JP5375590B2 - Quantization control circuit, quantization parameter changing method, and encoding apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce quantization noise. <P>SOLUTION: A quantization control circuit 13 calculates a non-linear detection value corresponding to a difference between values of adjacent pixels included in a macro block for every macro block into which a specified sized picture is divided. According to the non-linear detection value, the quantization control circuit 13 changes quantization parameters for controlling the quantization steps of a quantization part 12b which quantizes macro blocks. <P>COPYRIGHT: (C)2011,JPO&amp;INPIT

Description

The present invention relates to a quantization control circuit, a quantization parameter changing method, and an encoding device .

  Conventionally, discrete cosine transform (DCT) is used for encoding an image signal. Such an apparatus performs a DCT operation on the input image signal for each block, quantizes the value of the operation result, and encodes the quantized value.

  In the image compression method using the above DCT operation, in a portion where a sharp outline such as a black line is expressed on a background with little color change, for example, a fine haze like a large group of mosquitoes is visually observed. Noise, so-called Mosquito Noise, is generated. This is due to the fact that the information of the high frequency component is reduced by the DCT calculation.

  Therefore, a method for removing mosquito noise using a filter when decoding image data has been proposed in order to suppress deterioration in quality of the displayed image (see, for example, Patent Document 1).

JP 2000-295615 A

  However, in the method of removing noise at the time of decoding, the quality of the display image is reduced in an apparatus that does not use this method. For this reason, it is required to suppress degradation in image quality regardless of the apparatus.

According to one aspect of the present invention, a predetermined number of each sub-block obtained by dividing a block having pixel data, the sub-blocks included in the three adjacent consecutively in a predetermined direction of the pixel adjacent the two A linear prediction value is calculated in accordance with a difference between pixel values of pixels, and nonlinearity based on a difference between a pixel value of a pixel different from the adjacent two pixels of the three consecutive pixels and the linear prediction value A quantization parameter for controlling a quantization step when the sub-block is quantized is changed according to the detected value.

  According to one aspect of the present invention, quantization noise is reduced.

It is a block diagram of the encoding apparatus of one Embodiment. It is explanatory drawing of a macroblock. It is a block diagram of a quantization control circuit. (A) (b) is explanatory drawing of a nonlinear detection process. It is explanatory drawing of a non-linear detection process. It is explanatory drawing of a non-linear detection process. It is a flowchart of an encoding process. (A) (b) is explanatory drawing of a nonlinear detection process.

Hereinafter, an embodiment will be described with reference to FIGS.
As shown in FIG. 1, the encoding device includes an encoding circuit 12 that encodes image data stored in an image memory 11.

  The image memory 11 is a memory having a capacity for storing image data (pixel values) of a plurality of screens. The image memory 11 stores image data corresponding to the image signal Si. The image data stored in the image memory 11 is read in units of macro blocks.

  The encoding circuit 12 includes an orthogonal transform unit 12a, a quantization unit 12b, and an encoding unit 12c. The orthogonal transform unit 12a performs discrete cosine transform (DCT: Discrete Cosine Transform) on the macroblock unit image data read from the image memory 11, and outputs the transformed data (DCT coefficient value). The quantization unit 12b quantizes the output data of the orthogonal transform unit 12a, and outputs the quantized data (quantization coefficient value). The encoding unit 12c performs variable length encoding conversion on the output data of the quantization unit 12b, and outputs the converted data (encoded data So).

  The encoding device also includes a quantization control circuit 13. An image signal Si is input to the quantization control circuit 13. The quantization control circuit 13 generates control information for controlling the quantization step in the quantization unit 12b based on image data for one screen (one picture) based on the image signal Si. Then, the quantization control circuit 13 outputs the control information.

The quantization unit 12b generates data obtained by quantizing the DCT coefficient value in a quantization step corresponding to the control information generated by the quantization control circuit 13.
Specifically, the quantization control circuit 13 detects an edge included in one picture and generates control information according to the detection result. The edge is a portion where the difference between the pixel values of two adjacent pixels is large in a pixel row composed of a plurality of pixels, that is, a non-linear portion where the pixel value does not change linearly in the pixel row. Such non-linearity of the pixel value occurs in a portion where a sharp outline is expressed, for example, a black line on a background with little color change. In a portion where such a contour is expressed, quantization noise (mosquito noise) is generated if the quantization step is rough.

  The quantization control circuit 13 detects an edge included in one picture, and generates control information so that a quantization step for a picture including a contour is made fine according to the detection result. When the same image data is used, in the quantization unit 12b, the data quantized in the fine quantization step retains more high-frequency components than the data quantized in the coarse quantization step. Yes. Therefore, it is possible to suppress the deterioration of the image quality by making the quantization step fine.

In the present embodiment, as shown in FIG. 2, 16 × 16 pixels are defined as one macro block MB. When an individual pixel is shown, one direction (left and right direction in the figure) of the two-dimensionally arranged pixel arrangement direction is an X direction, a direction different from the X direction (up and down direction in the figure) is a Y direction, Numbers from 0 to 15 are assigned to the directions from the upper left to the lower right, and the numbers in the respective directions are used as pixels a _{x, y} . For example, the upper left pixel in the drawing is the pixel a _0,0 , the upper right pixel is the pixel a _15,0, and the lower right pixel is the pixel a _15,15 .

  As shown in FIG. 3, the quantization control circuit 13 has a setting register 21. The setting register 21 stores various types of information supplied to the units included in the quantization control circuit 13. These pieces of information are set by a CPU (not shown), for example. Therefore, the setting register 21 may have a function of an interface for connecting the encoding device and an external device (CPU), or the interface may include a setting register.

