JP5272940B2 - Image encoding device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image coder applying prediction coding to an input image that adaptively controls a coding parameter in the unit of pictures so as to attain high efficiency coding. <P>SOLUTION: The image coder applying prediction coding to an input image is provided with: an acquisition means that acquires at least any one statistic information among first statistic information that denotes the property of an input image read from a frame memory 1 such as an input image information statistic acquisition device 14, second statistic information based on correlation information in the process of prediction coding with a motion information statistics acquisition device 15 or the like and third statistic information as a result of coding or in the process of coding with a coding information statistic acquisition device 16 or the like; and a control means such as a coding control section 12 that uses a scene discrimination device 17 or the like to discriminate a scene on the basis of the statistic information acquired by the acquisition means, thereby applying adaptive control to a coding parameter in a coder 6 in the unit of frames or fields. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to an image coding apparatus that performs high-efficiency coding by switching coding parameters in units of pictures based on characteristics of an input image and statistical information in the coding process.

  MPEG-2 (Moving Picture Experts Group-2) has been internationally standardized as one of the moving image encoding systems, and is applied to the fields of DVD (Digital Video Disc / Digital Versatile Disc) video content and digital broadcasting. ing. Prior to the international standardization of MPEG-2, MPEG-1 was standardized as an encoding method targeting recording media such as CD-ROM (Video CD) and line use up to about 1.5 Mbps.

  The above-mentioned MPEG-2 incorporates countermeasures for improving image quality as compared with the previous MPEG-1. As a typical example, field-aware motion prediction can be performed as a countermeasure for higher resolution image encoding and interlaced image encoding. Encoding is also becoming more versatile. These functions are multiplexed into a stream (data string) as an additional function (extension). Basically, MPEG-2 has upward compatibility with MPEG-1, and one of the methods that can be specified in each additional function layer of MPEG-2 is the same as MPEG-1.

An example of one picture_coding_extension of the additional function layer in MPEG-2 is shown in FIG. 9, and a part of it will be described.
f_code: range of motion vector that can be expressed,
intra_dc_precise: accuracy of DCT DC component of intra-frame coding block,
picture_structure: coding structure,
top_field_first: input order of fields in a frame,
frame_pred_frame_dct: limit flag of frame prediction frame DCT,
q_scale_type: quantization scale type,
intra_vlc_format: DCT coefficient variable length code table selection,
alternate_scan: input order of DCT coefficients to variable-length coding unit,
progressive_frame: selection of whether the input signal is progressive,
Is shown.

  If the quantization coefficient is 5 bits and the change width of the quantization coefficient is 1 according to the MPEG standard, the quantization level is an integer value in the range of 1 to 31. Therefore, for example, the change width is set to 1 of the standard value, 0.25 of 1/4 of the standard value, and 4 times 4 of the standard value, and the quantization level is set to 0.25 to 7.75, 1 to 31. And means for enabling selection within a range from 4 to 124 have been proposed (see, for example, Patent Document 1).

JP-A-7-131789

  The parameters that can be selected in the extension layer added to the above-mentioned MPEG-2 basically improve the efficiency in moving picture coding, which was not good in MPEG-1. It is intended. However, depending on the nature of the input image and the coding rate, there are cases where more efficient coding can be achieved by adaptively using coding parameters in MPEG-1. In the conventional image encoding apparatus, such an additional function layer parameter is set to a preset fixed value to perform predictive encoding of an input image. Also, the image encoding means proposed by the above-mentioned Patent Document 1 performs quantization by selecting the change width of the quantization coefficient depending on the number of fixed bits. Optimum encoding is performed according to the nature of the input image. It was not enough to do.

  An object of the present invention is to provide an encoding means that can change an encoding parameter in accordance with the nature of an input image on a picture-by-picture basis and can produce a high-quality reproduced image even at the same encoding rate.

