JP3622226B2 - Encoder - Google Patents

Description

[0001]
[Industrial application fields]
The present invention relates to an encoding apparatus that encodes a moving image, for example.
[0002]
[Prior art]
Coding apparatuses that perform variable-length coding after a DCT transform and quantization of a moving image and output it are about to be put into practical use. In this encoding apparatus, a motion vector is obtained from the error of the current image data subjected to inverse quantization and inverse DCT transform and the previous image data. Based on the motion vector, motion-compensated predicted image data is generated, and difference data is obtained from the predicted image data and the next image data. Then, DCT transformation and quantization are performed on the difference data.
[0003]
FIG. 5 is a block diagram illustrating an example of an encoding apparatus. The current image data via the input terminal 51 is supplied to a subtracter 52 and a motion detector 63 described later. In the subtracter 52, output data of a motion compensator 62 described later is subtracted from the current image data. The output data of the subtractor 52 is DCT converted by, for example, a two-dimensional DCT converter 53 of (8 × 8) pixels. The image data after DCT conversion is quantized by a quantizer 54 and supplied to a variable length encoder 55 and an inverse quantizer 58. In the variable length encoder 55, the code length is changed according to the appearance frequency of the data. The variable length code data is output to the subsequent circuit via the output terminal 56.
[0004]
The output of the variable length encoder 55 is supplied to a constant rate controller 57. In the constant rate controller 57, a quantization scale for the quantizer 54 is set. The output of the constant rate controller 57 is supplied to the quantizer 54 and the inverse quantizer 58. The image data supplied to the inverse quantizer 58 is subjected to a process opposite to that performed by the quantizer 54 and is subjected to inverse DCT conversion by the inverse DCT converter 59. This data is supplied to the adder 60. The adder 60 is supplied with the predicted image data of the reference frame from the motion compensator 62, and these data are added. The output data of the adder 60 is supplied to the frame memory 61. The image data of the reference frame output from the frame memory 61 is supplied to the motion compensator 62 and the motion detector 63.
[0005]
The motion detector 63 obtains a motion vector of the image based on the current image data input via the input terminal 51 and the image data of the reference frame supplied from the frame memory 61. The detection output of the motion detector 63 is supplied to the motion compensator 62. The motion compensator 62 performs motion compensation prediction based on the detection output of the motion detector 63 and the image data of the reference frame supplied from the frame memory 61. The predicted image data of the motion compensator 62 is supplied to the subtracter 52 and the adder 60.
[0006]
[Problems to be solved by the invention]
As described above, the DCT-transformed image data is quantized. In the test model of the encoding apparatus, the weighting factor is quantized with a default value in the quantizer. For example, in the case of quantizing image data that has a strong motion and is difficult to follow visually, using the default value, the image data is quantized with some high-frequency components included. For this reason, visually useless codes are generated. As a result, the quantization scale increases, causing image quality degradation.
[0007]
On the other hand, in the case of image data with small motion, visual characteristics are good. If the image data has many high frequency components and is quantized using a default value, the high frequency components are reduced and quantized. For this reason, the image is output as a blurred outline. On the other hand, when the high-frequency component of the image data is small and quantized using the default value, mosquito noise or the like is conspicuous and causes deterioration in image quality.
[0008]
Accordingly, an object of the present invention is to provide an encoding apparatus that can change the weighting coefficient at the time of quantization corresponding to the degree of motion and high frequency of image data, and can prevent image quality deterioration.
[0009]
[Means for Solving the Problems]
The present invention relates to an encoding device for encoding image data.
DCT conversion means for DCT converting image data;
Quantization means for quantizing the image data converted by the DCT conversion means;
A degree-of-motion detecting means for detecting the degree of movement based on the magnitude of the motion vector of the image data;
High frequency detection means for detecting the high frequency of the image data based on the magnitude of the high frequency component of the DCT coefficient supplied from the DCT conversion means;
Based on the degree of motion determined by the degree of motion detection means and the high frequency detected by the high frequency detection means, the control means for variably controlling the quantization weight coefficient of the quantization means,
The control means is
If the degree of motion detected by the degree-of-motion detection means is greater than the first threshold, the quantization weight coefficient of the quantization means is controlled so as to reduce the high frequency side of the DCT coefficient,
When the degree of motion detected by the degree-of-motion detector is smaller than the first threshold and the high frequency detected by the high-frequency detector is smaller than the second threshold, the high frequency side of the DCT coefficient is reduced. Control the quantization weighting coefficient of the quantization means,
When the degree of motion detected by the degree-of-motion detecting unit is smaller than the first threshold and the high level detected by the high-frequency detecting unit is larger than the second threshold, the high frequency side of the DCT coefficient is increased. Thus, the encoding device is characterized by controlling the quantization weight coefficient of the quantization means.