The information stored in the setting register 21 includes threshold information, size information, and parameter set.
The threshold information includes a vertical threshold Vth, a horizontal threshold Hth, an oblique threshold VHth, a macroblock determination threshold MBth, and a picture determination threshold Qth. The threshold values Vth, Hth, and VHth are used to determine the nonlinearity of the macroblock. The threshold value MBth is used to determine the short wavelength property of the macroblock based on the nonlinear value. The threshold value Qth is used to determine whether or not to perform quantization step control on the picture based on the short wavelength property of the macroblock included in one picture. The vertical threshold value Vth, the horizontal threshold value Hth, and the oblique threshold value VHth are supplied to the nonlinear value calculation unit 22. The macroblock determination threshold value MBth is supplied to the MB short wavelength detector 30. The picture determination threshold value Qth is supplied to the subdivision determination unit 32.

  The size information includes a screen size setting value Sz and the number of macroblocks P. The screen size setting value Sz is the number of pixels included in one picture supplied from the encoding circuit 12 and the quantization control circuit 13 according to the input image signal Si, that is, an image sensor (not shown) that outputs the image signal Si. Is a set value corresponding to the number of pixels. The screen size setting value Sz is supplied to the screen end detection unit 25. The number of macroblocks P is the number of macroblocks included in one picture. The macro block MB includes 16 × 16 pixels a. Therefore, the macroblock number Sp is set based on the screen size setting value Sz and the number of pixels a included in the macroblock. The macroblock number P is supplied to the average value calculation unit 27 and the segmentation determination unit 32.

  The parameter set Ps includes a parameter QpBdOffsety, a constant k, and the like. The parameter QpBdOffsety is a parameter corresponding to the bit width per pixel, and is calculated by QpBdOffsety = 6 × bit_depth_luma_minus8. Bit_depth_luma_minus8 is a value obtained by subtracting “8” from the bit width per pixel. The parameter set Ps is supplied to the difference value calculation unit 33.

  The non-linear value calculation unit 22 calculates a non-linear detection value G for the macro block based on the pixel value of each macro block and the threshold values Vth, Hth, and VHth. The nonlinear detection value G corresponds to the number of points at which the pixel values (hereinafter simply referred to as pixel values) included in the macroblock do not change linearly, that is, the number of nonlinear points. That is, the non-linear value calculation unit 22 counts non-linear points and outputs the count value as a non-linear detection value G corresponding to the macroblock MB.

  More specifically, the nonlinear value calculation unit 22 determines, for each pixel included in the macroblock MB, whether or not the pixel column including the pixel includes a nonlinear point. The pixel column includes three pixels that are continuous in a predetermined direction (vertical direction (Y-axis direction), horizontal direction (X-axis direction), and diagonal direction) for the two-dimensionally arranged pixels a (see FIG. 2). The pixel values of the three pixels are defined as a first pixel value, a second pixel value, and a third pixel value according to the pixel arrangement direction.

  The non-linear value calculation unit 22 includes a central pixel value (second pixel value) of three pixel values included in the pixel column and one of the two pixel values adjacent to the pixel value (first pixel value). A linear prediction value is calculated based on the pixel value. That is, based on the first pixel value and the second pixel value, a pixel value (linear prediction value) when linearly changing from the second pixel value is calculated. Specifically, a difference between the first pixel value and the second pixel value is calculated, and the difference value is added to the second pixel value to calculate a linear prediction value. Of the pixel values adjacent to the second pixel value, the pixel value adjacent to the first pixel value in the direction different from the first pixel value (third pixel value) is obtained by changing the pixel value of this pixel column linearly. The value is within the prediction range based on the calculated linear prediction value. This prediction range is set by threshold values Vth, Hth, VHth. Therefore, the nonlinear value calculation unit 22 calculates the absolute value of the difference between the linear prediction value and the third pixel value, and when the absolute value is smaller than the threshold value, the first to third pixel values are Judged to be linear. On the other hand, if the absolute value is larger than the threshold value, the nonlinear value calculation unit 22 determines that the first to third pixel values are nonlinear, and determines that the pixel value includes a nonlinear point. Count up.

  A first example of the pixel value is shown in FIG. The pixel values of three pixels A, B, and C that are included in the pixel column and are adjacent in the horizontal direction are “5”, “20”, and “3”. The nonlinear value calculation unit 22 calculates a linear prediction value based on the pixel values of the pixels A and B. FIG. 4B shows the pixel values of the pixels A to C and the linear prediction value E as a bar graph. At this time, the linear prediction value E is “35”. The nonlinear value calculation unit 22 calculates a difference between the linear prediction value E and the pixel value C, and compares the absolute value of the difference with a threshold value (for example, a horizontal threshold value Hth). For example, if the horizontal threshold value Hth is “5”, the difference value (absolute value) is “32” in this case, which is larger than the threshold value Hth (= 5). Accordingly, the non-linear value calculation unit 22 determines that the non-linear point is included in this pixel row, and counts up the count value.