  An image encoding apparatus according to the present invention is an image encoding apparatus having an encoding unit that performs predictive encoding of an input image. In the process of predictive encoding, first statistical information indicating the nature of an input image is provided. Acquisition means for acquiring at least one statistical information of the second statistical information based on the correlation information and the third statistical information in the encoding result or the encoding process based on the encoding parameter And coding control means for performing scene determination based on statistical information obtained by the obtaining means and adaptively controlling coding parameters in the coding means in units of frames or fields, the obtaining means includes: The third statistical information has a configuration for obtaining an average value of effective coefficients after quantization per macroblock in a picture, and the encoding control means compares the average value of effective coefficients with a preset coefficient. You And it has a control structure to perform variable length coding by switching in the variable length code table in the encoding means based on the scene determination result.

  Further, the acquisition means has a configuration for acquiring, as the second statistical information, an average value of a horizontal component of a motion vector, a variance value of a horizontal component, an average value of a vertical component, and a variance of a vertical component, and the encoding means Can comprise a control configuration for switching the input scan order for variable length encoding in the encoding means based on a scene determination result comparing the second statistical information with a preset coefficient.

  The acquisition means has a configuration for acquiring an average quantized value as the third statistical information, and the encoding control means obtains a scene determination result for comparing the average quantized value with a preset coefficient. Based on this, it is possible to provide a control configuration for performing quantization by switching the quantization table in the encoding means.

  The present invention provides first statistical information such as an activity indicating the nature of an input image, second statistical information such as a motion vector of correlation information in the process of predictive coding, and a code based on a coding parameter. In the encoding means on a frame or field basis, using at least one or a plurality of statistical information of the third statistical information such as the amount of encoded information of the encoding result and the quantization average value of the encoding process There is an advantage that optimal encoding can be performed on an input image without increasing the amount of encoded information by controlling the application of encoding parameters.

It is explanatory drawing of Example 1 of this invention. It is explanatory drawing of a motion search. It is a flowchart of the determination process of embodiment of this invention. It is explanatory drawing of a variable-length code table. It is explanatory drawing of a variable-length code table. It is selection explanatory drawing of a variable-length code table. It is explanatory drawing of an input scan. It is explanatory drawing of a quantization table. It is explanatory drawing of an example of an additional function.

  Referring to FIG. 1, the image coding apparatus of the present invention is an image coding apparatus having coding means for performing predictive coding of an input image, and includes first statistical information indicating the nature of the input image. And / or second statistical information based on the correlation information in the predictive encoding process and encoding result based on the encoding parameter or third statistical information in the encoding process Acquisition means for acquiring one piece of statistical information, for example, acquisition means including an input image information statistical acquisition unit 14, a motion information statistical acquisition unit 15, an encoded information statistical acquisition unit 16 and the like, and statistical information acquired by the acquisition unit Coding means for adaptively controlling the coding parameters in the coding means for each frame or field by performing scene determination based on the scene, and the obtaining means includes one macro in the picture as third statistical information. The coding means has a configuration for obtaining an average value of effective coefficients after quantization per block, and the encoding control means is based on a scene determination result for comparing the average value of effective coefficients with a preset coefficient. In this case, the variable length code table is switched to perform variable length coding.

  FIG. 1 is an explanatory diagram of Embodiment 1 of the present invention, in which 1 is a frame memory, 2 is an original picture MB (macroblock) reading unit, 3 is a reference block reading unit, 4 is a motion vector searcher, and 5 is a prediction determination unit. , 6 is an encoder, 7 is a local decoder, 8 and 9 are switching units, 10 is an adder, 11 is a subtractor, 12 is an encoding control unit, 13 is a header information generation unit, and 14 is input image information. A statistical acquisition unit, 15 is a motion information statistical acquisition unit, 16 is an encoded information statistical acquisition unit, and 17 is a scene determination unit.

