JP2006303734A - Moving picture encoder and program thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a moving picture encoder for improving subjective image quality of a whole moving picture and to provide a program thereof. <P>SOLUTION: A moving picture encoder compresses a moving picture by using: an orthogonal conversion section 104 carrying out orthogonal conversion in the unit of blocks; a quantization section 105; a variable length coding section 106; a block distortion elimination filter 114 carrying out filtering in the unit of pictures; a means for selecting a filter offset value B affecting the filter processing depending on at least a size of the inputted moving picture prior to coding processing in order to control a filter ON / OFF threshold value; a means for determining a filter offset value A on the basis of at least the comparison between a variance of quantization values of whole coded pictures with a predetermined threshold value before the filter processing; and a means attaching the determined filter offset values A, B to coded data. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a moving image encoding apparatus and a program thereof that reduce block distortion that occurs when a moving image is orthogonally transformed and quantized.

  FIG. 7 shows ISO / IEC14496-10AVC (Non-Patent Document 1), which is a moving image coding technique for coding a moving image using a block unit orthogonal transform, quantization, variable length coding, and a filter for removing block distortion. ) Encoder block, intra prediction unit 1, motion compensation (MC) unit 2, motion vector detection (ME) unit 3, image memory (Frame Memory) 4, block distortion removing filter (Deblock Filter) 5, buffer The memory (DPB) 6, the transform unit 7, the quantization unit 8, the entropy coding unit 9, the coding unit 10, the inverse quantization unit 11, and the inverse transform unit 12 are configured. As the block distortion removal filter 5, a LoopFilter (loop filter) is mounted.

  FIG. 8 is a flowchart showing how to use the filter on / off threshold values α and β of Non-Patent Document 1. In Non-Patent Document 1, it is determined whether the filter is on or off before the block distortion removing filter 5 is applied. To determine whether the filter is on or off, first, a numerical value indicating the degree of block distortion at the block boundary, called Boundary Strength (BS), is obtained (step S601). Next, it is determined whether or not the obtained BS is 0 (step S602). When BS = 0, no filter is applied. When BS ≠ 0, filter on / off threshold values α and β are obtained (steps S603 and S604). Formulas for obtaining the filter on / off threshold values α and β are shown below.

IndexA = CLIP (0,51, QP + A)
IndexB = CLIP (0, 51, QP + B)
Index A and Index B are determined by the above formula, and filter on / off threshold values α and β corresponding to Index A and Index B are determined from the table of FIG. 9 (steps S603 and S604). CLIP (a, b, c) indicates that the value c is always determined between a and b. Filter on / off is determined using the determined filter on / off threshold values α and β (step S605). The filter on / off conditions are shown below.

  1. The absolute difference between p0 and q0 is smaller than α.

  2. The absolute difference between p1 and p0 is smaller than β.

  3. The absolute difference between q1 and q0 is smaller than β.

  When all the above conditions 1 to 3 are satisfied, the filter is turned on (step S607), and when the conditions are not satisfied, the filter is turned off.

  In the case of such a technique of Non-Patent Document 1, there is a tendency to use fixed values for the filter offset values A and B when obtaining the filter on / off threshold values α and β for turning on / off the filter. The same filtering process is performed for a moving image having a large size and hardly causing block distortion even for a moving image having a small image size and a large amount of block distortion. As a result, there is a problem in that a moving image having a large image size and hardly causing block distortion loses a sense of resolution due to excessive filtering, and a moving image having a small image size and a large amount of block distortion remains with block distortion.

  FIG. 10 is a block diagram of an MPEG-2 decoder having a block distortion removal filter described in Japanese Patent Laid-Open No. 2002-199403 (Patent Document 1). In the figure, 21 is a variable length decoder that decodes an MPEG-2 video code string that is a variable length code, 22 is an inverse quantizer that performs inverse quantization of the decoded variable length code, and 23 is an inverse quantum. 25 is a motion compensator that performs motion compensation on the output of the variable length decoder 21, and 24 is an output of the inverse DCT 23 and the motion compensator 25. An adder that performs addition with the output, 26 is an image memory that temporarily stores the decoded image data, and 27 detects the presence or absence of block distortion using the output of the image memory 26 and the output of the variable length decoder 21. The block distortion detector 28 is a block distortion remover 28 that removes block distortion using the output of the block distortion detector 27. In Patent Document 1, block distortion is detected and removed using a block distortion detector 27 and a block distortion remover 28.