[0010]
[Action]
The motion level detection block 23 detects the motion level of the image data. When this degree of motion is equal to or greater than a predetermined value, a large weight coefficient is supplied from the weight coefficient determination block 25 to the quantization weight coefficient block 21. If the degree of motion is less than or equal to a predetermined value, the high frequency detection block 24 detects the high frequency. A small weight coefficient is supplied to the quantization weight coefficient block 21 when the high frequency is equal to or greater than a predetermined value, and a large weight coefficient is supplied to the quantization weight coefficient block 21 when the high frequency is equal to or less than the predetermined value.
[0011]
【Example】
Embodiments of an encoding apparatus according to the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an encoding apparatus according to the present invention. The current image data via the input terminal 1 is supplied to the subtracter 2 and a motion detector 13 described later. In the subtracter 2, output data of a motion compensator 12 described later is subtracted from the current image data. The output data of the subtracter 2 is DCT converted by the two-dimensional DCT converter 3 of (8 × 8) pixels. The image data after DCT conversion is quantized by the quantizer 4 and supplied to the variable length encoder 5 and the inverse quantizer 8. The DCT coefficient generated by the DCT converter 3 is supplied to a quantization weight coefficient controller 14 described later. In the variable length encoder 5, the code length is changed according to the appearance frequency of data. This variable length code data is output to the subsequent circuit via the output terminal 6.
[0012]
The output of the variable length encoder 5 is supplied to the constant rate controller 7. In the constant rate controller 7, the quantization scale for the quantizer 4 is set. The output of the constant rate controller 7 is supplied to the quantizer 4 and the inverse quantizer 8. The image data supplied to the inverse quantizer 8 is subjected to a process reverse to the process performed by the quantizer 4 and is subjected to inverse DCT by the inverse DCT converter 9. This data is supplied to the adder 10. The adder 10 is supplied with the predicted image data of the reference frame from the motion compensator 12, and these data are added. The output data of the adder 10 is supplied to the frame memory 11. The reference frame image data output from the frame memory 11 is supplied to the motion compensator 12 and the motion detector 13.
[0013]
The motion detector 13 obtains an image motion vector based on the current image data input via the input terminal 1 and the reference frame image data supplied from the frame memory 11. The detection output of the motion detector 13 is supplied to the motion compensator 12 and the quantization weight coefficient controller 14. The motion compensator 12 performs motion compensation prediction based on the detection output of the motion detector 13 and the image data of the reference frame supplied from the frame memory 11. The predicted image data of the motion compensator 12 is supplied to the subtracter 2 and the adder 10.
[0014]
The quantization weight coefficient controller 14 controls the weight coefficient for each coefficient according to the image feature when the DCT-transformed data is quantized. The quantization weight coefficient controller 14 generates an optimum weight coefficient based on the data supplied from the DCT converter 3 and the detection output supplied from the motion detector 13. This weight coefficient is supplied to the quantizer 4 and the inverse quantizer 8.
[0015]
FIG. 2 is a detailed block diagram of the quantizer 4 and the quantization weight coefficient controller 14. The quantizer 4 quantizes the image data based on the quantization weight coefficient block 21 for reducing the high frequency components of the DCT block composed of (8 × 8) pixels and the quantization scale given from the constant rate controller 7. And a quantization block 22 to be converted. Further, the quantization weight coefficient controller 14 detects the degree of motion of the image data based on the motion vector supplied from the motion detector 13 and the DCT coefficient supplied from the DCT converter 3. Based on the high frequency detection block 24 that detects the high frequency component of the image data based on this, and the optimum weighting factor for the image data supplied to the quantizer 4, the detection output of the motion level detection block 23 and the high frequency detection block And a weighting factor determination block 25 which is determined based on 24 detection outputs. The quantization weight coefficient controller 14 configured as described above changes the weight coefficient matrix of the quantization weight coefficient block 21 according to the image feature.