  The frame memory 1 includes an area for accumulating input image information and an area for accumulating reference image information. The frame memory 1 is read from the input image area in units of macroblocks by the original image MB reading unit 2 and is within the search range of the reference image area. The data is read by the reference block reading unit 3 in units of macroblocks, and a motion vector is obtained by the motion vector search unit 4 and input to the prediction determination unit 5. The encoder 6 includes functions of orthogonal transform by DCT (Discrete Cosine Transform), quantization, and variable length coding. Note that the macroblock (MB) size in the MPEG-2 system is 16 × 16 pixels, and the block on which DCT is performed is 8 × 8 pixels obtained by dividing the macroblock into four.

  The switching units 8 and 9 switch to the adder 10 and subtractor 11 side when encoding within the frame and between frames, and switch between the adder 10 and subtractor 11 side when encoding within the field and between fields. Switching to the opposite side. The local decoder 7 includes functions of inverse quantization and inverse DCT, performs decoding using the quantized output of the encoder 6 before variable encoding, and reconstructs a reference image. Are stored in the reference image information area of the frame memory 1.

The DCT in the encoder 6 is a two-dimensional DCT. As described above, in the MPEG-2, the DCT is performed on a block of 8 × 8 pixels. When this block is f (x, y) and the coefficient of the DCT result is F (u, v), the following equation (1) is obtained.

By this DCT calculation, the block unit image information is converted into frequency components, and the effective components are concentrated on the low frequency component side, thereby reducing the coding information. The block unit DCT coefficients are set to Rec [x]. If the quantization scale is Qs, the quantization matrix value is Qm [x], the quantization result is Level [x], and the value of the processing process is Level '[x], the direct current of the quantization result in the intra coding An arithmetic process for obtaining a component (Intra DC), an alternating current component (Intra AC), and a quantization result (Non Intra) in non-intra coding can be expressed by the following equation (2). In addition, as for p and q in a formula, p = 3 and q = 4 are generally used. “//” indicates an operation for rounding off the fraction of the division result, and “/” indicates an operation for rounding down the fraction of the division result.

In addition, it is necessary to obtain inter-frame difference information in order to perform inter-frame encoding. For this purpose, the encoded data is decoded by the local quantizer 7 through inverse quantization and inverse DCT processing, and the image data is decoded. Reconstruction is performed and the image is stored in the frame memory 1 as a reference image. As described above, the local decoder 7 includes processing functions of inverse quantization and inverse DCT, and the inverse quantization can be performed by the process shown in the equation (3). It can be performed by the process shown in the equation (4).

  When the input image is actually encoded, the first picture (frame or field) has no picture to be referred to, so intra-picture encoding is performed, and inter-picture encoding is performed from the next picture. Become. Note that intra-picture encoding is generally performed at predetermined intervals in the sense of periodic refresh. In the inter-picture coding, the motion vector search unit 4 performs motion prediction. For example, as shown in FIG. 2, the macro block 22 of the original image 21 and the search range 24 of the reference image 23 are included. A motion vector is obtained by searching for a position where the accumulated value of the absolute difference value is minimum for each pixel with respect to the macroblock, and multiplexed on the encoded information.

  Coding means is constituted by the functional parts denoted by reference numerals 1 to 11 in FIG. 1 described above, and coding control means is constituted by the coding control unit 12 for controlling the coding parameters in the coder 6. Then, header information including encoding parameters is generated by the header information generation unit 13, and the header information is added to the encoded data in units of pictures from the encoder 6 and transmitted.

The input image information statistic acquisition unit 14 is means for acquiring first statistical information indicating the nature of the input image. For example, the statistic about the luminance signal is used as the feature information of the encoding target picture stored in the frame memory 1. Ask for information (activity). In this case, the luminance value of each pixel in the input frame is accumulated and divided by the accumulated number of pixels, whereby the frame luminance average can be obtained and used as the first statistical information. That is, when the pixel set in the frame is U, the luminance is Pixel_i, the number of pixels is Num_i, the frame luminance average is AveY, and the frame luminance variance is VarY, the following equations (5) and (6) are expressed. In the equation (6), A represents the frame luminance average AveY.