A second example of pixel values is shown in FIG. Now, the nonlinear value calculation unit 22 sets the pixel row to three pixels a _0,0 , a _1,1 and a _2,2 that are diagonally adjacent to each other. These pixel values are as shown in the figure, and the linear prediction value is “0”. At this time, if the diagonal threshold value VHth is “5”, the difference between the linear prediction value and the pixel value of the pixel a ₂ , ₂ is larger than the threshold value VHth. Accordingly, the non-linear value calculation unit 22 determines that the non-linear point is included in this pixel row, and counts up the count value. Next, when the pixel columns are the pixels a _1,1 , a _2,2 , and a _3,3 , the linear prediction value is “197”, and the difference between the linear prediction value and the pixel value of the pixel a _3,3 is Is smaller than the threshold value VHth. Therefore, the non-linear value calculation unit 22 determines that the non-linear point is not included in this pixel row, and holds the count value.

  Then, the non-linear value calculation unit 22 performs the above determination on all the pixel values included in the macro block for each macro block. That is, the nonlinear value calculation unit 22 calculates the nonlinear detection value G for the macroblock according to the following equation.

つまり、非線形値算出部２２は、線形予測値がしきい値越える画素列の判定値を「１」とし、線形予測値がしきい値以下の画素列の判定値を「０」とする。そして、非線形値算出部２２は、マクロブロックＭＢについて、画素列の判定値を累積加算し、その累積的な加算結果を非線形検出値Ｇとして出力する。

That is, the nonlinear value calculation unit 22 sets the determination value of the pixel column whose linear prediction value exceeds the threshold value to “1”, and sets the determination value of the pixel column whose linear prediction value is equal to or less than the threshold value to “0”. Then, the nonlinear value calculation unit 22 cumulatively adds the pixel column determination values for the macroblock MB, and outputs the cumulative addition result as the nonlinear detection value G.

  The non-linear detection value G corresponds to the number of non-linear points where the difference between adjacent pixel values is equal to or greater than a threshold value, and indicates the degree of change in pixel value in the macroblock. That is, the macro block having a large non-linear detection value G includes more edges than the macro block having a small non-linear detection value G, that is, the high-frequency component is included in the orthogonally transformed data. Therefore, calculating the non-linear detection value G is equivalent to calculating the high frequency degree (the intensity of change in pixel value), that is, the short wavelength degree in units of macroblocks.

  In the above equation (1), the first term on the right side indicates the determination and count for the pixel row including three pixels adjacent in the vertical direction for the pixel a included in the macroblock shown in FIG. Similarly, the second term on the right side indicates determination and count for a pixel column including three pixels adjacent in the horizontal direction. Further, the third term on the right side indicates determination and count for a pixel column including three pixels adjacent in the oblique direction from the upper left to the lower right with respect to the pixel a included in the macroblock illustrated in FIG. Indicates determination and count for a pixel column including three pixels adjacent in an oblique direction from upper right to lower left.

  As described above, the nonlinear value calculation unit 22 detects an edge extending along an arbitrary direction included in the macroblock MB by calculating the nonlinear detection value G using three pixel values adjacent to each other in the four directions. Can do.

  The screen head detection unit 23 detects the timing at which the first pixel data is input in each picture, that is, the input start timing of the pixel value, based on the image signal Si, and outputs the head detection signal Ss and the clear signal Scl. The input start timing is detected by a signal readout pulse, a vertical synchronization signal, a transfer start pulse signal, and the like. The head detection signal Ss is supplied to the write control unit 24, and the clear signal Sc is supplied to the cumulative addition unit 26.

  The writing control unit 24 responds to the head detection signal Ss output from the screen head detection unit 23, the nonlinear detection value G output from the nonlinear value calculation unit 22, and the macroblock MB in which the nonlinear detection value G is calculated. Control is performed to write to the area (memory address) of the memory 28 corresponding to the. For example, the write control unit 24 supplies the memory 28 with a write control signal WC including an address signal corresponding to a macroblock every time the nonlinear detection value G is output from the nonlinear value calculation unit 22. In response to the write control signal WC, the memory 28 stores the non-linear detection value G output from the non-linear value calculator 22 at a memory address specified by the address signal included in the control signal WC.

  The screen end detection unit 25 detects the timing at which the last pixel data is input in each picture, that is, the input end timing of the pixel value, based on the image signal Si, and outputs the end detection signal Se. The input end timing is detected by a pixel value count result, a horizontal synchronization signal count result, a transfer end pulse signal, and the like. The end detection signal Se is supplied to the read control unit 29, the picture average value calculation unit 27, and the segmentation determination unit 32.

  The cumulative addition unit 26 cumulatively adds the non-linear detection value G output from the non-linear value calculation unit 22 and outputs the value of the addition result. That is, the cumulative addition unit 26 has a register for storing a value, adds the nonlinear detection value G output from the nonlinear value calculation unit 22 to the stored value stored in the register, and stores the stored value (= cumulative value). Output. Further, the cumulative addition unit 26 clears the stored value (= 0) in response to the clear signal Sc output from the screen head detection unit 23. The clear signal Sc is output from the screen head detection unit 23 according to the head of the picture. Therefore, the cumulative addition value is cleared at the beginning of the picture. Accordingly, the cumulative addition unit 26 cumulatively adds the nonlinear detection value G output from the nonlinear value calculation unit 22 in units of pictures.

  The average value calculation unit 27 divides the cumulative value SG output from the cumulative addition unit 26 by the number of macroblocks (total number of macroblocks) P included in one picture, and outputs the calculation result. The accumulated value is the total value of the non-linear detection values G corresponding to the macroblocks included in one picture, and the result of dividing the total value by the total number of macroblocks is the result for all macroblocks included in one picture. This is the average value (EdgeAverage) of the nonlinear detection values G. This average value EA is supplied to the difference value calculation unit 33.