[0016]
That is, when the motion level is detected as greater than the first threshold level in the motion level detection block 23, there is no visual problem even if the spatial resolution is lowered. The block 21 is given a weighting factor that reduces the coefficient of the high frequency component . The weighting coefficient is for DCT coefficient data generated as a result of DCT. The weighting coefficient changes with a slope from the weighting coefficient for the lowest frequency component of the DCT block to the weighting coefficient for the highest frequency component. When reducing the coefficient of the high frequency component, processing for increasing the slope of the weighting coefficient is performed. Conversely, when increasing the coefficient of the high frequency component, processing for decreasing the slope of the weighting coefficient is performed. Based on this weighting coefficient, the coefficient data by the quantization weighting coefficient block 21 is weighting processing. As a result, the high-frequency component is reduced to reduce the data, and image quality deterioration due to an unnecessary high-frequency component can be prevented.
[0017]
On the other hand, when the motion level detection block 23 detects that the motion level is smaller than the second threshold level, the high frequency level detection block 24 uses the image data based on the DCT coefficient supplied from the DCT converter 3. High frequency is detected. When the high frequency detected by the high frequency detection block 24 is equal to or higher than the third threshold level, the high frequency component in the image data is transmitted more faithfully. That is, in this case, the weighting factor determination block 25 reduces the slope of the weighting factor supplied to the quantization weighting factor block 21. Therefore, the screen is clear and the image quality as a whole can be improved. When the motion level detection block 23 detects that the motion level is smaller than the second threshold level and the high frequency level detection block 24 detects that the high frequency level of the image data is equal to or lower than the fourth threshold level, A clearer image is obtained by transmitting unnecessary high frequency components. That is, in this case, the weighting factor determination block 25 increases the gradient of the weighting factor supplied to the quantization weighting factor block 21 and reduces unnecessary high frequency components. Therefore, the screen is clear and the image quality as a whole can be improved.
[0018]
FIG. 3 is a circuit diagram of the quantization weight coefficient controller 14. As the weighting factor, any one of the default value, the weighting factor when the inclination is large, and the weighting factor when the inclination is small is appropriately used. The detection output of the motion detector 13 is supplied to the motion detector 23. The degree of motion at the degree-of-motion detector 23 is detected as the sum of absolute values of motion vectors (Σ | MV |). In addition, two threshold levels TL1 and TL2 (TL1> TL2) are set in the motion detector 23. When the degree of motion is large (Σ | MV |> TL 1), the sum of absolute values is supplied from the motion degree detector 23 to the OR circuit 31. At this time, since the high frequency component of the image data is irrelevant, the weighting coefficient when the inclination is large with respect to the quantization weighting coefficient block 21 connected in the subsequent stage is output as it is.
[0019]
When the motion level detector 23 detects the motion level of the image data as TL1> Σ | MV |> TL2, the motion level detector 23 supplies the absolute value sum of the motion levels to the OR circuit 32. The Also at this time, the high frequency component of the image data is not related to determining the weighting coefficient. Therefore, a default value is supplied to the quantization weight coefficient block 21 connected to the subsequent stage based only on the output supplied from the motion detector 23.
[0020]
Hereinafter, a case where the motion detector 23 detects that the degree of motion is small (TL2> Σ | MV |) will be described. In this case, the absolute value sum is supplied from the motion detector 23 to the AND circuits 33, 34 and 35. If it is detected that the motion vector is small, the high frequency component of the image data is related to determining the quantization scale. That is, in this case, based on the DCT coefficient supplied from the DCT converter 3, the high frequency detector 24 detects the absolute value sum (Σ | HF |) of the high frequency. In the high frequency detector 24, two threshold levels TL3 and TL4 (TL3> TL4) are set. When there are many high frequency components (Σ | HF |> TL3), the high frequency detector 24 supplies the absolute value sum of the high frequency at that time to the AND circuit 35. By giving two inputs to the AND circuit 35, the weighting coefficient (small inclination) when the inclination is small is output to the quantization weighting coefficient block 21, and the high frequency component in the image data is transmitted more faithfully.
[0021]
When the high frequency detector 24 detects TL 3> Σ | HF |> TL 4, the absolute value sum of the high frequencies in the image data is supplied to the AND circuit 34. As a result, a default value is output from the OR circuit 32. Further, when it is detected that TL4> Σ | HF | (the number of high frequency components is small), the absolute value sum of the high frequency in the image data is supplied to the AND circuit 33. As a result, the weighting coefficient (large inclination) when the inclination is large is output from the OR circuit 31 to the quantization weighting coefficient block 21, and the high frequency component in the image data is reduced. As can be seen from the above description, Σ | MV | (sum of absolute values of motion) takes precedence over Σ | HF | (sum of absolute values of high frequency components).