  The motion information statistic acquisition unit 15 is means for acquiring second statistical information indicating correlation information between frames or fields. For example, the motion information statistic acquisition unit 15 accumulates motion vectors obtained by macroblocks in the motion vector searcher 4. By calculating and dividing by the number of macroblocks, an average value of motion vectors can be obtained. Alternatively, the variance of the motion vector can be obtained by obtaining the square sum of the difference between the motion vector and its average value and dividing the result by the number of macroblocks.

なお、（８）式の中のＡＨは、水平成分平均ＡｖｅＨＶを示し、又（１０）式の中のＡＶは、垂直成分平均ＡｖｅＶＶを示す。 That is, the set of macroblocks in the frame is V, the horizontal and vertical components of each motion vector are VecH_i and VecV_i, the horizontal component average is AveHV, the horizontal component variance is VerHV, the vertical component average value is AveVV, and the vertical component variance is Assuming that VerVV is expressed by the equations (7) to (10).

In the equation (8), AH represents the horizontal component average AveHV, and AV in the equation (10) represents the vertical component average AveVV.

The encoded information statistic acquisition unit 16 is means for acquiring third statistical information in the encoding process, for example, accumulates information obtained as a result of encoding each macroblock, and generates a generated information amount or quantization. Find the average of the values. In this case, the set of macroblocks in the frame is V, the generated information amount of each macroblock is Bit_i, the generated information amount of pictures is SumB, the quantization scale value of each macroblock is Qs_i, and the average quantized value is AveQ. Then, the generated information amount SumB and the average quantization value AveQ of the picture are expressed by Expressions (11) and (12).

Further, assuming that the effective coefficient of each macroblock after quantization is Coef_i and the average value of effective coefficients per macroblock in the picture is AveC, this effective coefficient average value AveC is expressed by Expression (13).

  The scene determination unit 17 performs scene determination based on at least one of the first, second, and third statistical information, and an input image as a means for acquiring the first statistical information. Frame luminance average AveY and variance VarY from the information statistics acquisition unit 14, and horizontal component average AveHV, horizontal component variance VerHV, vertical component average AveVV from the motion information statistics acquisition unit 15 as means for acquiring second statistical information, One or more of the vertical component variance AerVV, the picture generation information amount SumB, the average quantization value AveQ, the effective coefficient average value AveC, etc. from the encoded information statistics acquisition unit 16 as means for acquiring the third statistical information. For example, a determination is made for a scene with intense motion or a scene with flat brightness, and the encoding control unit 12 adaptively controls the encoding parameter. Te is intended to perform the encoding of the in the input image to the encoder 6.

  FIG. 3 is a flowchart of the determination process according to the embodiment of the present invention. The input image is read from the frame memory 1 (a1), the header information generation unit 13 generates a header for each picture (a2), and the motion The vector searcher 4 performs motion search (a3), and the encoder 6 performs MB (macroblock) encoding (a4) to determine whether or not the picture End, that is, one picture is finished. (A5) If not completed, the process proceeds to step (a3). If completed, necessary information is acquired (a10).

  In addition, the input image information statistics acquisition unit 14 acquires the first statistical information (a6), and the motion information statistics acquisition unit 15 calculates the statistics based on the motion vector search result as the second statistical information. The information is acquired (a7), and the encoded result by the encoder 6 or the third statistical information in the encoding process is acquired by the encoded information statistics acquirer 16 (a8), and the average value is obtained. Is obtained, an averaging process is performed (a9). The average processing result is acquired as necessary information acquisition (a10) in the scene determination unit 17, and it is determined whether or not a predetermined condition is satisfied (a11), and the encoding parameter 1 (a12) is selected based on the determination result. Alternatively, the encoding parameter 2 (a13) is selected, and it is determined whether or not it is the encoding end (a14). If not completed, the process proceeds to step (a1). In the determination step (a11), it is also possible to control selection switching of a larger number of encoding parameters in accordance with a plurality of types of determination conditions.