  In response to the termination detection signal Se, the read control unit 29 generates a read control signal RC for reading the nonlinear detection value G stored in the memory 28. The end detection signal Se is output corresponding to the data of the last pixel of one picture. The read control unit 29 generates a read control signal RC including an address signal so as to read the non-linear detection value G corresponding to the macro block MB processed in the circuit 12 if desired. Accordingly, the number of nonlinear detection values G corresponding to one picture is stored in the memory 28, and these nonlinear detection values G are sequentially read out by the read control unit 29. The nonlinear detection value RG output from the memory 28 is supplied to the difference value calculation unit 33. In addition, when distinguishing the nonlinear detection value RG of each macroblock MB, the code | symbol Gmb which attached | subjected the number mb of macroblock MB shall be used.

  The short wavelength detection unit 30 compares the non-linear detection value Gmb output from the non-linear value calculation unit 22 with the macroblock determination threshold value MBth supplied from the setting register 21 and detects the detection value according to the comparison result. Output MM. The short wavelength detection unit 30 outputs a detection value MM having a first predetermined value (for example, “1”) when the nonlinear detection value Gmb is larger than the threshold value MBth, and the nonlinear detection value Gmb is equal to or less than the threshold value MBth. In this case, a detection value MM having a second predetermined value (for example, “0”) is output.

  The macroblock determination threshold MBth is a threshold for the non-linear point count value in the macroblock, and determines whether the quantization step needs to be changed for the frequency component included in the image corresponding to the macroblock. It is a set value for The short wavelength component is a high frequency component and is generated by a non-linear point. Therefore, the threshold value MBth is set so as to determine that the quantization step needs to be changed for a macroblock including many short wavelength components. Therefore, when the detection value MM is the first predetermined value “1”, it indicates that the quantization step needs to be changed for the macroblock MB corresponding to the detection value MM, and the detection value MM is the second predetermined value. A value of “0” indicates that the quantization step need not be changed.

  The cumulative addition unit 31 cumulatively adds the detection values MM output from the short wavelength detection unit 30, and outputs a cumulative value H as a result of the addition. That is, the cumulative addition unit 31 has a register for storing a value, adds the detection value MM output from the short wavelength detection unit 30 to the stored value stored in the register, and stores the addition result in the register. . Then, the cumulative addition unit 31 outputs the cumulative value H stored in the register. The accumulated value H is supplied to the segmentation determination unit 32.

  The cumulative addition unit 31 clears the stored value (= 0) in response to the clear signal Sc output from the screen head detection unit 23. Accordingly, the cumulative addition unit 31 cumulatively adds the detection value MM output from the short wavelength detection unit 30 in units of pictures.

  That is, the short wavelength detection unit 30 and the cumulative addition unit 31 detect the number of macroblocks MB that need to be changed in the quantization step among the macroblocks MB included in one picture, that is, one picture. That is, the short wavelength detection unit 30 and the cumulative addition unit 31 execute processing corresponding to the following equation and output the cumulative value H.

細分化判定部３２は、累積加算部３１から供給される累積値Ｈと、設定レジスタ２１から供給されるピクチャ判定用しきい値Ｑｔｈ，総マクロブロック数Ｐに基づいて、量子化ステップの変更に対する判定フラグＦＩを設定する。この判定フラグＦＩは、差分値算出部３３に供給される。

The subdivision determination unit 32 responds to the change of the quantization step based on the cumulative value H supplied from the cumulative addition unit 31, the picture determination threshold value Qth supplied from the setting register 21, and the total number of macroblocks P. A determination flag FI is set. This determination flag FI is supplied to the difference value calculation unit 33.

  The picture determination threshold value Qth is a threshold value for determining whether or not to change the quantization step for one picture, and the quantization for the total number of macroblocks P included in one picture It is set as a ratio of the number of macroblocks determined to require a step. Accordingly, the same threshold value Qth can be used for images having different picture sizes.

  That is, the subdivision determination unit 32 divides the cumulative value H supplied from the cumulative addition unit 31 by the total number of macroblocks P supplied from the setting register 21, and the calculation result and the threshold supplied from the setting register 21. The value Qth is compared. Then, when the calculation result is larger than the threshold value Qth, the subdivision determination unit 32 determines that the quantization step needs to be changed for the picture, and a value indicating that (for example, “1”). The determination flag FI is output. On the other hand, when the calculation result is equal to or less than the threshold value Qth, the subdivision determination unit 32 determines that the quantization step does not need to be changed for the picture, and a value indicating that (for example, “0”, ) Determination flag FI is output.

  That is, as shown in FIG. 6, one picture Ip includes a first macroblock MBa that requires a change in the quantization step and a second macroblock MBb that requires a change in the quantization step.

  When the ratio of the second macroblock MBb to the picture Ip is small, the quantization noise generated in the second macroblock MBb is not noticeable with respect to the entire picture Ip. In other words, there is little deterioration in the quality of the displayed image visually. On the other hand, when the ratio of the second macroblock MBb to the picture Ip is large, the quantization noise generated in the second macroblock MBb is easily noticeable with respect to the entire picture Ip. That is, the quality of the displayed image is visually deteriorated. That is, the subdivision determination unit 32 can also set the determination flag FI depending on whether quantization noise is conspicuous. In this case, the threshold value Qth may be set based on the ratio of the macroblock MB in which quantization noise is conspicuous with respect to the picture Ip.

