JP2000350211A - Method and device for encoding moving picture - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a moving picture encoding device which can encode an image with picture quality matching its scene contents while maintaining the encoding bit rate at a specified value. SOLUTION: This device has a feature quantity calculation part 31 which divides an input moving picture signal 100 into temporally successive scenes consisting of at least one frame and calculates statistical feature quantities by the scenes, an encoding parameter generation part 32 which generates encoding parameters by the scenes according to the calculated statistical feature quantities, an encoding quantity decision part 33 which decides how much the generated code quantity 133 of a code sequence 111 generated by an encoding part 10 encoding the input moving picture signal 100 is larger or smaller than a target code quantity 134, and an encoding parameter correction part 34 which corrects the encoding parameters according to the decision result; and the code sequence generated by encoding the input moving picture signal 100 according to the corrected encoding parameters is sent out of a buffer 21 as an encoding output 200 to a transmission line or storage medium.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture compression coding technique such as the MPEG system used for a moving picture transmission system or a picture database system over the Internet or the like. The present invention relates to a moving picture coding method and a moving picture coding apparatus capable of providing a coherent and easy-to-view decoded image for each scene without increasing the data size by performing coding.

[0002]

2. Description of the Related Art MPEG which is an international standard for video coding
The method is a technique for compressing a moving image by combining motion compensation prediction, DCT (discrete cosine transform) and variable-length coding, as is well known. For example, reference 1: "MPEG" edited by the Television Society, Ohmsha, And so on.

[0003] In a conventional moving picture encoding apparatus based on the MPEG system, the compressed moving picture data is transmitted over a transmission path with a prescribed transmission rate or recorded on a storage medium having a limited recording capacity. A process called rate control is performed in which encoding is performed by setting encoding parameters such as a frame rate and a quantization width so that the bit rate of the encoded bit stream to be transmitted has a specified value. In the conventional rate control, a method of determining a frame rate in accordance with a generated code amount in encoding of a previous frame for a fixed quantization width is often used.

FIG. 10 is a diagram showing a conventional rate control. A frame rate is determined based on a difference (margin) between a frame skip threshold preset in accordance with a capacity of a buffer for temporarily storing an encoded bit stream and a current buffer amount, and a buffer amount is determined. When the value is smaller than the value, encoding is performed at a constant frame rate, and when the buffer amount exceeds a threshold value, frame skip is performed to control the frame rate to be reduced.

However, in this method, when the code amount generated in the previous frame is large as shown in FIG. 10, the frame is skipped until the buffer amount becomes equal to or less than the frame skip threshold. Is too wide, which makes it unnatural for a movie.

That is, in the conventional rate control, since the frame rate and the quantization width are basically set irrespective of the content of the image, the frame interval becomes too wide in a scene in which the motion of the object in the image is severe. In some cases, the motion becomes unnatural or the image is distorted due to an inappropriate quantization width, making the image visually difficult to see.

On the other hand, there is also known a method of performing rate control by a technique called two-pass encoding. For example, as shown in Document 2: Japanese Patent Laid-Open No. Hei 10-336675, the same moving image file is encoded twice, and the characteristics of the entire moving image file are analyzed in the first encoding. ,
Based on this, the second encoding is performed by setting an optimal encoding parameter, and the encoded bit stream obtained in the second encoding is transmitted or recorded. However, also in the two-pass encoding, conventionally, the encoding parameters are basically set irrespective of the content of the image, and have the same problem as described above.

[0008]

As described above, in the conventional moving picture coding apparatus, the coding parameters such as the frame rate and the quantization width are set at the time of rate control regardless of the contents of the picture. In particular, in a scene in which the movement of an object in an image is severe, the frame rate sharply lowers and the movement becomes unnatural, or the image quality is easily deteriorated such that the image is distorted due to an inappropriate quantization width. Was.

SUMMARY OF THE INVENTION It is therefore an object of the present invention to provide a moving picture coding method and apparatus capable of coding with an image quality suitable for the scene content of an image while maintaining a coding bit rate at a designated value.

[0010]

When compressed moving image data is recorded on a storage medium having a limited storage capacity or downloaded over the Internet or the like, it is suitable for a scene situation under a predetermined data size as much as possible. It is important to perform efficient coding at the set frame rate and quantization width. For this purpose, the amount of generated code is not always related to the content of the scene, so in order to obtain an easy-to-view image, it is necessary to determine the encoding parameters based on the movement of objects in the scene and the content of the scene. desired.

In order to solve the above problems, the present invention divides an input moving image signal into scenes composed of at least one frame that is temporally continuous, calculates a statistical feature for each scene, and The basic feature is that an encoding parameter is generated for each scene based on the feature amount, and the input moving image signal is encoded using the encoding parameter.

Here, the statistical feature amount is calculated by, for example, summing up the magnitude and distribution of motion vectors existing in each frame of the input moving image signal for each scene. The encoding parameters include, for example, at least a frame rate and a quantization width.