  For example, when adaptive control of intra_vlc_format (selection of DCT coefficient variable length code table) is performed as a parameter of the additional function (picture_coding_extension) shown in FIG. Selection can be made between 0 and table = 1 in FIG. 4 and 5 show a part of a variable length code table including a variable length code, a run, and a level, and s in the final bit indicates whether the level is positive or negative. In the sign, 0 indicates positive and 1 indicates negative. 1s indicates the first DCT coefficient of the block, and 11s indicates the next DCT coefficient.

  In addition, when table = 1 in FIG. 5 is equal to table = 0 in FIG. 4, it is possible to assign a variable length code by equalizing a bit shorter to some extent. Therefore, there are more effective coefficients in the block. If it exists, it is more effective to select table = 1 in FIG. 5 and perform variable length coding. In practice, the number of effective coefficients decreases when the average quantization value increases, and the number of effective coefficients decreases when the activity is small on a flat screen. Even when the average quantization value is large, the number of effective coefficients is small when the activity is small, and conversely, when the activity is large, the number of effective coefficients is large.

  FIG. 6 is an explanatory diagram of selection of the variable-length code table. When intra_vlc_format = 0, table = 0 for intra blocks and non-intra blocks, table = 1 for intra blocks, and table = 1 for intra blocks when intra_vlc_format = 1. Table = 0 is selected and variable length coding is performed.

Using the average quantization value AveQ obtained by the above equation (12) and VarY as the activity of the input image obtained by the equation (6),
AveQ> VarY * α ₁ + β ₁ (14)
If this condition is satisfied, the number of effective coefficients is small, so that table = 0 is selected, and if not, table-1 is adaptively controlled to select table = 1. Α ₁ and β ₁ indicate weighting coefficients.

For further simplification, the average value AveC of the effective coefficients obtained from the equation (13) is used.
AveC <α ₂ (15)
If this condition is satisfied, table = 0 is selected, and if not satisfied, variable-length encoding is performed by adaptively switching control so that table = 1 is selected. be able to. Α2 represents a coefficient. That is, the fact that the average value AveC of the effective coefficient per macroblock is smaller than the preset coefficient α2 corresponds to the case where the activity of the input image is small, and variable length coding is performed by switching to select table = 0. The encoding efficiency is better when this is done.

  Further, as a parameter of alternate_scan (input order of variable length coding of DCT coefficients), when alternate_scan = 0, the DCT coefficient scanning order is a zigzag scan (scan type 0) shown in FIG. If alternate_scan = 1, the alternate scan (scan type 1) shown in FIG. 7B can be performed. In this case, compared with the scan type 0 of (A), the scan type 1 of (B) preferentially encodes the coefficient of the vertical component in the frequency component. As a factor for generating a large number of effective coefficients in such a vertical component, for example, when an odd / even field image in encoding of an interlaced image is greatly different, that is, there is a movement such as panning or tilt. .

  MPEG-2 can perform motion prediction between fields in consideration of fields, and can perform field DCT that cuts out every other line when cutting out a block from a macroblock as an input to DCT. As shown in FIG. 2, the motion vector search range is generally set to a predetermined range 24 centered on the 0 vector. Therefore, when performing field prediction, the motion vector search range moves so much that the search range is insufficient. In such a case, the frequency vertical component direction becomes larger for the distribution of the effective coefficient.

