JP3200518B2 - Image signal encoding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a highly efficient coding method for television signals and the like, in which a limited amount of information is adaptively allocated according to local characteristics of an image, thereby improving the image quality of a decoded image. The present invention relates to an image signal encoding device for improving.

[0002]

2. Description of the Related Art In highly efficient coding of an image signal such as a television signal, a region where coding noise is easy to detect is quantized by a fine quantization step, and a region where coding noise is hard to detect is coarse quantization step. Is often used.

FIG. 5 is a diagram for explaining an example of a conventional encoding device. In this example, it is assumed that standard motion compensation and discrete cosine transform are used as an image encoding device. Also, as one of the human visual characteristics, masking is such that a flat area with little change in pixel value is noticeable when the pixel value fluctuates even a little, and an area with drastic change in pixel value is inconspicuous even if the pixel value fluctuates slightly. Because it is effective,
Consider a case where the variance of pixel values in a frame is used as a feature amount for determining a quantization step.

First, an image input from an input terminal 901 is divided into N × N small blocks by a small block dividing unit 902. From the small block image signal 903 and the image signal of the previous frame stored in the frame memory 914, a motion vector 905 is obtained in a motion vector detection unit 904, and based on the motion vector 905, a motion compensation unit 906 Motion compensation is performed, and the subtractor 9
After the motion compensation prediction error signal 908 is obtained at 07,
The discrete cosine transform unit 909 performs a discrete cosine transform on the motion compensation prediction error signal 908.

On the other hand, the image signal 90 divided into small blocks is
For 3, the intra-frame variance 916 is calculated by the following calculation in the intra-frame variance calculation unit 915 and sent to the quantization step calculation unit 917.

V = Σ _j Σ _i (s _ij −s _M ) ² / (N × N) where s _ij represents a pixel value in a small block, and s _M represents an average thereof. Also, は_j represents the sum from j = 1 to N, and Σ _i represents the sum from i = 1 to N.

Next, in the quantization step calculator 917, instead of the average value of the intra-frame variance of the encoding target frame, the average value of the intra-frame variance V ′ _M of the frame encoded immediately before the encoding target frame is used. Is used to determine the magnitude of the intra-frame variance (V) 916 obtained for each small block. The parameter A representing this magnitude relationship is
It is calculated from the following equation.

A = (2 · V + V ′ _M ) / (V + 2 · V ′ _M ) A is the value of the intra-frame variance (V) 916 obtained for each small block of the frame encoded immediately before the encoding target frame. When the average value of the intra-frame variance is smaller than V ′ _{M, the} value is smaller than 1.0 with 0.5 as the minimum value, and the intra-frame variance (V) obtained for each small block
If 916 is larger than the average value V ′ _M of the intra-frame variance of the frame coded immediately before the frame to be coded, the value is larger than 1.0 with 2.0 as the maximum value.

The quantization step Q obtained from the buffer memory occupancy 918 to output encoded data at a constant bit rate is corrected by A, and the actually used quantization step Q _act is determined. .

Q _act = A · Q The quantization step Q obtained from the buffer memory occupancy 918 is corrected by A, so that when the intra-frame variance (V) 916 obtained for each small block is large, the coarse quantum The process for setting the conversion step and setting the small portion finely is performed.

The determined quantization step (Q _act ) 91
9, the discrete cosine transform coefficient 920 is calculated by the quantization unit 9.
After being quantized at 10 and subjected to variable-length coding by the variable-length coding unit 921, the data is input to the buffer memory 922 and output to the coded data output terminal 923 at a constant bit rate.

The quantized discrete cosine transform coefficients are inversely quantized by an inverse quantizer 911 and inverse discrete cosine transformed by an inverse discrete cosine transform unit 912, and then added to the motion-compensated previous frame data. The data is added by the unit 913 and stored in the frame memory 914 for motion compensation prediction of the next frame. Further, the intra-frame variance average value calculation unit 92
In step 4, the average value of the intra-frame variance of the encoding target frame is calculated and used as the estimated value of the average value of the intra-frame variance of the next frame to be encoded.

According to such a method, a portion where the pixel value is flat can be finely quantized and a portion where the pixel value changes rapidly can be roughly quantized, so that a limited amount of information can be distributed according to the masking effect. As a result, the image quality of the encoded image can be improved. In addition, since the average value of the intra-frame variance of the immediately preceding frame is used as a predicted value of the average value of the intra-frame variance of the current frame, the intra-frame variance of the current frame before encoding is performed. It is not necessary to find the average value of.