[0013] In addition to counting the size and distribution of motion vectors present in each frame as statistical features for each scene, the camera used to obtain the input video signal from the size and distribution of motion vectors. A scene may be classified according to the type of the frame based on the motion and the motion of the object in the image, and the encoding parameter may be generated in consideration of the classification of the scene.

When a quantization width in units of macroblocks is generated as an encoding parameter, among macroblocks in a frame to be encoded, a macroblock whose luminance value variance from an adjacent macroblock is equal to or larger than a predetermined value is determined. The quantization width of a macroblock block in which an edge of an object exists may be made relatively smaller than that of other macroblocks.

As described above, according to the present invention, an encoding parameter used for encoding an input video signal is generated for each scene based on a statistical feature calculated for each scene of the input video signal. Further, it is possible to prevent the frame rate from decreasing when the movement of the object or the camera is intense and the visual quality of the decoded image from decreasing.

Further, by reflecting the feature amount of the image based on the motion of the object in the image, the motion of the camera, and the like in the coding parameter, the frame rate is changed based on the feature amount, or the quantization width is changed for each macroblock. It is possible to obtain a coherent and good decoded image for each scene even with the same generated code amount.

Further, the present invention can be applied to a moving picture coding apparatus for coding an input moving picture signal of the same moving picture file twice or more times.
That is, a first code for encoding an input video signal using a first encoding parameter generated for each scene based on a statistical feature amount calculated for each scene of the input video signal And determining whether the generated code amount of the code string generated by this coding is larger or smaller than the target code amount, and correcting the first coding parameter based on the determination result to obtain the second coding parameter. The second coding parameter is used to perform a second coding on the input video signal to generate a code sequence, which is output as a coded output.

The coding parameters generated as described above are corrected while constantly monitoring the generated code amount, and
By repeating the coding one or more times, it is possible to realize coding capable of obtaining good decoded image quality with a data size equal to or smaller than the target code amount.

[0019]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration of a video encoding device according to an embodiment of the present invention. The input moving image signal 100 is a moving image signal (video signal) reproduced by a video recording / reproducing device such as a digital VTR or a DVD system capable of repeatedly reproducing the same signal a plurality of times, and is input to the encoding unit 10. . Encoding unit 10
Has the same configuration as the moving picture coding apparatus based on the MPEG system in this example.

In the encoding unit 10, first, the frame memory 11 stores the image signal of the encoding target frame selected from the input moving image signal 100. The moving image signal 101 of the encoding target frame read from the frame memory 11 is input to the subtractor 12 and the motion compensation predictor 19. The motion compensation predictor 19 has a built-in motion vector detection circuit, and generates a prediction signal 109 by motion compensation prediction.

Although there are three types of picture types, ie, I picture, P picture and B picture, in the current frame to be coded, the motion compensation predictor 19 does not perform motion vector detection when the current frame to be coded is an I picture. Prediction signal 10
When the encoding target frame is a P picture or a B picture, a motion vector is detected for each macro block from the video signal 101 of the encoding target frame, and a prediction signal 109 is generated.

The detection of the motion vector is performed by converting the macroblock between the moving picture signal 101 of the frame to be coded and the reference picture signal, which is stored in the video memory 18 and is a local decoded signal of the already coded frame. This is done in units called. Specifically, by detecting the macroblock having the highest correlation of the reference image signal with respect to the macroblock of the moving image signal 101 of the encoding target frame,
Information indicating from which macro block of the reference image signal the macro block of the moving image signal 101 has been detected is detected as a motion vector.

The motion compensation predictor 19 performs motion compensation on the reference image signal using the motion vector, and generates a prediction signal 109. That is, in the motion compensation predictor 19, a preferred prediction mode is selected from the motion compensation inter-frame prediction and the intra-frame encoding (prediction signal = 0) for directly encoding the moving image signal 101 of the encoding target frame. The prediction signal 109 corresponding to the selected prediction mode is output.

In the subtracter 12, the video signal 101 of the encoding target frame and the prediction signal 1 from the motion compensation
09 and a prediction error signal 102 is generated. The prediction error signal 102 is calculated by the discrete cosine converter 1
3, a discrete cosine transform (DCT) is performed for each block of a fixed size. The DCT coefficient data 103 obtained by the discrete cosine transform is quantized by the quantizer 14. DCT coefficient data 10 quantized by the quantizer 14
4 is bifurcated, and is input to the variable length encoder 20 on the one hand, and is inversely quantized by the inverse quantizer 15 on the other hand. The inversely quantized DCT coefficient data 105 is further subjected to inverse discrete cosine transform (inverse DCT) by the inverse discrete cosine transformer 16.