Therefore, using the horizontal component average value AveHV in the equation (7) and the horizontal component variance VerHV in the equation (8),
AveHV> α ₃ (16)
VerHV <β ₃ (17)
It is determined whether or not the above condition is satisfied. In the case of large movements that are aligned to some extent in the horizontal direction so that this condition is satisfied, the vertical component increases, and therefore, adaptive scan = 1 is selected, and otherwise, adaptive_scan = 0 is selected. To perform variable length coding. Note that α ₃ and β ₃ are constants, and the motion search range is generally different depending on the device, and can be set in advance based on the motion search range.

  Further, as the third statistical information acquired by the encoded information statistical acquisition unit 16, the quantization scale type q_scale_type can be selected using the average quantization value AveQ. That is, in the quantization table shown in FIG. 8, when q_scale_type = 0, the change in the quantization scale value is linear, and the quantization scale value (punizer_scale_code) changes from 2 to 62. On the other hand, in the case of q_scale_type = 1, in order to cover a wide range of quantization scale values, non-linear changes occur, and the finer quantization part is finer and the coarser quantization part is more coarsely quantized. Thus, it changes from 1 to 112.

If the quantization scale value does not become extremely small or extremely large, there is no significant difference in selecting any of the quantization tables, but the average is used when the coding rate is high. When the quantization value is large, control is performed so as to select q_scale_type = 1. For example, for the average quantization value AveQ according to equation (12),
AveQ <α ₄ (18)
AveQ> β ₄ (19)
If this condition is satisfied, that is, if the average quantization value is extremely small or extremely large, the quantization scale value becomes a non-linear change q_scale_type. = 1 is selected, and otherwise, adaptive control is performed so that q_scale_type = 0 is selected.

  The present invention is not limited only to the above-described embodiments, and various additions and modifications can be made. Adaptive encoding parameters can be changed by switching the quantization table or variable-length code table. In addition to switching control, it is also possible to perform switching control of other encoding parameters, and obtain statistical information of a picture to be encoded before encoding the picture, and perform encoding by feedforward control. It is also possible to do this. When applying to storage media, provisional encoding of pictures is performed using multiple types of encoding parameters, and final determination is made based on the provisional encoding results, and header information including optimal encoding parameters is encoded. It can also be added to data and stored.

DESCRIPTION OF SYMBOLS 1 Frame memory 2 Original image MB reading part 3 Reference block reading part 4 Motion vector searcher 5 Predictive decision unit 6 Encoder 7 Local decoder 8, 9 Switching part 10 Adder 11 Subtractor 12 Encoding control part 13 Header information Generating unit 14 Input image information statistical acquisition unit 15 Motion information statistical acquisition unit 16 Encoding information statistical acquisition unit 17 Scene determination unit

Claims (2)

In an image coding apparatus having coding means for performing predictive coding of an input moving image,
First statistical information indicating the nature of the input video, second statistical information based on correlation information in the process of predictive encoding, and an encoding result or encoding based on an encoding parameter An acquisition means for acquiring at least one statistical information of the third statistical information in the process;
Coding determination means for performing scene determination based on statistical information acquired by the acquisition means, and adaptively controlling encoding parameters in the encoding means in units of frames or fields;
The acquisition means acquires, as the second statistical information, a horizontal component average value, a horizontal component variance value, a vertical component average value, and a vertical component variance value of the motion vector, and the third statistical information. It has a structure and to obtain the average value of the effective coefficient after quantization of one macro block per the picture as,
The encoding control means controls the variable length encoding by switching the variable length code table in the encoding means based on a scene determination result comparing the average value of the effective coefficients with a preset coefficient. configuration and that a control arrangement for switching the input scan order for the in variable-length coding on the coding unit based on the scene determination result of comparing the coefficient set in advance as the second statistical information An image encoding device. The acquisition unit has a configuration for acquiring an average quantized value as the third statistical information, and the encoding control unit is based on a scene determination result for comparing the average quantized value with a preset coefficient. 2. The image encoding apparatus according to claim 1, further comprising a control configuration for performing quantization by switching a quantization table in said encoding means.

Images