[0022]
FIG. 4 is a block diagram when the weighting coefficient is partially changed in one screen. One screen of video data from the input terminal 40 is divided into a plurality of screens by a demultiplexer 41. The output of the demultiplexer 41 is supplied to each of the DCT converters 42a to 42n. The image data of each DCT converter is supplied to each of the corresponding quantizers 43a to 43n. Further, the DCT coefficient of each DCT converter is supplied to each of the corresponding quantization weight coefficient controllers 45a to 45n. Each quantization weight coefficient controller is supplied with a motion vector of each part in the screen from the corresponding motion detectors 44a to 44n. The quantized outputs of the quantizers 43 a to 43 n are supplied to the multiplexer 46. The multiplexed data for one screen is synthesized by the multiplexer 46, and the output of the multiplexer 46 is output from the output terminal 47.
[0023]
For example, in the case of image data containing a lot of high-frequency components such that the upper half of the screen is blue sky and the lower half is a flower garden, the upper half outputs the weighting coefficient when the inclination is large to the quantization weighting coefficient block. As a result, the upper half of the image data is transmitted with its high-frequency component reduced. On the other hand, the lower half outputs the weighting coefficient when the slope is small to the quantization weighting coefficient block. As a result, for the lower half of the image data, the high frequency component is faithfully transmitted.
[0024]
【The invention's effect】
According to the present invention, quantization is performed with a weight coefficient having a large gradient when an image is moving rapidly. Therefore, the code generation amount can be suppressed to a quantization scale decreases, thereby improving the overall image quality. On the other hand, when the image is small in motion and has many high frequency components, the gradient of the weight coefficient is reduced. Thereby, even a high frequency component is transmitted and a clearer screen is obtained. In addition, even if the movement is small, the gradient of the weighting coefficient is increased when the image has few high frequency components. This makes it possible to such conspicuous mosquito noise strange.
[Brief description of the drawings]
FIG. 1 is a block diagram of an encoding apparatus according to the present invention.
FIG. 2 is a detailed block diagram of a quantizer and a quantization weight coefficient controller.
FIG. 3 is a circuit diagram of a quantization weight coefficient controller.
FIG. 4 is a block diagram when a weighting coefficient is partially changed within one screen.
FIG. 5 is a block diagram illustrating an example of an encoding device.
[Explanation of symbols]
3 DCT converter 4 Quantizer 13 Motion detector 14 Quantization weight coefficient controller 21 Quantization weight coefficient block 22 Quantization block 23 Motion degree detection block 24 High frequency detection block 25 Weight coefficient determination block

Claims (2)

In an encoding device for encoding image data,
DCT conversion means for DCT converting the image data;
Quantization means for quantizing the image data converted by the DCT conversion means;
A degree of motion detecting means for detecting a degree of motion based on the magnitude of the motion vector of the image data;
High frequency detection means for detecting the high frequency of the image data based on the magnitude of the high frequency component of the DCT coefficient supplied from the DCT conversion means;
Based on the degree of motion determined by the degree-of-motion detection means and the high frequency detected by the high frequency detection means, the control means for variably controlling the quantization weight coefficient of the quantization means,
The control means includes
If the degree of motion detected by the degree-of-motion detection means is greater than a first threshold, the quantization weight coefficient of the quantization means is controlled so as to reduce the high frequency side of the DCT coefficient,
When the degree of motion detected by the degree-of-motion detection means is smaller than the first threshold value and the high frequency degree detected by the high-frequency degree detection means is smaller than the second threshold value, the high frequency range of the DCT coefficient Controlling the quantization weight coefficient of the quantization means so as to reduce the side,
When the degree of motion detected by the degree-of-motion detection means is smaller than the first threshold value and the high frequency degree detected by the high-frequency degree detection means is larger than the second threshold value, the DCT coefficient is high. An encoding apparatus for controlling a quantization weighting coefficient of the quantization means so as to increase a band side. Screen division means for dividing the image data of the image data into a plurality of screens;
2. The encoding apparatus according to claim 1, wherein the control means determines a quantization weight coefficient of the quantization means corresponding to each image data divided by the screen dividing means.

Images