Output 10 from inverse discrete cosine transformer 16
6 is added to the prediction signal 109 by the adder 17 to form a local decoded signal 107, which is stored in the video memory 18 as a reference image signal. The reference image signal stored in the video memory 18 is read out by a motion compensation predictor 19, and motion compensation inter-frame prediction is performed.

The motion compensation predictor 19 also outputs a prediction mode of motion compensation prediction and prediction mode / motion vector information 110 indicating a motion vector, and inputs the information to the variable length encoder 20. In the variable length encoder 20, the quantized DCT
Coefficient data 104 and prediction mode / motion vector information 11
0 are respectively subjected to variable-length coding, and a code string of the obtained variable-length code (hereinafter, referred to as a coded bit stream) 111
Is output. The coded bit stream 111 is temporarily stored in the buffer 21.

In this embodiment, the same moving image file,
That is, the input moving image signal 100 of the same content such as the same movie
In contrast, two encodings are performed by the basic operation described above. Further, in the present embodiment, the image feature amount calculation unit 31, the encoding parameter generation unit 32, and the code amount determination unit 3
3 and an encoding parameter correction unit 34, which calculate an image feature amount and set an encoding parameter based on the image feature amount at the first encoding, and the encoded bit stream 111 at the second encoding. Is determined and the encoding parameters are corrected based on the determination.

That is, at the time of the first encoding, the input moving image signal 100 is input to the image characteristic amount calculating section 31 before being input to the frame memory 11, where the statistical characteristic amount of the image (hereinafter referred to as image 130 is calculated. As will be described in detail later, the image feature amount 130 is a statistical feature amount in which at least the magnitude and distribution of a motion vector in each frame of the input moving image signal 100 are totaled for each scene. When calculating the image feature amount 130,
First, a frame whose luminance changes rapidly from the inter-frame difference value of the input moving image signal 100 is detected, and the detected frame is set as the first frame of a scene break. By calculating the number, distribution, size, variance of luminance / color difference, and the like, and averaging these for each scene, a representative value of the feature amount for each scene is obtained as the image feature amount 130.

The information on the image characteristic amount 130 for each scene calculated in this way is input to the encoding parameter generation unit 32. The encoding parameter generation unit 32 generates an encoding parameter 131 for performing encoding such that the data size of the encoded bit stream 111 generated by the variable length encoder 20 is equal to or smaller than the size specified by the user. By applying the image feature amount 130 to an encoding parameter creation formula described later, the encoding parameter 13
1 is generated. Encoding parameter 131 obtained here
Is a frame rate FR and a quantization width QP.

At the time of the first encoding, the moving image signal 101 of the encoding target frame is selected from the input moving image signal 100 in accordance with the value of the frame rate FR generated by the encoding parameter generating unit 32. Memory 10
Is stored in If this frame is an I picture such as a scene break, motion vector detection is not performed and intra-frame coding is performed. If this frame is a P picture or B picture, inter-frame coding based on motion compensation prediction is performed. Become.

In either case of intra-frame encoding or inter-frame encoding, the DCT coefficient data 103 output from the discrete cosine transformer 13 is encoded by the encoding parameter generator 32 in the quantizer 14 for each scene. The quantization is performed by the quantizer 14 according to the generated quantization width QP. The quantized DCT coefficient data 104 is encoded by the variable length encoder 20 together with the prediction mode / motion vector information 110 by the variable length encoder 20 as described above, and further combined with the information of the quantization width QP. And output to the buffer 21 as an encoded bit stream 111.

When the first encoding is completed,
The coding amount determination unit 33 determines whether the generated code amount 133 of the coded bit stream 111 stored in the buffer 21 is larger or smaller than the target code amount 134. Then, according to the determination result, the encoding parameter generated by the encoding parameter generation unit 32 is
4 to be modified.

That is, when the code amount determining section 33 determines that the difference between the generated code amount 133 and the target code amount 134 specified by the user exceeds the threshold, the coding parameter correcting section 34 generates the generated code amount. The coding parameter is corrected so that the amount 133 approaches the target code amount 134, and the second coding is performed according to the corrected coding parameter 136. When the difference between the generated code amount 133 and the target code amount 134 becomes equal to or smaller than the threshold value by the second encoding, the encoded bit stream stored in the buffer 21 is output as the encoded output 200.
This encoded output 200 is sent out on a transmission line or stored in a storage medium.

In the moving picture coding apparatus of the present embodiment,
The image feature amount 130 calculated by the image feature amount calculation unit 31 as described above is a value indicating the intensity of the motion of the image of each scene and the fineness of the image, which is generated by the encoding parameter generation unit 32. Encoding parameter 131 or encoding parameter 1 modified by the encoding parameter modifying unit 34
The encoding is performed by reflecting the result on the line 36.