  FIG. 11 is a block diagram showing an internal configuration of the block distortion detector 27 of Patent Document 1. The block distortion detector 27 includes a pixel value checker 27-1, a motion vector checker 27-2, and a block distortion determiner. 27-3. The block distortion detector 27 receives the decoded image data from the image memory 26 and the motion vector from the variable length decoder 21. In Patent Document 1, a pixel value inspector 27-1 calculates an absolute difference value of pixels near an arbitrary block boundary from the image memory 26, and determines whether or not the block boundary is distorted based on a predetermined threshold value. Further, the motion vector checker 27-2 corrects block distortion detection using the motion vector information used for encoding, and the block distortion determiner 27-3 determines the block distortion.

  The pixel value inspector 27-1 determines that the block distortion occurs when the following condition is satisfied for the pixel in FIG.

  4). The absolute value difference value between the pixel q0 and the pixel p0 is larger than the predetermined value Th1.

  5. The absolute value difference value between the pixel p0 and the pixel p1 and between the pixel p1 and the pixel p2 is smaller than the predetermined value Th2.

  Block distortion is detected when the above conditions 4 and 5 are satisfied.

In the technique of Patent Document 1 as well, since the predetermined threshold values Th1 and Th2 are used as the threshold values used for block distortion detection, the block is the same as in Non-Patent Document 1. There has been a problem that distortion cannot be detected accurately.
JP 2002-199403 A ISO / IEC14496-10AVC

  The present invention has been made in view of the above-mentioned problems of the prior art. A moving image having a large image size and a small block distortion retains a sense of resolution, and a moving image having a small image size and a large block distortion. Then, it aims at providing the moving image encoder which can reduce block distortion more, and its program.

  A moving image encoding apparatus according to a first aspect of the present invention includes a first filter offset value determination unit that determines a first filter offset value based on an image size of an input image signal, and blocks the input image signal. An orthogonal transform unit that performs orthogonal transform in units, a quantization unit that quantizes the orthogonally transformed signal, a variable length coding unit that encodes the quantized signal, and the quantized data as 1 A quantization determination unit that holds slices and obtains a dispersion value of quantization data of a block group included in one slice; a second filter offset value determination unit that determines a second filter offset value based on the dispersion value; A multiplexing unit for writing the first and second filter offset values into a slice header of the encoded image encoded data, and inverse quantization on the quantized data. A quantizing unit, an inverse transform unit that performs orthogonal inverse transform on the inversely quantized data, and addition data of the inversely transformed data and intra-screen prediction data or inter-screen prediction data, decoded image data A decoded image memory that is held as a filter, and the data input from the decoded image memory is subjected to filter on / off determination using the first and second filter offset values, and deblocking filter processing when the filter is on A deblocking filter processing unit for performing motion vector detection, a buffer memory for storing the data subjected to the deblocking filtering processing as reference image data, and motion for performing motion vector detection using image data to be encoded and the reference image data Motion compensation is performed using a vector detection unit, the image data to be encoded, and the reference image data It is obtained by a can compensator.

  A moving image encoding program according to a second aspect of the invention includes a first filter offset value determination process for determining a first filter offset value based on an image size of an input image signal, and the input image signal. Orthogonal transformation processing for orthogonal transformation in units of blocks, quantization processing for quantizing the orthogonally transformed signal, variable length coding processing for coding the quantized signal, and the quantized data A quantization determination process that holds one slice and obtains a variance value of a quantization value of a block group included in one slice, and a second filter offset value determination process that determines a second filter offset value based on the variance value A multiplexing process for writing the first and second filter offset values in the slice header of the encoded image encoded data, and the quantized data. An inverse quantization process, an inverse transform process on the inversely quantized data, an image decoding process to obtain decoded image data by adding the inversely transformed data and intra prediction data or inter prediction data; A deblocking filter process for determining whether the decoded image data is filtered on / off using the first and second filter offset values, and performing a deblocking filter process at the time of filter on determination; Buffer memory processing for holding block filtered data as reference image data, motion vector detection processing for performing motion vector detection using image data to be encoded and the reference image data, and encoding from now on Causing a computer to execute motion compensation processing for performing motion compensation using image data and the reference image data It is intended.