  Based on the flag FI supplied from the subdivision determination unit 32, the difference value calculation unit 33 uses the average value EA supplied from the average value calculation unit 27 and the memory 28 when the quantization step needs to be changed. Based on the read nonlinear detection value G and the parameter set Ps supplied from the setting register 21, a difference value mb_qp_delta (hereinafter referred to as a difference value MD) is calculated by the following equation. On the other hand, based on the flag FI, the difference value calculation unit 33 outputs the difference value MD having a value of “0” without performing arithmetic processing when the quantization step does not need to be changed.

上記式において、「Ｃｌｉｐ３」は、値を設定された範囲内の値に制限する処理を示す。例えば、「Ｒ＝Ｃｌｐｉ３（ｘ、ｙ、ｚ）」は、ｘが範囲の下限値を、ｙが範囲の上限値を、ｚが対象の値（演算結果）を、Ｒが処理結果を示す。値ｚが下限値以下の場合には処理結果Ｒ＝ｘとし、値ｚが上限値ｙ以上の場合には処理結果Ｒ＝ｙとし、値ｚが下限値ｘと上限値ｙとの間の値の場合には処理結果Ｒ＝ｚとする。

In the above formula, “Clip 3” indicates processing for limiting the value to a value within the set range. For example, “R = Clpi3 (x, y, z)” indicates that x is the lower limit value of the range, y is the upper limit value of the range, z is the target value (calculation result), and R is the processing result. When the value z is less than or equal to the lower limit value, the processing result R = x. When the value z is greater than or equal to the upper limit value y, the processing result R = y, and the value z is a value between the lower limit value x and the upper limit value y. In this case, the processing result R = z.

  The difference value MD is a value corresponding to the quantization parameter QP used in the quantization unit 12b (see FIG. 1) of the encoding circuit 12. More specifically, the difference value MD is a difference value between the quantization parameter QP of the macroblock processed in the quantization unit 12b and the quantization parameter QP of the macroblock processed immediately before. A quantization parameter QP (Quantization Parameter) corresponds to the quantization step. For example, H.M. In the H.264 / AVC coding standard, the quantization parameter QP and the logarithm of the quantization step are proportional. Therefore, changing the value of the difference value MD means changing the quantization parameter QP, that is, changing the quantization step. When the difference value MD is “0”, the quantization step for the immediately preceding macroblock is maintained, and when the difference value MD is other than “0”, the quantization step is changed.

  As described above, the difference value MD is calculated based on the average value EA of the non-linear detection values G in one picture and the non-linear detection value G of each macroblock. Therefore, the quantization control circuit 13 controls the quantization step for the macroblock according to the non-linear detection value G of the macroblock. Further, the quantization control circuit 13 determines control / non-control of the quantization step according to the number of nonlinear macroblocks included in one picture.

  Subtracting the non-linear detection value G of each macroblock from the average value EA generates a difference value MD calculated from the non-linear detection value G and the average value EA so as to be inversely proportional to the non-linear detection value G within the above setting range. It is to be. Thereby, the larger the nonlinear detection value G, the smaller the difference value MD and the finer the quantization step. That is, the quantization step is made finer for a macroblock MB having many nonlinear points, that is, a macroblock MB having a short wavelength (high frequency).

  The difference value MD becomes zero (0) when the nonlinear detection value G is equal to the average value EA, and the quantization step is not changed in the quantization unit 12b. When the nonlinear detection value G is smaller than the average value EA, the difference value MD is larger than “0”, and the quantization step in the quantization unit 12b is more than the quantization step using the nonlinear detection value G equal to the average value EA. Also becomes rough.

  Making the quantization step finer increases the amount of code data generated by variable-length coding conversion, and making the quantization step coarser means that the amount of code data generated by variable-length coding conversion. Reduce. Therefore, by increasing the quantization step, it is possible to suppress an increase in the amount of code data generated by variable length coding conversion.

Next, the operation of the encoding apparatus configured as described above will be described.
Steps 51 to 67 shown in FIG. 7 indicate processing executed by the encoding device. Steps 51 to 59 show processing executed by the quantization control circuit 13 shown in FIG. 1, and Steps 60 to 67 show processing executed by the encoding circuit 12 shown in FIG. Therefore, the operation of the circuit corresponding to each step will be described.

  In step 51, when the quantization control circuit 13 detects the start of reception of the input image, the quantization control circuit 13 clears the cumulative value SG for the non-linear detection value G and the cumulative value H for the short wavelength detection value MM.

  In step 52, the quantization control circuit 13 counts the non-linear points for each macro block MB of the input image, and sets the count value as the non-linear detection value G. Then, the quantization control circuit 13 cumulatively adds the nonlinear detection value G and sets the addition result as a cumulative value SG. Further, the quantization control circuit 13 compares the nonlinear detection value G with the threshold value MBth to detect the short wavelength property of the macroblock MB, cumulatively adds the detection value MM, and sets the addition result as the cumulative value H. .

In step 53, the quantization control circuit 13 stores the calculated nonlinear detection value G in an area corresponding to the macro block MB in the memory 28.
In step 54, the quantization control circuit 13 determines whether or not the reception of the input image for one picture has been completed. If the reception has not been completed, the process proceeds to step 52. Move on to the step.

  In step 55, the quantization control circuit 13 calculates the ratio of macroblocks determined to be short wavelength in one picture based on the accumulated value H and the total number of macroblocks P for one picture, and sets the ratio value as the ratio value. The threshold value Qth is compared, and the determination flag FI is set according to the comparison result. That is, the quantization control circuit 13 determines whether it is necessary to change the quantization step for one picture, and sets the determination flag FI.