As a result, as shown in FIG. 2, encoding can be performed according to the encoding parameters (frame rate, quantization width) suitable for the contents of the scene. That is, in scene (j) with little motion, the frame rate is lowered,
In addition, by changing the quantization width for each macroblock in each frame, the decoded image is made to be a relatively fine image, and the image quality of an area that is easily noticed, such as an object or a telop, is prevented. To

On the other hand, in a scene (j + 1) in which the movement is intense, the frame rate is increased and the quantization width is increased, so that a decoded image free from unnatural movement can be obtained. When the quantization width is increased, the decoded image becomes coarse, but in a scene with a lot of motion, the image roughness is inconspicuous, so that there is not much problem.

The processing procedure of the moving picture coding apparatus according to the present embodiment will be described with reference to the flowchart shown in FIG. First, the moving image signal 100 is input (step S1).
1) The image feature amount, which is the statistical feature amount of the image for each scene described above, is calculated (step S12). In the present embodiment, the image feature amount calculating step S12 includes scene division,
It consists of three processes of feature amount calculation and scene classification.

Next, an encoding parameter is generated (step S13). This encoding parameter generation step S1
Reference numeral 3 includes four processes of calculating a frame rate, calculating a quantization width, adjusting a frame rate, and setting a quantization width for each macroblock.

Next, the moving picture signal is coded according to the generated coding parameters (step S14). The processing of the encoding step S14 is as described above.

When the first encoding is completed in the encoding step S14, the code amount is determined, that is, whether the difference between the generated code amount and the target code amount is equal to or smaller than a threshold is determined (step S15).

As a result of the determination in the code amount determination step S15, if the difference between the generated code amount and the target code amount exceeds the threshold, the encoding parameter is set so that the difference between the generated code amount and the target code amount becomes small. Is corrected (step S16), and the process returns to step S16 to perform the second encoding.

If the difference between the generated code amount and the target code amount is equal to or smaller than the threshold value in the code amount determination step S15, the coding is terminated, and the coded bit stream obtained in the coding step S14 is used as a coded output. Output (Step S17). Therefore, the encoding may be completed once, or may be performed twice or more times.

Next, the individual processes of the image feature amount calculation unit 31, the coding parameter generation unit 32, the code amount determination unit 33, and the coding parameter correction unit 34, which are characteristic parts of the present embodiment, will be described in further detail.

[Regarding Feature Amount Calculating Unit 31] The feature amount calculating unit 31 first divides a scene as described below, calculates a feature amount, and finally classifies a scene.

<Scene Division> The input moving image signal 100 is
The scene is divided into a plurality of scenes excluding frames such as a flash frame and a noise frame by a difference between adjacent frames. Here, the flash frame is, for example, an interview scene in a news program, and is a frame in which the luminance sharply increases, such as at the moment when a flash (flash) is emitted. A noise frame is a frame in which an image is greatly deteriorated due to camera shake or the like. A specific example of the scene division processing will be described with reference to the flowchart shown in FIG. 4 and FIGS.

First, a difference value of luminance between the i-th frame and the (i + 1) -th frame (hereinafter, referred to as an inter-frame difference value) is calculated (step S21). Then, this inter-frame difference value is set to a certain threshold T predetermined by the user.
hre (step S22). As a result of this comparison, if the inter-frame difference value is less than Theh, i
= I + 1 (step S23), and the process returns to step S21.

If the inter-frame difference value between the i-th frame and the (i + 1) -th frame is equal to or greater than Theh, then i +
An inter-frame difference value between the first frame and the (i + 2) th frame is calculated (step S24). Then, the inter-frame difference value is similarly compared with the threshold value Threshold (step S25).

As a result of the comparison in step S25, when the inter-frame difference value is equal to or greater than Theh, that is, as shown in FIG. 5, the inter-frame difference value between the i-th frame and the (i + 1) -th frame, the i + 1-th frame and the (i + 2) -th frame If all of the inter-frame difference values of the frames are equal to or larger than the threshold value Thr (large difference), it is determined that the i-th frame and the (i + 1) -th frame belong to different scenes, and i + 1
The ith frame is a scene break. That is, i
The frame at the end is the last frame of the scene break,
The (i + 1) th frame is set as the first frame of a scene break.

On the other hand, as a result of the comparison in step S25, if the inter-frame difference value is less than Theh, that is, the inter-frame difference value between the i-th frame and the (i + 1) -th frame is equal to or larger than the threshold value Thr, but the (i + 1) -th frame And the (i + 2) th frame have a threshold value Thr
If the number of frames is less than e, for example, as shown in FIG. 6, the (i + 1) th frame is a flash frame (or a noise frame), and the (i) th, (i + 1) th and (i + 2) th frames belong to the same scene, and (i + 1) th Frame is not a scene delimiter, and i = i + 2
(Step S26), and returns to step S21.

Similarly, in order to cope with an image having a large number of flash frames, the following method can be used as a method for preventing erroneous determination as a scene break when k flash frames are continuous. .

