr2, the quantization scale is encoded and not transmitted. The bits saved by this are DC
It is expected that the T coefficients will be effectively allocated and the coding efficiency will be improved.

The thresholds thr1 and thr2 may be variable according to the bit transmission rate. In other words, when the bit transmission rate is high, even if the interval between the threshold values thr1 and thr2 is made small or equal, the transmission capacity is large, so that the transmission of the quantization scale does not pose a problem. It is also possible to change these thresholds according to the region characteristics. That is, since the quantization scale is fine (the bit allocation amount is large) in the encoding of the macroblock having the edge region or the flat region described above,
In this case, the intervals between the threshold values thr1 and thr2 are made smaller or equal, while the threshold values thr1 and thr2 are set in the macroblock having the texture area.
It is also possible to make the interval between and relatively large.

Table 2 shows the amount of calculation required to obtain Q_step3 per macroblock in this embodiment,
The calculation amount by step3 of the conventional TM is shown. It can be seen that the method according to the present embodiment is about 10-20% of the calculation amount of the TM step3 algorithm.

[0031]

[Table 2]

Simulations were performed using several test sequences in comparison with TM step3. Simulation conditions and coding parameters are as follows: 704 × 480 pixels as input image, 4.0 Mbit / sec, N = 15, M = 3, frame structure picture,
An adaptive frame / field prediction mode and an adaptive frame / field DCT. To get an accurate comparison, TM speed control steps 1, 2 were used. The threshold values thr1 and thr2 are 0.75 and 1.
The sub-block structure was set to 25 and a 4 × 4 square block was selected. Table 3 shows the average SNR of 60 frames.
Shown in

[0033]

[Table 3] Table 3: SNR results

Further, from the test image used for the simulation, noise in the edge region is appropriately suppressed, and in comparison with the conventional TMstep3, this embodiment shows that the Y-SNR is
Under the conditions described above, it is possible to provide a high image quality having a 0.11-0.85 dB gain.

As described above, the present embodiment enables high quality image reproduction with a small amount of calculation.
It can be applied to adaptive quantization for MPEG-2 rate control. 10-20 by simulation
While providing a calculation amount of%, it is possible to generate an image quality as good as or better than that of the MPEG-2TM speed control step 3.

[0036]

According to the present invention, the image data can be encoded by using the quantization scale depending on the flat component, the edge component and the texture component to realize the quantization in accordance with the region characteristic. , It is possible to generate higher quality images at the same encoded data rate.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an encoding device to which an adaptive quantization method according to an embodiment of the present invention is applied.

FIG. 2 is a diagram showing a configuration of sub-blocks.

[Explanation of symbols]

 10 Input Image 12 DCT Section 14 Quantization Scale Determining Section 16 Quantization Section

Claims (7)

[Claims] 1. For an image area composed of image data of m × n size, a quantization scale depending on edge area characteristics and texture area characteristics in the image area is obtained, and the image data is quantized by the quantization scale. To do
Adaptive quantization method for image data. 2. A quantization scale using an identifier that changes according to the characteristics of the image area is obtained for an image area consisting of m × n size image data, and the image data is quantized by the quantization scale. , Adaptive quantization method for image data. 3. The claim according to claim 2, wherein the identifier is
An adaptive quantization method for image data, comprising first and second parameters that change according to the region characteristic, and obtaining a quantization scale by a combination of the first and second parameters. 4. The third parameter according to claim 3, wherein the first parameter is changed so as to make the quantization scale of the edge region finer than that of the texture region, and the second parameter is changed from the texture region and the edge region. Is an adaptive quantization method for image data, which changes so that the quantization scale of the flat region is finer. 5. The m × n size image data is quantized by a predetermined quantization scale, the predetermined quantization scale is compared with the quantization scale of the image data transmitted immediately before, and the comparison value is obtained. Is outside a certain range, an adaptive quantization method for image data, which encodes and transmits the predetermined quantization scale. 6. The fixed range according to claim 5,
An adaptive quantization method for image data, which is variable according to the data transmission rate. 7. A means for obtaining a quantization scale depending on edge area characteristics and texture area characteristics in an image area composed of m × n size image data, and the means are connected to the means and are obtained. An image data adaptive quantizing device, comprising: means for quantizing the image data according to a quantization scale.