  In step 56, based on the determination flag FI, the quantization control circuit 13 proceeds to the next step 57 when the determination flag is 眞 (= true: 1), and the determination flag FI is false (= false: 0). ) Go to Step 60. That is, the quantization control circuit 13 executes the processing of steps 57 to 59 when it is determined that the quantization step needs to be changed for one picture, and when it is determined that the quantization step does not need to be changed. Steps 57 to 59 are skipped.

In step 57, the quantization control circuit 13 calculates an average value EA (EdgeAverage) of the nonlinear detection values G in one picture based on the accumulated value SG.
In step 58, the quantization control circuit 13 reads the nonlinear detection value G from the memory 28. In step 59, the quantization control circuit 13 calculates a difference value MD based on the read nonlinear detection value G.

  In step 60, the encoding circuit 12 performs orthogonal transform (discrete cosine transform: DCT) on the image data in units of macroblocks, and quantizes the converted data in a quantization step based on the difference value MD. Note that the difference value MD is cleared at the start of execution of the steps shown in FIG. Therefore, when the quantization control circuit 13 skips steps 58 to 59, the cleared difference value MD (= 0) is input to the encoding circuit 12. Therefore, the encoding circuit 12 quantizes the image data of the current macroblock MB in the quantization step for the previously processed macroblock MB based on the cleared difference value MD.

  In step 61, the encoding circuit 12 performs variable length encoding conversion on the image data generated by the conversion processing in step 60, and outputs the converted data (encoded data So).

  In step 62, the encoding circuit 12 dequantizes the image data (quantization coefficient) generated by the conversion process in step 60 to reproduce the conversion coefficient. Further, the encoding circuit 12 performs inverse orthogonal transform (inverse discrete cosine transform: inverse DCT) on the reconstructed transform coefficient. That is, the encoding circuit 12 decodes the quantized coefficient generated by the quantization.

In step 63, the encoding circuit 12 performs a process using a deblocking filter, and removes block distortion in the decoded image data.
In step 64, the encoding circuit 12 performs an intra-screen prediction process, and generates a prediction image for each block with reference to a reference frame stored in a frame memory (not shown).

In step 65, the encoding circuit 12 performs a motion compensation process, and generates a motion vector with reference to a reference frame stored in a frame memory (not shown).
In step 66, the encoding circuit 12 generates a local decoded image (predicted frame) based on the processing results of steps 64 and 65.

  In step 67, the encoding circuit 12 acquires a difference between the local decoded image and the input image. The encoding circuit 12 performs orthogonal transformation and quantization on the processing in step 60 including the difference acquired in step 67.

As described above, according to the present embodiment, the following effects can be obtained.
(1) The quantization control circuit 13 calculates a non-linear detection value G for each macro block MB obtained by dividing a picture Ip of a predetermined size according to a difference between adjacent pixel values included in the macro block MB, and detects the non-linear detection thereof. The quantization parameter for controlling the quantization step when the macroblock MB is quantized is changed according to the value G.

  The nonlinear detection value G corresponds to the number of points at which the pixel values (hereinafter simply referred to as pixel values) included in the macroblock do not change linearly, that is, the number of nonlinear points. A macroblock having a large non-linear detection value G includes more edges than a macro block having a small non-linear detection value G, that is, it indicates that there are many high-frequency components in orthogonally transformed data. Therefore, the quantization noise can be reduced by making the quantization step finer for the macroblock MB having more nonlinear points, that is, the macroblock MB having a higher short wavelength property (high frequency property).

  (2) The non-linear value calculation unit 22 of the quantization control circuit 13 counts up the non-linear detection value G for each macro block MB according to the difference in pixel values in the pixel columns in the macro block MB. The difference value calculation unit 33 calculates a difference value for controlling the quantization step based on the nonlinear detection value. Accordingly, the quantization parameter (difference value) corresponding to the nonlinear detection value G can be easily calculated.

  (3) The cumulative addition unit 26 of the quantization control circuit 13 cumulatively adds the nonlinear detection values. The average value calculation unit 27 calculates the average value EA of the nonlinear detection values G in the picture Ip based on the accumulated value SG and the screen size setting value Sz of the picture Ip. Then, the difference value calculation unit 33 calculates a difference value MD for controlling the quantization step based on the average value EA and the non-linear detection value RG. Therefore, the larger the non-linear detection value G, the smaller the difference value MD and the finer the quantization step.

  (4) The quantization control circuit 13 detects the short wavelength picture Ip according to the non-linear detection value, and indicates whether or not to change the quantization step for the picture Ip based on the detection result. A determination flag is set, and a difference value calculated for a picture determined to require a change in quantization step is output in accordance with the determination flag, and a predetermined value is set for a picture determined to require no change in the quantization step. The difference value of was output. Therefore, control / non-control of the quantization step can be easily determined.

  (5) The short wavelength detection unit 30 compares the nonlinear detection value G with the threshold value MBth and generates a detection value MM indicating that the short wavelength macroblock MB has been detected. The cumulative addition unit 31 cumulatively adds the detection value MM. Based on the accumulated value H and the number P of macroblocks MB included in the picture Ip, the subdivision determination unit 32 calculates a ratio of the macroblocks MB determined to be a short wavelength included in the picture Ip, and the ratio and threshold value The determination flag FI is set by comparing with Qth. Therefore, the control / non-control of the quantization step for each picture Ip can be easily determined.