That is, the inter-frame difference values between the i-th frame and the (i + 1) -th, (i + 2) -th,... I + k-th frames are all greater than or equal to Thr, and the inter-frame difference values between the i-th frame and the (i + k + 1) -th frame are also T
hre or more, the i-th frame and the (i + 1) -th frame belong to different scenes, and i
The + 1st frame is defined as a scene break.

On the other hand, the i-th frame, the (i + 1) -th frame,
The inter-frame difference value of the (i + 2) -th, (i + k) -th frame is greater than or equal to Thr, and the i-th frame and i + k +
When the inter-frame difference value of the first frame is less than Threshold, the (i + 1) th, (i + 2) th,... I + kth frames are flash frames or noise frames, and the (i) th, ((i + k) th, (i + k + 1) th) frames are noise frames. The frames are determined to belong to the same scene, and are not set as scene breaks.

It is desirable that the user can select and set in advance whether to treat consecutive flash frames or noise frames as scene breaks as described above.

<Characteristic Calculation> Next, the input moving image signal 10
For all 0 frames, the image feature amount such as the number of macroblocks in which a motion vector exists in the frame (the motion vector is not 0), the average of the magnitude of the motion vector, and the variance of luminance and color difference are calculated. At this time, the feature amount is calculated only for frames other than the frames determined to be any of the scene-separated frame, the flash frame, and the noise frame by the above-described scene division. further,
The feature amount is averaged for each scene determined by the scene division, and the average value is used as a representative value of the feature amount of each scene.

<Scene Classification> In the present embodiment, the following scene classification is performed using a motion vector in addition to the scene division and the calculation of the feature amount as described above.

After calculating the motion vector for each frame, the distribution of the motion vector is examined to classify the scene. Specifically, first, the distribution of motion vectors in one frame is calculated, and it is checked which of the five types shown in FIGS. 7A to 7E each frame belongs to.

(A) There is almost no motion vector in the frame (the number of macroblocks whose motion vector is not 0 is Mmin or less). (B) Motion vectors having the same direction and magnitude are distributed over the entire screen (the number of macroblocks in which the motion vectors appear is equal to or greater than Mmax, and the magnitude and direction are within a certain range). (C) A motion vector appears only in a specific portion in a frame (the position of a macroblock where a motion vector appears is concentrated in a specific portion). (D) Motion vectors are distributed radially in the frame. (E) The number of motion vectors in the frame is large, and the directions are also irregular.

Each of the cases (a) to (e) is closely related to the camera used to obtain the input moving image signal 100 and the movement of an object in a captured image. That is, in (a), both the camera and the object are stationary. In (b), the camera is moving in parallel. In (c), the object is moving in a stationary background. In (d), the camera is performing zooming. In (e), the camera and the object are both moving.

Next, as described above, the frames are (a) to
After the type is classified as to which type of (e), the scene divided as described above is classified by frame type. That is, it is determined whether each scene is composed of what type of frames (a) to (e). Then, using the classification result of the determined scene (the type of a frame constituting the scene) and the feature amount calculated as described above, the encoding parameter generation unit 32 uses the encoding parameter as follows. A certain frame rate and quantization width are determined for each scene.

[About the encoding parameter generation unit 32]
The encoding parameter generation unit 32 sequentially calculates the frame rate and the quantization width as described below, and further corrects the calculated frame and quantization width. Further, a process of changing the quantization width for each macroblock block is also performed.

<Calculation of Frame Rate> The encoding parameter generator 32 first determines a frame rate. It is assumed that the feature value calculation unit 31 has calculated the representative value of the motion vector as the representative value of the feature value for each scene as described above. For example, a motion vector representative value MVnum _ j of the j-th scene is a value related to the average number of macroblocks for which the motion in the scene as follows. That is, assuming that the number of macroblocks whose motion vector is not 0 in the i-th frame is MVnum (i), MVn
um _ j is represented by the following equation. MVnum _ j) = average of (j-th MVnum (i of all frames included in the scene)) × (fixed multiple) ... (1) motion vector representative value of the j-th scene MVnum _
j, the frame rate FR of the same j-th scene
(J) is calculated by the following equation. FR (j) = a × MVnum _ j + b + w _ FR ... (2) where, a, b are coefficients for the bit rate or data size user specifies, the w _ FR is a weighting parameter, which will be described later.

[0062] Equation (2) means that motion vector representative value MVnum _ j in the scene increases, to increase the frame rate FR (j) in the same scene. That is, the frame rate is higher for a scene with a larger motion.

[0063] In addition, the motion vector representative value MVnum _ j
In addition to the number of motion vectors in a frame, the sum of absolute values of the magnitudes of motion vectors in a frame, the density, and the like may be used.

<Calculation of Quantization Width> After calculating the frame rate for each scene as described above, the quantization width for each scene is calculated. similar to the quantization scale QP (j) is the frame rate FR (j) for the j-th scene is calculated by the following equation using the motion vector representative value Mvnum _ j in the scene. QP (j) = c × MVnum _ j + d + w _ QP ... (3) where, c, d are coefficients for the bit rate or data size user specifies, the w _ QP is a weighting parameter, which will be described later.