  According to the present invention, since the filter offset value suitable for each slice can be made timely variable by using at least the input image size and the variance value of the quantization value used for encoding, In comparison, a moving image with a large image size and a small block distortion can maintain a sense of resolution, and a moving image with a small image size and a large amount of block distortion can further reduce the block distortion. Improve image quality.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram of a moving picture encoding apparatus using the MPEG-4 AVC encoding technique according to one embodiment of the present invention, and FIG. 2 is a flowchart of a processing function program according to the present embodiment. In FIG. 1, 101 is an input unit, 102 is a filter offset value B determination unit, 103 is an intra prediction unit, 104 is a DCT processing conversion unit, 105 is a quantization unit, 106 is an encoding unit, and 107 is a filter offset value A. , B in the header of the encoded data, 108 is an output unit that outputs the image encoded data, 109 is a filter offset value A determining unit, 110 is a quantization determining unit, 111 is an inverse quantization unit, 112 is an inverse transform unit, 113 is a decoded image memory, 114 is a loop filter for removing block distortion (LoopFilter), 115 is a buffer memory (DPB), 116 is a motion compensation (MC) unit, and 117 is a motion vector detection (ME). Indicates the part.

  The operation of the video encoding apparatus having the above configuration will be described. The input unit 101 receives moving image data to be encoded, outputs image data to the intra prediction unit 103 or the MC unit 116 and the ME unit 117, and sets the image size of the input image to the filter offset value B determination unit 102. Output (step S201).

  The filter offset value B determination unit 102 determines filter offset value B candidates based on the size of the moving image input from the input unit 101, and outputs the result to the multiplexing unit 107 (step S202).

  FIG. 3 shows a method for determining the filter offset value B according to the present embodiment, and FIG. 4 shows a table of arithmetic expressions. The filter offset value B is a numerical value related to the absolute value difference values of p0 and p1, q0 and q1, p1 and p2, q1 and q2, p0 and p2, and q0 and q2 in FIG. It is a numerical value related to. Even if the same moving image has different image sizes, the information held by the pixels in the block is different. When the image size is large, the pixel information in one block is small, and when the image size is small, the information in one block is large. For this reason, the numerical value of the filter offset value B is reduced as the image size S increases as shown in the table of FIG. For example, if the image size of the input moving image is 720 × 480, the filter offset value B is determined to be 0 from the tables of FIGS. In this way, by controlling the filter offset value B according to the size of the image size S, the condition 2 (the absolute difference value between p1 and p0 is smaller than β) and the condition 3 (the absolute difference value between q1 and q0 is greater than β) A filter on / off threshold value β for turning on / off a small filter can be controlled. That is, in the case of a moving image having a large image size S, β is decreased to reduce the filter effect. In the case of a moving image with a small image size S, the effect of the filter is increased by increasing β. As a result, a sense of resolution can be maintained in the case of a moving image with a large image size S, and block distortion detection can be facilitated and a block distortion removal filter can be used effectively for a moving image with a small image size. can do.

  The intra prediction unit 103 performs intra prediction at the pixel level and outputs the result to the subtracter. This subtracter performs subtraction between the input image and the output image of the intra prediction unit 103 or MC unit 116, and outputs a difference signal to the conversion unit 104 (step S204).

  The conversion unit 104 receives the difference signal from the subtracter and performs DCT orthogonal conversion in units of 4 × 4 or 8 × 8. The converted data is output to the quantization unit 105 (step S207).

  The quantizing unit 105 inputs a signal subjected to DCT orthogonal transformation, quantizes the coefficient component at an appropriate quantization step, and outputs quantized data to the encoding unit 106 and the inverse quantization unit 111. In addition, the quantization value is output to the quantization determination unit 110. The quantization determining unit 110 stores the quantized value in units of slices (Steps S208 and S209).