[0014]

In the conventional method, the distribution of the feature of the immediately preceding frame is used as the estimated value as the distribution of the feature of the frame to be encoded. In the case where the distribution state of the feature amount changes greatly, there is a problem that the estimation is deviated because the distribution state of the feature amount differs greatly between the immediately preceding frame and the encoding target frame.

As an example, FIG. 6 shows a case where the average value of the intra-frame variance greatly changes between frames. As shown in FIG. 6, the quantization control in the case where the distribution state of the feature amount greatly changes between frames is performed as follows.

First, as shown in FIG. 7A, when the average value of the intra-frame variance of the immediately-encoded frame used as the estimated value is smaller than the average value of the intra-frame variance of the encoding target frame, If the average value of the intra-frame variance of the frame to be coded is used, a part of the small block to be set so that the fine quantization is performed is estimated as the estimated value of the intra-frame variance of the immediately coded frame. By using the average value, coarse quantization may be performed.

On the other hand, as shown in FIG. 7B, the average value of the intra-frame variance of the immediately-encoded frame used as the estimated value is larger than the average value of the intra-frame variance of the encoding target frame. In this case, if the average value of the intra-frame variance of the frame to be coded is used, a part of the small block to be set so as to perform coarse quantization is used as an estimated value within the frame of the immediately coded frame. By using the average value of the variance, fine quantization may be performed.

As described above, in the case of an image signal in which the distribution state of the feature amount of the immediately preceding frame and the distribution state of the feature amount of the encoding target frame are significantly different, an inappropriate quantization step is set. There was a problem that the quantization control did not work properly.

[0019]

SUMMARY OF THE INVENTION The present invention has been made to solve this problem, and has the following means.

FIG. 1 is a block diagram showing the principle of the present invention.
First, the small block dividing means 1 receives a digital image signal such as a television signal and divides the image data of the encoding target frame into small blocks. The small block feature value calculation means 2 calculates a feature value representing the property of each divided small block. The frame feature amount distribution state calculation unit 3 checks the distribution state of the feature amount over one frame from the result of the small block feature amount calculation unit 2 and stores the result in the frame feature amount distribution state storage unit 4.
The frame feature amount distribution state storage means 4 stores the feature amount distribution states of a plurality of past or future frames.

The current frame feature distribution state estimating means 5
The distribution state of the characteristic amount of the encoding target frame is estimated from the distribution state of the characteristic amount of a plurality of past or future frames stored in the frame characteristic amount distribution state storage unit 4. The quantization characteristic determination unit 6 refers to the estimated distribution state of the feature amount of the encoding target frame and the feature amount obtained for each small block in the encoding target frame by the small block feature amount calculation unit 2. Thereby, the quantization characteristic of the small block to be encoded is determined.

The quantization means 7 performs quantization for each small block according to the quantization characteristics determined by the quantization characteristic determination means 6. The variable length coding means 8 codes the result.

[0023]

According to the present invention, the distribution state of the feature amounts of a plurality of past or future frames encoded before the frame to be encoded is stored. Next, as shown in FIG. 2, the distribution state estimation value of the feature amount of the encoding target frame is estimated using the stored distribution state of the feature amount of a plurality of past or future frames.

D ′ _n = f (D _ni ,..., D _n−1 ,
D _{n + 1} ,..., D _{n + j} ) where D ′ _n is the estimated value of the distribution of the feature of the encoding target frame, and D _k is the distribution of the feature of the already encoded past or future frame. The states i and j represent the number of past and future frames used for estimating the feature distribution state of the encoding target frame, respectively.

In this manner, when estimating the distribution of the feature of the encoding target frame, the error between the distribution of the feature of the encoding target frame and the estimated value can be reduced.

[0026]

FIG. 3 is a diagram showing a block configuration of an embodiment of the present invention. In the present embodiment, MPE is used as the image encoding method.
Using motion compensation + discrete cosine transform coding as used in G2, etc., using the intra-frame variance as a feature value, the average value of the entire frame as an index of the distribution state, and the average value of the luminance variance of the encoding target frame Is estimated from the average value of the intra-frame variances of the two frames encoded immediately before by linear approximation. Also, assume that a frame using only motion compensation from the past (hereinafter, referred to as a P picture) and a frame using motion compensation from the past and the future (hereinafter, referred to as a B picture) are encoded with the following structure. .