The equation (3) is equivalent to the motion vector representative value MVnu.
m _ j becomes larger, which means increasing the quantization width QP (j). In other words, a scene with a large motion has a larger quantization width, and a scene with a smaller motion has a smaller quantization width to sharpen an image.

<Correction of Frame Rate and Quantization Width> After calculating the frame rate FR (j) for each scene as described above, the image feature amount calculation unit 31 obtained the result of the <scene classification> processing. scene classification results using (type of frames constituting the scene), the weight parameter w _ FR in the formula (2), the weighting parameters in equation (3) w _
The frame rate and the quantization width are corrected by adding the QPs respectively.

That is, when the scene classification result indicates that there is almost no motion vector in the frame (a),
Decrease the frame rate and reduce the quantization width ( w_F
R, w _ QP together to reduce). In the case of (b),
The movement of the camera is raised as much as possible frame rate so as not to be unnatural, quantization width is larger (w _ FR, w
_ Increase both QP). In the case (c), when the motion of the moving object, that is, the size of the motion vector is large, the frame rate is corrected ( w_F).
R is increased). For the (d), since for most object during zooming does not seem to be of interest, the quantization width is made large, increasing as much as possible the frame rate (increasing the w _ FR, also increases w _ QP ). Even in the case of (e), a large quantization width increasing the frame rate (w _ FR, w _ QP together to increase).

[0068] In this way, the set weighting parameter w _ FR, by adding each of w _ QP formula (2) (3), to correct the frame rate and the quantization width.

<When the quantization width is changed for each macroblock> For a macroblock determined to have mosquito noise likely to appear in a frame or a macroblock determined to have a strong edge such as a telop character in a frame,
The image quality may be improved by setting the quantization width smaller than that of other macroblocks.

That is, in the coding procedure shown in FIG. 5 (steps S11 to S15), if the quantization width is changed for each macroblock by the user, mosquito noise is generated in the frame. For macroblocks that are determined to be easy to appear or macroblocks that are determined to have strong edges such as telop characters, partial image quality is set by setting the quantization width relatively smaller than other macroblocks Make improvements.

Here, a method for further correcting the quantization width QP (j) for each scene obtained by the equation (5) will be described. For the encoding target frame, as shown in FIG. 8, the variance of the luminance value is calculated for each small block obtained by further dividing the macroblock into four. When a block having a large variance is adjacent to a block having a small variance, mosquito noise is likely to occur if the quantization width is large. In other words, when a block with a flat texture exists next to a block with a complicated texture, mosquito noise is likely to appear.

Therefore, when a small block is adjacent to a block having a large variance of the luminance value, the quantization width of the small block is determined by the quantization width QP (j) generated as an encoding parameter as described above. Set to be smaller than Conversely, a quantization width larger than QP (j) is set for a block for which it is determined that mosquito noise is unlikely to appear, thereby preventing the generated code amount from increasing.

For example, suppose that there are four small blocks in the macroblock for the mth macroblock in the jth frame. At this time, as shown in FIG. 8, if there is a small block satisfying the combination of (variance of block k) ≧ MBVarThre and (variance of block adjacent to block k) <MBVarThre (4), this m-th block It is determined that the macroblock is a macroblock in which mosquito noise is likely to appear. For such a macroblock in which mosquito noise is likely to appear, the quantization width QP (j) _m is reduced as QP (j) _m = QP (j) -q1 (5).

On the other hand, for a macroblock for which it is not determined that mosquito noise is likely to occur, the quantization width is increased as follows: QP (j) _ (MB where noise is unlikely to occur) = QP (j) + q2 (6) This prevents an increase in the generated code amount.

However, in equation (4), MBVarThr
e is a threshold defined by the user. Q in equation (5)
1, q2 in equation (6) is a positive number, and QP (j) -q1
≧ (minimum value of quantization width), QP (j) + q2 ≦ (maximum value of quantization width). Here, the scenes classified as the above-described camera translation scene (b) and the camera zooming scene (d) are visually noticed to objects in the image because they are governed by camera movement. Are considered to be low, q1 and q2 are set small. In the still scene of (a) and the scene in which moving parts are concentrated in (c), q1 and q2 are set to be large because it is considered that the degree of visual attention to the object in the image is high.

Also, for a macroblock having an edge such as a character, by reducing the quantization width,
Make the character parts clear. Specifically, as shown in FIG. 9, an edge emphasis filter is applied to the luminance value data of the frame, and a pixel having a strong gradient of the gray value is examined for each macroblock. Then, the positions of such pixels having a strong gradient of the gray value are totaled, and a macroblock in which pixels having a large gradient are partially concentrated is determined to be a macroblock having an edge. The quantization width is reduced for the macroblock, and the quantization widths of the other macroblocks are increased according to equation (6).