  The encoding unit 106 encodes the input data. The encoded data is output to the multiplexing unit 107 (step S210).

  The multiplexing unit 107 writes the filter offset values A and B from the filter offset value A determination unit 109 and the filter offset value B determination unit 102 in the header of the encoded data in units of slices from the encoding unit 106. That is, the process of rewriting the values of A and B so far to the input values is performed. The multiplexed encoded data is output to the output unit 108. Further, the filter offset values A and B are output to the Loop Filter 114 (step S215).

  The output unit 108 outputs the encoded data input from the multiplexing unit 107 to the outside.

  The filter offset value A determination unit 109 determines the filter offset value A based on the quantization determination data input from the quantization determination unit 110, and outputs the result to the multiplexing unit 107 (steps S213 and S214).

  The filter offset value A will be described. This filter offset value A is a numerical value related to p0 and q0, that is, the absolute value difference of pixels between blocks, as shown in FIG. From condition 1 (the absolute value of the difference between p0 and q0 is smaller than α), if the filter offset value A is increased, the α also increases, making it easier to filter. On the other hand, if the filter offset value A is reduced, the α is also reduced, making it difficult to apply a filter. That is, the filter offset value A is closely related to the on / off state of the filter, and the degree of filter application can be controlled by making the filter offset value A variable. In the present embodiment, quantization is performed in units of MB (macroblock) using a quantizer, and distortion occurs between adjacent MBs due to a difference in quantization values. Therefore, it is possible to control the filter processing by changing the filter offset value A according to the quantization value. Therefore, in the present embodiment, the filter offset value A is changed in units of slices, and the filter offset value A is used as an index that can be varied by the dispersion value of the quantization value of the MB group included in one slice.

  The filter offset value A determination unit 109 of the present embodiment will be described using the tables in FIGS. The quantization determination unit 110 holds the quantization value of the MB group in units of one slice, and obtains the dispersion value F of the quantization value of the MB group for each rice. Then, it is determined whether the obtained dispersion value F is larger or smaller than predetermined threshold values Th1 and Th2 (steps S301 and S303). If Th1 = 10 and Th2 = 50, if the obtained dispersion value F is smaller than 10, A = -1 (step S302) from the table of FIG. 6, and if F is 30, A from the table of FIG. = 0 (step S304), and if F is 60, A = 1 (step S305). As described above, by making the filter offset value A variable in units of slices by the dispersion value F of the quantized value of MB, the filter offset value A corresponding to each slice can be used, and a lot of block distortion occurs. In the case of a slice that has a small block distortion, it is easy to apply a block distortion filter by increasing the filter offset value A, and the block distortion is reduced. Therefore, it is difficult to apply a filter, and a sense of resolution can be maintained.

  In the quantization determination unit 110, the quantization value used for encoding is input from the quantization unit 105, the quantization value of MB for each slice unit is stored, and the variance value F is calculated using the stored quantization value. Ask. The obtained variance value F is output to the filter offset value A determination unit 109 (step S213).

  The inverse quantization unit 111 performs an inverse quantization process on the quantized data input from the quantization unit 105 and outputs the result to the inverse transform unit 112 (step S211).

  The inverse transform unit 112 performs an inverse transform process on the data input from the inverse quantization unit 111 and outputs the result to the adder (step S212).

  The decoded image memory 113 holds decoded image data that is the addition result of the data input from the inverse transform unit 112 and the data input from the intra prediction unit 103 or the MC unit 116, and outputs the data to the Loop Filter 114. .

  The Loop Filter 114 performs a filtering process on the data input from the decoded image memory 113, and outputs the filtered data to the buffer memory (DPB) 115 (Step S216).

  The DPB 115 holds the decoded image data when the data input from the Loop Filter 114 is used as a reference image, and outputs the data to the ME unit 117 as necessary.

  The ME unit 117 performs motion vector detection using image data to be encoded from now on and image data to be referred to at the time of encoding. As the image data to be referred to, the current image or the decoded image may be used (step S205).