... B _n−2 P _n−1 B _n−1 P _n B _n P _{n + 1} B _{n + 1} P _{n + 2} B _{n + 2} P _{n + 3} . It is as follows. It is _{... P n-1 B n-} 2 P n B n-1 P n + 1 B n P n + 2 B n + 1 P n + 3 B n + 2 ... First, inputted from the input terminal 101 image, The small block dividing unit 102 divides the image into N × N small blocks.
A motion vector detection unit 104 obtains a motion vector 105 from the image signal 103 of the small block and the image signal of the previous frame stored in the frame memory 114, and the motion compensation unit 1
At 06, motion compensation is performed. After the motion compensation prediction error signal 108 is obtained by the subtractor 107, the discrete cosine transform unit 109 performs discrete cosine transform on the motion compensation prediction error signal 108.

On the other hand, the image signal 10 divided into small blocks
For 3, the intra-frame variance calculation unit 115 obtains the intra-frame variance 116 by the following calculation, and sends it to the quantization step calculation unit 117.

V = Σ _j Σ _i (s _ij −s _M ) ² / (N × N) where s _ij represents a pixel value in a small block, and s _M represents an average thereof. Also, は_j represents the sum from j = 1 to N, and Σ _i represents the sum from i = 1 to N.

Subsequently, the intra-frame variance average value estimating unit 12
5, the intra-frame variance average memories 127 to 12
8, the average values D _n-1 and D _{n of the} intra-frame variances of the two frames coded immediately before the frame to be coded.
_{From n-2} , the average value of the intra-frame variance of the encoding target frame is estimated by linear approximation. This linear approximation is performed as follows.

First, in the case of a P picture, _assuming that the frame to be encoded is P _n , the two frames encoded immediately before are D _n−2 = P _n−1 and D _n−1 = B _n−2. Becomes Therefore, D _Pn is estimated by linear approximation from the average values of the intra _- frame variances of P _n-1 and B _n-2 . FIG. 4A shows a method of estimating the average value of the intra-frame variance in the case of a P picture. As is clear from the figure, the estimated value D' _Pn is obtained from the following equation.

D ′ _Pn = −2 · D _Bn−2 + 3 · D _{Pn−1 In} the case of a B picture, the frame to be coded is
_Assuming that _n , the two frames encoded immediately before are D _n−2 =
B _n-1 and D _n-1 = P _{n + 1} . Therefore, this B
_DBn is estimated by linear approximation from the average value of the intra-frame variances of _n-1 and _{Pn + 1} . FIG. 4B shows a method of estimating the average value of intra-frame variance in the case of a B picture. As is clear from the figure, the estimated value D' _Bn is obtained from the following equation.

D '_Bn= (D_Bn-1+ 2 · D_{Pn + 1}) / 3 Next, in the quantization step calculation unit 117, the frame
The encoding target file estimated by the internal variance average value estimation unit 125
Estimated value of the intra-frame variance average (D ′ _n) 12
6, the amount within the frame obtained for each small block
The magnitude of the scatter V is determined. A parameter that represents this magnitude relationship
Data A is calculated by the following equation.

A = (2 · V + D ′ _n ) / (V + 2 · D ′ _n ) A is the intra-frame variance (V) 116 obtained for each small block is the estimation of the intra-frame variance average value of the encoding target frame. If the value (D ' _n ) is smaller than 126,
Becomes a value less than 1.0 to 0.5 as the minimum value, from each small block basis in the frame variance obtained (V) 116 is an estimate of the frame in a distributed average values of the encoding target frame (D _'n) 126 If it is larger, the maximum value will be 2.0 and the value will be larger than 1.0. With this A, the quantization step Q obtained from the buffer memory occupancy 118 is corrected, and the quantization step (Q _act ) actually used is corrected.
119 is determined.

Q _act = A · Q By correcting the quantization step Q obtained from the buffer memory occupancy 118 by A, a portion where the intra-frame variance (V) 116 obtained for each small block is large is a coarse quantum The process for setting the conversion step and setting the small portion finely is performed.

The determined quantization step (Q _act ) 11
9, the discrete cosine transform coefficients 120
After being quantized by 10 and subjected to variable-length encoding by the variable-length encoding unit 121, the data is input to the buffer memory 122 and output to the encoded data output terminal 123 at a constant bit rate.