  (6) The nonlinear value calculation unit 22 calculates a linear prediction value from two of the three pixel values, calculates a difference between the pixel value not used for calculating the linear prediction value and the linear prediction value, and The non-linear detection value G is calculated by comparing the difference value with the threshold value. As a result, nonlinear points such as edges included in the macroblock MB can be detected, and the number of nonlinear points in the macroblock MB can be easily calculated.

In addition, you may implement the said embodiment in the following aspects.
In the non-linear value calculation unit 22, in a pixel column including three pixel values, a linear prediction value is calculated from the first pixel value and the third pixel value at both ends included in the pixel column, and the linear calculation value A difference value from the second pixel value at the center of the pixel row is calculated, and the linear value is determined by comparing the difference value with a threshold value.

  The number of pixels included in the pixel column in the nonlinear value calculation unit 22 is changed as appropriate. For example, a non-linear point is determined for a pixel column including two pixels. The nonlinear value calculation unit 22 calculates an absolute value of a difference value between two pixel values, compares the absolute value with a threshold value corresponding to the direction of the pixel column, and when the absolute value is larger than the threshold value. , It is determined that the pixel column includes a non-linear point. Therefore, the non-linear value calculation unit 22 calculates the non-linear detection value G for the macroblock MB by the following equation.

上記の構成により、マクロブロック単位での高周波度（画素値変化の激しさ）を算出することができる。

With the above configuration, it is possible to calculate the high-frequency level (the intensity of pixel value change) in units of macroblocks.

For example, as shown in FIG. 8A, the pixel values of three pixels A, B, and C are “5”, “6”, and “30”. FIG. 8B shows the pixel values of the pixels A to C as a bar graph.
The nonlinear value calculator 22 calculates a difference value D (= 1) between the pixel values of the pixels A and B, and compares the difference value D with a threshold value (for example, a horizontal threshold value Hth). For example, if the horizontal threshold value Hth is “5”, the difference value D is smaller than the threshold value Hth, so the nonlinear value calculation unit 22 determines that the pixel A is not a nonlinear point. .

  Next, the nonlinear value calculation unit 22 calculates a difference value D (= 26) between the pixel values of the pixels B and C, and compares the difference value D with a threshold value (for example, a horizontal threshold value Hth). In this case, since the difference value D is larger than the threshold value Hth, the nonlinear value calculation unit 22 determines that the pixel B and C are nonlinear points. Thus, by counting the non-linear points, it is possible to detect the high frequency (intensity of change in pixel value) in units of macroblock MB.

  Change the size of the macro block as appropriate, such as the number of pixels other than 16 × 16 pixels, for example, 8 × 8 pixels, 4 × 4 pixels, etc. Further, the nonlinear detection value G is calculated for each macroblock in which the number of vertical pixels and the number of horizontal pixels are set to different values, for example, 8 × 16 pixels.

  The quantization control circuit 13 reads out the image data stored in the image memory 11 in units of macroblocks to generate control information, that is, determines the subdivision of the quantization step for the picture stored in the image memory 11. As described above, when the image data stored in the image memory 11 is read, the screen head detection unit 23 outputs the clear signal Sc and the head detection signal Ss at the start of reading the image data, and the screen end detection unit 25 reads the data. What is necessary is just to change so that the termination | terminus detection signal Se may be output at the time of completion | finish. Further, both the detection units 23 and 25 may be omitted by using a signal indicating the start of reading as the clear signal Sc and the head detection signal Ss and using a signal indicating the end of reading as the end detection signal Se.

-The short wavelength detection part 30, the accumulation addition part 31, and the subdivision determination part 32 which are shown in FIG. 3 are abbreviate | omitted, and the difference value MD is calculated about all the pictures.
Steps 55 and 56 shown in FIG. 7 are omitted, and difference values MD are calculated for all the pictures.

The present invention is applied to an encoding circuit of an image processing apparatus that performs image code conversion (for example, from standard MPEG2 to standard H.264).
At least one of various parameters stored in the setting register 21 is set as a fixed value in a processing unit that uses the parameter.

  The encoding circuit 12 and the quantization control circuit 13 are implemented as one circuit. Further, at least one of the encoding circuit 12 and the quantization control circuit 13 is a processor (for example, CPU) that executes a program including the steps shown in FIG.

The following notes are disclosed regarding the above embodiments.
(Appendix 1)
For each sub-block obtained by dividing a block of a predetermined size, a quantum for controlling a quantization step when the sub-block is quantized according to a non-linear detection value calculated according to a difference between adjacent pixel values included in the sub-block A quantization control circuit characterized by changing a quantization parameter.
(Appendix 2)
A non-linear value calculation unit that counts up a non-linear detection value according to a difference in pixel values in a pixel column in the sub block for each sub block;
A parameter calculation unit for calculating a quantization parameter for controlling the quantization step based on the non-linear detection value;
The quantization control circuit according to appendix 1, characterized by comprising:
(Appendix 3)
A first cumulative addition unit that cumulatively adds the nonlinear detection values;
An average value calculation unit that calculates an average value of nonlinear detection values in the block based on the cumulative value of the first cumulative addition unit and the size of the block;
Including
The parameter calculation unit calculates a quantization parameter for controlling the quantization step based on the average value and the nonlinear detection value;
The quantization control circuit according to appendix 2, characterized in that:
(Appendix 4)
A determination unit configured to detect a short wavelength block according to the non-linear detection value and set a determination flag for indicating whether or not to change a quantization step for the block based on the detection result; ,
According to the determination result of the determination unit, a calculated value or a predetermined value is output as a quantization parameter.
The quantization control circuit according to any one of Appendices 1 to 3, characterized in that:
(Appendix 5)
The determination unit
A short wavelength detection unit that generates a detection value indicating that a short wavelength sub-block has been detected by comparing the nonlinear detection value with a threshold value;
A second cumulative addition unit that cumulatively adds the detection values;
Based on the cumulative value of the second cumulative adder and the number of sub-blocks included in the block, a ratio of sub-blocks determined to be short wavelengths included in the block is calculated, and the ratio is compared with a threshold value. A subdivision determination unit for setting the determination flag;
The quantization control circuit according to appendix 4, characterized by comprising:
(Appendix 6)
The nonlinear value calculation unit calculates a linear prediction value from two of the three pixel values, calculates a difference between the pixel value not used to calculate the linear prediction value and the linear prediction value, and the difference The quantization control circuit according to appendix 2, wherein the value is compared with a threshold value, and the non-linear detection value is counted up according to the comparison result.
(Appendix 7)
The non-linear value calculation unit calculates a difference between two adjacent pixel values, compares the difference value with a threshold value, and counts up the non-linear detection value according to the comparison result. A quantization control circuit according to 1.
(Appendix 8)
A head detection unit that detects the head of the image data of the block and generates a head detection signal;
An end detection unit that detects the end of the image of the block and generates an end detection signal;
A write control unit for performing control to write the non-linear detection value in a memory in response to the head detection signal;
A read control unit that performs control to read a non-linear detection value from the memory in response to the termination detection signal;
Have
The parameter calculation unit calculates the quantization parameter based on a non-linear detection value read from the memory;
4. The quantization control circuit according to appendix 2 or 3, characterized by the above.
(Appendix 9)
For each sub-block obtained by dividing a block of a predetermined size, a quantum for controlling a quantization step when the sub-block is quantized according to a non-linear detection value calculated according to a difference between adjacent pixel values included in the sub-block A quantization parameter changing method characterized by changing a quantization parameter.
(Appendix 10)
For each sub-block obtained by dividing a block of a predetermined size, a quantum for controlling a quantization step when the sub-block is quantized according to a non-linear detection value calculated according to a difference between adjacent pixel values included in the sub-block A quantization control circuit for changing the quantization parameter;
An encoding circuit that sets a quantization step according to the quantization parameter, quantizes the image data in the quantization step, and converts the quantized data into code data;
An encoding circuit.

DESCRIPTION OF SYMBOLS 12 Coding circuit 12b Quantization part 13 Quantization control circuit 22 Nonlinear value calculation part 33 Difference value calculation part (quantization parameter calculation part)
G, Gmb Nonlinear detection value H Cumulative value MD Difference value (quantization parameter: mb_qp_delta)
Ip picture (block)
MB macroblock (sub-block)
a _{x, y} pixel, pixel value

Claims (6)

For each sub-block obtained by dividing a block having a predetermined number of pixel data, depending on the difference between the pixel values of two adjacent pixels of the three pixels adjacent in succession in a predetermined direction included in the sub-block A linear prediction value is calculated , and the sub-block is determined according to a non-linear detection value based on a difference between a pixel value of a pixel different from the adjacent two pixels of the three consecutive pixels and the linear prediction value. A quantization control circuit characterized by changing a quantization parameter for controlling a quantization step at the time of quantization. A non-linear value calculation unit that counts up a non-linear detection value according to a difference between a pixel value of a pixel different from the two adjacent pixels and the linear prediction value ;
A parameter calculation unit for calculating a quantization parameter for controlling the quantization step based on the non-linear detection value;
The quantization control circuit according to claim 1, comprising: A first cumulative addition unit that cumulatively adds the nonlinear detection values;
An average value calculation unit that calculates an average value of nonlinear detection values in the block based on the cumulative value of the first cumulative addition unit and the size of the block;
Including
The parameter calculation unit calculates a quantization parameter for controlling the quantization step based on the average value and the nonlinear detection value;
The quantization control circuit according to claim 2. A determination unit configured to detect a short wavelength block according to the non-linear detection value and set a determination flag for indicating whether or not to change a quantization step for the block based on the detection result; ,
According to the determination result of the determination unit, a calculated value or a predetermined value is output as a quantization parameter.
The quantization control circuit according to any one of claims 1 to 3. For each sub-block obtained by dividing a block having a predetermined number of pixel data, depending on the difference between the pixel values of two adjacent pixels of the three pixels adjacent in succession in a predetermined direction included in the sub-block A linear prediction value is calculated , and the sub-block is determined according to a non-linear detection value based on a difference between a pixel value of a pixel different from the adjacent two pixels of the three consecutive pixels and the linear prediction value. A quantization parameter changing method, characterized by changing a quantization parameter for controlling a quantization step at the time of quantization. For each sub-block obtained by dividing a block having a predetermined number of pixel data, depending on the difference between the pixel values of two adjacent pixels of the three pixels adjacent in succession in a predetermined direction included in the sub-block A linear prediction value is calculated , and the sub-block is determined according to a non-linear detection value based on a difference between a pixel value of a pixel different from the adjacent two pixels of the three consecutive pixels and the linear prediction value. A quantization control circuit for changing a quantization parameter for controlling a quantization step at the time of quantization;
An encoding circuit that sets a quantization step according to the quantization parameter, quantizes the image data in the quantization step, and converts the quantized data into code data;
An encoding device .

Images