  The MC unit 116 performs motion compensation using the image data to be encoded and the decoded image data stored in the DPB 115, and outputs the result to the subtracter (step S206).

  According to the present embodiment, it is possible to make the filter offset value suitable for each slice timely variable by using at least the input image size and the variance value of the quantization value used for encoding. For example, a moving image with a large image size and a small block distortion can maintain a sense of resolution, and a moving image with a small image size and a large amount of block distortion can further reduce the block distortion. Can be improved.

  INDUSTRIAL APPLICABILITY The present invention is effective for a moving image encoding apparatus used in a recording device or a mobile device that requires high image quality.

1 is a block diagram of a moving picture encoding apparatus according to an embodiment of the present invention. The flowchart of the moving image encoding process by the said embodiment. The flowchart of the process which calculates | requires the filter offset value B. A table for determining a filter offset value B. The flowchart of the process which calculates | requires the filter offset value A. A table for determining a filter offset value A. The block diagram of the encoder of a prior art example. The flowchart of the LoopFilter pre-process in the said prior art example. A table of filter on / off threshold values α and β in the conventional example. FIG. 10 is a block diagram of another conventional decoder. The block diagram of the block distortion detector in the said prior art example. The figure which shows the position of a block boundary and a pixel.

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 Input part 102 Filter offset value B determination part 103 Intra prediction part 104 Conversion part 105 Quantization part 106 Encoding part 107 Multiplexing part 108 Output part 109 Filter offset value A determination part 110 Quantization determination part 111 Inverse quantization part 112 Inverse transformation unit 113 Decoded image memory 114 LoopFilter
115 DPB
116 MC part 117 ME part

Claims (2)

A first filter offset value determining unit that determines a first filter offset value based on the image size of the input image signal;
An orthogonal transform unit that orthogonally transforms the input image signal in units of blocks;
A quantization unit for quantizing the orthogonally transformed signal;
A variable length encoding unit for encoding the quantized signal;
A quantization determination unit that holds the quantized data for one slice and obtains a dispersion value of quantized data of a block group included in one slice;
A second filter offset value determination unit for determining a second filter offset value based on the variance value;
A multiplexing unit for writing the first and second filter offset values in a slice header of the encoded image encoded data;
An inverse quantization unit that performs inverse quantization on the quantized data;
An inverse transform unit that performs orthogonal inverse transform on the inversely quantized data;
Decoded image memory that holds addition data of the inversely transformed data and intra prediction data or inter prediction data as decoded image data;
A deblocking filter processing unit for determining whether to turn on / off the data input from the decoded image memory using the first and second filter offset values and performing a deblocking filter process when the filter is on; ,
A buffer memory for holding the deblock filtered data as reference image data;
A motion vector detection unit for performing motion vector detection using the image data to be encoded and the reference image data;
A video encoding apparatus comprising: a motion compensation unit that performs motion compensation using the image data to be encoded and the reference image data. A first filter offset value determination process for determining a first filter offset value based on the image size of the input image signal;
Orthogonal transformation processing for orthogonally transforming the input image signal in units of blocks;
A quantization process for quantizing the orthogonally transformed signal;
A variable length encoding process for encoding the quantized signal;
A quantization determination process for holding the quantized data for one slice and obtaining a dispersion value of a quantization value of a block group included in one slice;
A second filter offset value determination process for determining a second filter offset value based on the variance value;
A multiplexing process for writing the first and second filter offset values in a slice header of the encoded image encoded data;
An inverse quantization process for the quantized data;
An inverse transform process for the inversely quantized data;
An image decoding process for adding the inversely transformed data and intra prediction data or inter prediction data to obtain decoded image data;
A deblocking filter process for determining the filter on / off using the first and second filter offset values for the decoded image data and performing a deblocking filter process at the time of the filter on determination;
Buffer memory processing for holding the deblocked filter processed data as reference image data;
A motion vector detection process for performing motion vector detection using the image data to be encoded and the reference image data;
A moving image encoding program for causing a computer to execute a motion compensation process for performing motion compensation using the image data to be encoded and the reference image data.

Images