[Regarding the Encoding Unit 10] The processing of the encoding unit 10 is the same as that of the conventional moving picture encoding apparatus of the MPEG system, but here, the encoding parameter is generated by the encoding parameter generation unit 31 as described above. First according to the coding parameters
Perform the third encoding. In this case, encoding is basically performed at a fixed frame rate and a fixed quantization width in each divided scene. That is, for the j th scene, frame rate F from the first frame of a scene (j_start) to the last frame (j _ end The)
R (j) and the quantization width QP (j).

However, in the case where the above-described processing of <in the case where the quantization width is changed for each macroblock> is performed in the encoding parameter generation unit 31, the setting is made for each macroblock even in the same scene or the same frame. [Regarding the Code Amount Determining Unit 33] for Encoding Using the Quantization Width The code amount determining unit 33 generates the generated code amount 13 of the encoded bit stream 111 output from the encoding unit 10 as described above.
3 and the target code amount 134, and if the difference between the generated code amount 133 and the target code amount 134 exceeds the threshold, the code amount determination unit 33 instructs the coding parameter correction unit 34 to correct the coding parameter. And the encoding unit 10 is controlled via a system control unit (not shown) so as to perform the second encoding after the encoding parameters are corrected. Generated code amount 1
If 33 is a value equal to or less than the target code amount 134 and the difference is equal to or less than the threshold value and the two are close to each other within a substantially reasonable range, the encoding is terminated, and the encoded bit stream stored in the buffer 21 is encoded. To the transmission path or storage medium.

[Regarding the Coding Parameter Correction Unit 34]
The coding parameter correction unit 34 corrects the coding parameter based on the determination result of the code amount determination unit 33. For example, when the generated code amount 134 is larger than the target code amount 134, the frame rate is reduced as a whole, the quantization width is increased, or the gradient applied to the quantization width is reduced. When the generated code amount 133 is smaller than the target code amount 134, the frame rate is increased as a whole, the quantization width is reduced, or the gradient applied to the quantization width is increased. Increasing the quantization width means that the quantization width of the central macroblock in the frame is reduced and the quantization width of the peripheral macroblock is increased in the frame.
This is to provide a difference in quantization width depending on the position of a macroblock in the same frame.

The modification of the encoding parameters is performed by adding weights w_refFR and w_refQP to the frame rate FR (j) and the quantization width QP (j) generated by the encoding parameter generation unit 31, respectively. Is realized by: The corrected frame rate FR '(j),
The quantization width QP '(j) is as follows. FR ′ (j) = FR (j) + w_refFR (7) QP ′ (j) = QP (j) + w_refQP (8) In the encoding unit 10, the frame rate FR ′ ( j), quantization width QP '
The second encoding is performed using (j).

If a large difference exceeding the threshold value still remains between the generated code amount 131 and the target code amount 134 even after the second encoding is completed, the weight w_ref is used.
Step S in FIG. 3 while modifying FR and w_refQP
14 to S16 are repeated, and finally the generated code amount 1
When the value of 33 becomes equal to or less than the target code amount of 134 and is almost within a reasonable range, the encoding is terminated, and the encoded output 200 is output.

Although the above embodiment has been described with respect to the case where two-pass encoding is performed, the present invention relates to ordinary moving image encoding in which a moving image signal of one moving image file is completed only once. It is also applicable to the device.

[0083]

As described above, according to the present invention, an input moving image signal is divided into scenes consisting of at least one frame that is temporally continuous, and a statistical feature amount is calculated for each scene. By generating an encoding parameter for each scene based on the statistical feature and encoding the input moving image signal, for example, in a scene with a large motion, the frame rate is set high to smooth the motion of the object, In such a scene, it is possible to set a small quantization width around an edge or a telop character portion, which is a portion that is easily noticed in the image, to clarify the image.

Therefore, according to the present invention, it is possible to obtain a united decoded image for each scene as compared with a conventional moving image encoding apparatus which does not consider the movement of an object in an image or the movement of a camera. Good image quality improvement effects can be obtained while maintaining the bit rate at a value specified according to the transmission rate of the transmission path, the capacity of the storage medium, and the like.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a moving picture coding apparatus according to an embodiment of the present invention.

FIG. 2 is a view showing an example of encoding according to the contents of a scene for explaining the outline of the embodiment;

FIG. 3 is an exemplary flowchart showing an encoding processing procedure according to the embodiment;

FIG. 4 is an exemplary flowchart showing a procedure of scene division in the embodiment;

FIG. 5 is an exemplary view for explaining an operation for determining a scene break for scene division in the embodiment;

FIG. 6 is an exemplary view for explaining a flash frame determination operation in the embodiment;

FIG. 7 is an exemplary view for explaining frame classification by a motion vector in the embodiment;

FIG. 8 is a view for explaining determination of a macroblock in which mosquito noise is likely to occur in the embodiment.

FIG. 9 is an exemplary view for explaining determination of a macroblock having an edge in the embodiment;

FIG. 10 is a diagram illustrating a relationship between a buffer amount, a frame rate, and an encoded frame for describing a problem of rate control in a conventional moving image encoding device.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 10 ... Encoding part 11 ... Frame memory 13 ... Discrete cosine transformer 14 ... Quantizer 15 ... Inverse quantizer 16 ... Inverse discrete cosine transformer 18 ... Video memory 19 ... Motion compensation predictor 20 ... Variable length encoder DESCRIPTION OF SYMBOLS 21 ... Buffer 31 ... Image feature quantity calculation part 32 ... Coding parameter generation part 33 ... Code quantity judgment part 34 ... Coding parameter correction part 100 ... Input moving image signal 130 ... Image feature quantity 131 ... Coding parameter (first 133: generated code amount 134: target code amount 135: code amount determination result 136: corrected coding parameter (second coding parameter)

 ────────────────────────────────────────────────── ─── Continuing from the front page (72) Inventor Yoshihiro Kikuchi 1 Tokoba R & D Center, Komukai Toshiba-cho, Kawasaki-shi, Kanagawa F-term (reference) 5C059 KK02 KK25 MA00 MA23 MA32 MB04 MC31 NN03 NN28 NN36 PP04 SS08 TA46 TA50 TA63 TB02 TC12 TC19 TC37 TC42 TD04 TD10 TD12 UA02

Claims (8)

[Claims] 1. An input moving image signal is divided into scenes composed of at least one frame that is temporally continuous, a statistical feature amount is calculated for each scene, and each of the scenes is calculated based on the statistical feature amount. A coding parameter is generated, and the input moving image signal is coded using the coding parameter. 2. A feature amount calculating step of dividing an input moving image signal into scenes composed of at least one frame that is temporally continuous, and calculating a statistical feature amount for each scene; An encoding parameter generating step of generating a first encoding parameter for each of the scenes based on the statistical feature amount obtained, and encoding the input video signal using the first encoding parameter. A first encoding step of generating a sequence; a determining step of determining whether the generated code amount of the code sequence generated by the first encoding step is more or less than a target code amount; An encoding parameter modification step of modifying a first encoding parameter to obtain a second encoding parameter based on the first encoding parameter, A second encoding step of encoding the dynamic video signal to generate a code sequence; and an output step of outputting the code sequence generated by the second encoding step as an encoded output. Video encoding method. 3. A feature amount calculating unit that divides an input moving image signal into scenes composed of at least one temporally continuous frame and calculates a statistical feature amount for each scene; Encoding parameter generating means for generating an encoding parameter for each of the scenes based on the statistical feature amount obtained, and encoding means for encoding the input video signal using the encoding parameter. A video encoding device characterized by the above-mentioned. 4. A feature amount calculating unit that divides an input moving image signal into scenes composed of at least one frame that is temporally continuous, and calculates a statistical feature amount for each scene; Coding parameter generating means for generating a first coding parameter for each of the scenes based on the statistical features obtained, and a second coding parameter obtained by correcting the first coding parameter or the first coding parameter. Encoding means for encoding the input video signal using an encoding parameter of the formula (1) to generate a code sequence; and wherein the encoding means encodes the input video signal according to the first encoding parameter. Determining means for determining whether the generated code amount of the generated code string is more or less than a target code amount; and correcting the second coding parameter based on a determination result of the determining means. Coding parameter correction means for obtaining the coding parameter of the following; and a coding sequence generated when the coding means codes the input video signal in accordance with the second coding parameter, as a coded output. A moving picture coding apparatus comprising output means. 5. The statistical feature amount, wherein the feature amount calculating means tabulates at least the magnitude and distribution of a motion vector present in each frame of the input moving image signal for each scene. The moving picture coding apparatus according to claim 3. 6. The moving picture coding apparatus according to claim 3, wherein said coding parameter generating means generates at least a frame rate and a quantization width as coding parameters. 7. The feature amount calculating means sums up at least the magnitude and distribution of a motion vector present in each frame of the input moving image signal for each scene to obtain the statistical feature amount. Classifying scenes according to the type of frame based on the motion of the camera and the motion of an object in the image used to obtain the input video signal from the size and the distribution of the input video signal; 3. The moving picture coding apparatus according to claim 2, wherein the means generates the coding parameter in consideration of the scene classification. 8. An encoding parameter generating unit comprising: a macroblock and an object having a difference in variance of luminance values between adjacent macroblocks of a macroblock in a frame encoded by the encoding unit which is equal to or greater than a predetermined value. 7. The moving picture coding apparatus according to claim 6, wherein a quantization width of a macroblock block in which the edge of the macroblock exists is made relatively smaller than that of other macroblocks.