The quantized discrete cosine transform coefficients are inversely quantized by an inverse quantizer 111, inverse discrete cosine transformed by an inverse discrete cosine transform unit 112, and added to data of a motion-compensated previous frame. The data is added by the adder 113 and stored in the frame memory 114 for motion compensation prediction of the next frame.

When the frame head detecting section 129 detects the input of the frame to be coded, the switches 130 to 131 are closed, and the coding before the frame stored in the intra-frame variance average value memory 127 is performed. The intra-frame variance average of the obtained frame is transferred to the intra-frame variance average value memory 128, and is stored in the intra-frame variance average value memory 127 for the immediately encoded frame calculated by the intra-frame variance average value calculation unit 124. The intra-frame variance average is stored.

In the embodiment described above, the image signal encoding method is motion compensation + discrete cosine transform encoding, the feature amount is the variance in the frame, and the index of the distribution state of the feature amount is the average value. is not. Further, a past or future frame and its number used for estimation, a method of estimating the distribution state of the feature amount of the encoding target frame, and a method of calculating the quantization control parameter can be arbitrarily determined in the embodiment of the present invention.

[0040]

As described above, according to the present invention,
When performing adaptive quantization control on local features of an image signal, the distribution of the feature amount of the encoding target frame is not determined in advance, and the region where degradation is conspicuous and the region where degradation is not conspicuous are determined. It is possible to set an appropriate quantization step. As a result, in regions where deterioration is conspicuous, fine quantization can be used to make the deterioration inconspicuous, and in regions where deterioration is not noticeable, the amount of information can be reduced by performing coarse quantization. Visual deterioration can be reduced under the amount of information, and image quality can be improved.

[Brief description of the drawings]

FIG. 1 is a principle block diagram of the present invention.

FIG. 2 is a diagram illustrating an estimated value of a distribution state of a feature amount of an encoding target frame.

FIG. 3 is a block diagram of an embodiment of the present invention.

FIG. 4 is a diagram illustrating a method for estimating an average value of intra-frame variance.

FIG. 5 is a diagram illustrating a configuration example of a conventional encoding device.

FIG. 6 is a diagram illustrating a change in a distribution state of a feature amount and an estimated value.

FIG. 7 is a diagram illustrating an influence on a quantization control due to a prediction error of a distribution state of a feature amount.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Small block division means 2 Small block feature quantity calculation means 3 Frame feature quantity distribution state calculation means 4 Frame feature quantity distribution state storage means 5 Current frame feature quantity distribution state estimation means 6 Quantization characteristic determination means 7 Quantization means 8 Variable length Encoding means 101 Input terminal 102 Small block division unit 103 Small block image signal 104 Motion vector detection unit 105 Motion vector 106 Motion compensation unit 107 Subtractor 108 Motion compensation prediction error signal 109 Discrete cosine transform unit 110 Quantization unit 111 Inverse quantum Transformer 112 Inverse discrete cosine transform 113 Adder 114 Frame memory 115 In-frame variance calculator 116 In-frame variance 117 Quantization step calculator 118 Buffer memory occupancy 119 Quantization step 120 Discrete cosine transform coefficient 121 Variable length encoder 1 Reference Signs List 22 buffer memory 123 output terminal 124 intra-frame variance average value calculation unit 125 intra-frame variance average value estimation unit 126 estimated value of intra-frame variance average value 127-128 intra-frame variance average value memory 129 frame head detection unit 130-131 switch

Claims (1)

(57) [Claims] 1. An image signal encoding apparatus for encoding a digital image signal by changing a quantization characteristic according to a distribution state of a feature amount of image data, wherein the image data of an encoding target frame is divided into small blocks. means for, and a feature quantity that represents the nature of the small block for each small block
Means for determining a frame variance for the pixel value each, the small blanking as distribution of the characteristic quantity of over one frame
Means for storing check the average value in the frame is determined for each lock dispersion, means for estimating the distribution of the feature values of the encoding target frame from the feature quantity of distribution of the plurality of frames in the past or future Means for determining the quantization characteristic of the encoding target small block by referring to the estimated distribution state of the characteristic amount of the encoding target frame and the characteristic amount obtained for each small block in the encoding target frame; An image signal encoding device comprising: