KR100804451B1 - 1/4 quarter pixel interpolation method for imaging process and processor thereof - Google Patents

Abstract

A quarter pixel interpolation method for image processing and a processor thereof are provided to perform horizontal interpolation and vertical interpolation by using an SIMD(Single Instruction Multiple Data) command, thereby reducing the amount of operations and accordingly increasing a data processing rate. A quarter pixel interpolation method for image processing comprises the following steps of: obtaining a first half pixel and a first quarter pixel by using information about a neighboring integer pixel(230,240) among pixels(250); obtaining a second half pixel(270) and a second quarter pixel(280,290) by using information about the pixel where horizontal interpolation is completed(260); and obtaining a third quarter pixel by using a third half pixel located in a diagonal line and a third half pixel located in a horizontal line among the third quarter pixels and the third half pixels generated after the above two steps. The above three steps are performed by an SIMD(Single Instruction Multiple Data) command supported in a processor.

Description

1/4 pixel interpolation method of image processing and its processor {1/4 QUARTER PIXEL INTERPOLATION METHOD FOR IMAGING PROCESS AND PROCESSOR THEREOF}

FIG. 1A is a diagram for describing a method of performing a pixel according to the position of each pixel when performing 1/4 pixel interpolation in the image processing apparatus, and FIG. 1B is a diagram illustrating a performance of performing a corresponding module in the image processing apparatus. One drawing.

2 is a diagram schematically illustrating a structure of a pixel generated after performing a method of quarter pixel interpolation according to the present invention.

3A to 3C are structural diagrams of a register for explaining a process in which horizontal interpolation is completed.

4 is a structural diagram illustrating a process of completing longitudinal interpolation according to the present invention.

5 is a structural diagram showing pixel information generated by horizontal interpolation and vertical interpolation in detail according to an exemplary embodiment of the present invention.

6 is a structural diagram illustrating a method of obtaining a third quarter pixel through diagonal longitudinal interpolation of the present invention.

7A to 7C are graphs and tables for comparing quarter quarter pixel interpolation using the SIMD instruction of the present invention and execution time of JM Reference.

<Explanation of symbols for the main parts of the drawings>

230, 240: integer pixel 250: horizontal interpolation

260: vertical interpolation 270: second half pixel

280, 290: second quarter pixel

The present invention relates to a 1/4 pixel interpolation method and a processor thereof for efficiently processing data for removing temporal redundant data of an image.

In general, image processing refers to processing an image signal using a computer. Here, in relation to video signals, standards such as JPEG, MPEG, and H.264 employ a block-based video compression scheme. Among these, as an example, H.264 / AVC uses MC (Motion Compensation) and ME (Motion Estimation) methods to remove temporal redundant data of an image. In this process, MV (Motion Vector) is 1 /. It uses a 4-pixel interpolation method.

Therefore, in H.264 / AVC of image processing, a 1/4 pixel interpolation method is adopted in image data processing to make MV of 1/4 pixel unit.

FIG. 1A is a diagram for describing a method of performing pixel interpolation according to the position of each pixel when the image processing apparatus is performed. FIG. 1B is a diagram illustrating the performance of each corresponding module in the image processing apparatus. Drawing. Here, FIG. 1A defines integer pixels, half pixels, and quarter pixels in the H.264 standard document. At this time, the position of each pixel is not shown at all.

Referring to FIG. 1A, a pixel used in H.264 is composed of an integer pixel, a half pixel, and a quarter pixel. The integer pixel is a unit of a pixel representing the original image, and as shown in the H.264 standard document, A, B, C, D, E, F, G, H, I, J, K, L, M, N Located at, P, Q, R, S, T, U In addition, the half pixel is a unit of a pixel to express the original image in more detail, and is located at aa, bb, b, cc, dd, h, j, m, ee, ff, s, gg, and hh. In addition, the quarter pixel is a unit of a pixel to express an image more finely than an integer pixel and is located at a, c, d, e, f, g, I, k, n, p, q, and r.

The quarter-pixel interpolation method is performed by generating a half pixel and a quarter pixel using the integer pixel. More specifically, the half pixel may be represented by a six-tap filtering method using an integer pixel. It can be expressed as an average between half pixels produced by 6-tap filtering and neighboring integer pixels. The method for obtaining the half pixel and the quarter pixel can be expressed by Equation 1 defined in the H.264 standard.

Half pixel

b1 = (E-5F + 20G + 20H-5I + J), b = ((b1 + 16) >> 5)

h1 = (A-5C + 20G + 20M-5R + T), h = ((h1 + 16) >> 5)

Quarter pixel

a = (G + b + 1) >> 1; (Left and right directions), d = (G + h + 1) >> 1; (Up and down direction)

e = (b + h + 1) >> 1; (Left diagonal direction), g = (b + m + 1) >> 1; (Right diagonal direction) (Equation 1)

Equation 1 is a formula for obtaining a half pixel of b and h and a quarter pixel of d and g among half and quarter pixels.

According to Equation 1, b and h are half pixels, d and g are quarter pixels, E, F, G, H, I, and J are integer pixels, and constants 1, -5, 20, 20, -5 , 1 means the filter coefficient used for 6-tap filtering with symmetrical values. Therefore, the quarter-pixel interpolation method is completed by obtaining the half pixel and the quarter pixel through the above-described equation (1).

However, the conventional quarter-pixel interpolation method sequentially interpolates. For example, the pixels on the line where the integer pixels A and C are located are interpolated, the pixels on the line where E and H are positioned are interpolated, and the pixels on the line where R and T are located are interpolated. Therefore, in the conventional 1/4 pixel interpolation method, operation operations such as load, store, and move of a memory occur a lot.

In addition, MC (Motion Compensation), in which the 1/4 pixel interpolation method is used in H.264 image processing, uses integer transform, entropy decoding, and deblocking in the entire process as shown in FIG. It requires the most amount of computation compared to Deblocking.

 An object of the present invention is to provide a 1/4 pixel interpolation method of image processing capable of high speed processing by reducing unnecessary computation amount.

The 1/4 pixel interpolation method of the present invention for achieving the above object is a horizontal interpolation process of obtaining a first half pixel and a first quarter pixel using neighboring integer pixel information among pixels, and after the horizontal interpolation, The vertical interpolation process of obtaining the second half pixel and the second quarter pixel by using the horizontal interpolation pixel information and the third half pixel and the third quarter pixel generated after the horizontal interpolation and the vertical interpolation are performed. The method may include a diagonal longitudinal interpolation process of obtaining a third quarter pixel by using the third half pixel positioned in a horizontal direction and the third half pixel disposed in a diagonal direction.

The equation for obtaining the first half pixel and the second half pixel is b = (((G + H) * 4-F-I) * 5 + E + J + 16) / 32, and h = ((G + M) * 4-C-R ) * 5 + A + T + 16) / 32, and b and h are arbitrary positions of the first half pixel and the second half pixel, and the A, C, E, F, G, H, I, J, M, R, T Is the position of neighboring integer pixels of b and h.

The horizontal interpolation may include storing a predetermined integer pixel in a first register so as to perform 6-tap filtering and generating a first half pixel by performing the 6-tap filtering operation on the data stored in the first register. And generating a first quarter pixel by obtaining an average value between any integer pixels among the integer pixels adjacent to the first half pixel and the first half pixel.

The vertical interpolation may include storing a first half pixel and an integer pixel adjacent to the first half pixel in a second register among pixel information obtained from the horizontal interpolation after the horizontal interpolation. Generating a second half pixel by performing the six-tap filtering operation on the data stored in the second quarter pixel, and obtaining an average value between adjacent integer pixels of the second half pixel and the second half pixel. (quarter pixel) may be generated.

In addition, the diagonal longitudinal interpolation may include storing values of the third half pixel positioned in the horizontal direction and the third half pixel positioned in the diagonal in a third register, and the two third half pixels positioned in the horizontal direction. Obtaining a center pixel value of the third quarter pixel by using a value and calculating a value of the remaining pixels other than the center value of the third quarter pixel by using a value of the third half pixel positioned diagonally. Can be.

In addition, the horizontal interpolation, the vertical interpolation, and the diagonal vertical interpolation may be performed by a single instruction multiple data (SIMD) instruction supported by a processor.

The SIMD instruction may be performed by the processor with at least one of PADDW, PSLLW, PSUBW, PMULW, PADDWM, PSRLW, PUNKxx, and PAVGB instructions.

In addition, the processor according to the present invention after the horizontal interpolation and the horizontal interpolation to obtain a first half pixel (first half pixel) and the first quarter pixel (quarter pixel) by using the adjacent integer pixel information of the pixel, the horizontal direction Vertical interpolation is performed to obtain a second half pixel and a second quarter pixel using the interpolated pixel information.

In addition, the horizontal interpolation and the vertical interpolation may be performed by a single instruction multiple data (SIMD) command.

The third half pixel and the third half pixel, which are formed after performing the horizontal interpolation and the vertical interpolation, are disposed between two third half pixels positioned in a horizontal direction and two third half pixels disposed in a diagonal direction. It may be performed by further obtaining a third quarter pixel existing between.

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the present invention.

FIG. 2 is a diagram schematically illustrating a structure of a pixel generated after performing a 1/4 pixel interpolation method according to the present invention.

As shown in Fig. 2, the 1/4 pixel interpolation method of the present invention includes horizontal interpolation, longitudinal interpolation, and diagonal longitudinal interpolation.

The horizontal interpolation 250 includes the first half pixels h00 to h07 existing between two integer pixels using two integer pixels when the integer pixel values are regularly arranged on the x-axis from x00 to x12. The values of the first quarter pixels q00 to q15 are obtained.

For example, a first half pixel h00 existing between two integer pixels is obtained by using an integer pixel of x02 and an integer pixel of x03. Further, the first quarter pixel q00 is obtained by using the integer pixel x02 and the first half pixel h00. Also, the first quarter pixel q01 is obtained using the integer pixel x03 and the first half pixel h00.

As described above, the first half pixel and the first quarter pixel existing between neighboring integer pixels may be obtained using integer pixels having a regular arrangement on the x-axis. Therefore, the horizontal interpolation is completed by obtaining the first half pixel and the first quarter pixel arranged on the x-axis. As the horizontal interpolation 250 scans each row at the same time as scanning from left to right, the number of accesses of the memory can be reduced.

After the horizontal interpolation 250 is completed, the vertical interpolation 260 obtains the second half pixel and the second quarter pixel by using integer pixel information formed on the y-axis in which the horizontal interpolation is completed. Herein, the integer pixels on the y-axis in which the horizontal interpolation is completed are integer pixels regularly arranged on the y-axis, for example, the integer pixels regularly arranged on the y-axis from y00 to y50. The second half pixel and the second quarter pixel to be generated between any two rows of the horizontal interpolation using the integer pixel (y00 to y50) information or the first half pixel 230 and 240 information generated by the horizontal interpolation. Will be obtained.

For example, the second half pixel 270 existing between the two integer pixels is obtained by using the integer pixel y00 and the integer pixel y10. In addition, the second quarter pixel 280 existing between the two pixels may be obtained using the integer pixel y00 and the second half pixel 270. In addition, the second quarter pixel 290 existing between the two pixels may be obtained using the integer pixel y10 and the second half pixel 270.

In addition, after the horizontal interpolation is completed, two first half pixels (eg, two first half pixels 230, 240 of the first half pixels generated by the horizontal interpolation, for example, arranged on the y-axis) are used. The second half pixel and the second quarter pixel existing between 230 and 240 may be obtained. Although not illustrated, the second half pixel and the second quarter pixel are positioned at the center of the vertical interpolation region 260. As described above, the longitudinal interpolation 260 is performed after the horizontal interpolation is completed, and the vertical interpolation 260 is completed by simultaneously performing the scanning from the left to the right.

However, although the region of the vertical interpolation 260 is not displayed, interpolation is not performed in some regions, and the third half pixel generated after the horizontal interpolation 250 and the vertical interpolation 260 is performed. A diagonal longitudinal interpolation is performed to generate a new third quarter pixel by using a portion of the third quarter pixel. As described above, the method of obtaining the first and second half pixels by the horizontal interpolation 250 and the vertical interpolation 260 and the method of obtaining the third quarter pixel by the diagonal longitudinal interpolation will be described later with reference to FIGS. The process in 5 will be explained more clearly. First, the horizontal interpolation is as follows.

3A to 3C are structural diagrams of a register for explaining a process in which horizontal interpolation is completed.

Horizontal interpolation includes storing a predetermined integer pixel in a first register, generating a first half pixel, and generating a first quarter pixel.

Referring to FIG. 3A, horizontal interpolation according to an exemplary embodiment of the present invention includes first storing a predetermined integer pixel in a first register to enable 6-tap filtering. The first register is composed of six registers having variables of xmm0 to xmm5, and each of the first registers has values of eight integer pixels. The value of the integer pixel is a value where the integer pixel described with reference to FIG. 2 is located.

That is, an integer pixel value from x00 to x07 is stored in the first register of xmm0. In addition, an integer pixel value from x01 to x08 is stored in a first register of xmm1, an integer pixel value from x02 to x09 is stored in a first register of xmm2, and an integer pixel from x03 to x10 is stored in a first register of xmm3. The pixel value is stored, the first register of xmm4 stores integer pixel values from x04 to x11, and the first register of xmm5 stores integer pixel values from x05 to x12. Here, the above-described integer pixel value, for example, x00 is 0 rows and 0 columns on the x axis, x01 is 0 rows and 1 column on the x axis, x02 is 0 rows and 2 columns on the x axis, and x12 is 0 rows 12 on the x axis. Means heat. Here, only one line is shown, but the remaining lines can be the same as the above-described method. Accordingly, all integer pixel values related to the first half pixel to be obtained in FIG. 2 are stored in the first register.

Thereafter, horizontal interpolation has a step of generating a first half pixel by performing a six-tap filtering operation on the value of the integer pixel stored in the first register. The six-tap filtering operation generates the first half pixel according to the equation for obtaining the first half pixel, and the first half pixel is expressed by Equation 2 below.

b = ((E-5 * F + 20 * G + 20 * H-5 * I + J) +16) / 32

  = (((E + J)-(F + I) * 5 + (G + H) * 20) +16) / 32

  = (((G + H) * 4-F-I) * 5 + E + J + 16) / 32

h = ((A-5 * C + 20 * G + 20 * M-5 * R + T) +16) / 32

  = (((A + T)-(C + R) * 5 + (G + M) * 20) +16) / 32

  = ((G + M) * 4-C-R) * 5 + A + T + 16) / 32-(Equation 2)

Equation 2 is a form in which Equation 1 described above is rearranged to reduce the amount of computation, and b and h are positions of first half pixels, and A, C, E, F, G, H, I, J, M, R and T are the positions of the adjacent integer pixels of b and h, which are shown based on FIG.

In Equation 2, '* 4' is a multiplication, but is calculated and processed by a 2-byte shift. '* 5', on the other hand, is directly processed by the multiplication operation. Therefore, in Formula 1, b and h need four multiplications (*) four times, respectively. In Equation 2, b and h multiply (*) one time each. Therefore, Equation 2 can improve the data processing speed by the processor by reducing the number of multipliers.

Here, the operation of 6-tap filtering to obtain the first half pixel by applying Equation 2 is as shown in FIG. 3B. Here, FIG. 3B applies b = (((G + H) * 4-F-I) * 5 + E + J + 16) / 32 of Formula 2 as an example.

Referring to FIG. 3B, when (G + H) is first applied in Equation 2, an integer pixel value of x02 to x09 is stored in the register 1. In addition, the register 2 stores integer pixel values from x03 to x10. In addition, arithmetic of xmm6 = xmm2 + xmm3 is performed in the register 3, and the corresponding data value is stored. In this case, the addition operation uses a PADDW instruction among SIMD instructions supported by the processor. Subsequently, when (G + H) * 4 is applied in Equation 2, the operation of xmm6 = (xmm2 + xmm3) * 4 is performed in the register 4, and the corresponding data value is stored. At this time, '* 4' is performed by the shift command of PSLLW, 2 among SIMD instructions. Subsequently, when (G + H) * 4-F is applied in equation 2, the integer pixel values from x01 to x08 are stored in the register 5. Therefore, the operation of xmm6 = xmm6-xmm1 is performed in the register 6, and the corresponding data value is stored. Here, m00 to m07 means calculated data from x02 + x03 << 2 to x09 + x10 << 2. The subtraction operation is performed by the SIMD instruction of the PSUBW.

Subsequently, when (G + H) * 4-F-I is applied in Equation 2, when an integer pixel value of xmm4 is stored in the register 7, an operation of xmm6 = xmm6-xmm4 is performed in the register 8, thereby correspondingly. The data value is stored. Subsequently, if ((G + H) * 4-F-I) * 5 is applied in Equation 2, the operation of xmm6 = xmm6 * 5 is performed in the register 9, and the corresponding data value is stored. At this time, the multiplication operation (*) is performed by the SIMD instruction of PMULW.

Subsequently, when ((G + H) * 4-F-I) * 5 + E is applied in Equation 2, when the integer pixel value of xmm0 is stored in the register 10, the operation of xmm6 = xmm6 + xmm0 is performed in the register 11. Data values corresponding to m00 + x00, m01 + x01, m02 + x02, m03 + x03, m04 + x04, m05 + x05, m06 + x06 and m07 + x07 are stored. Subsequently, if ((G + H) * 4-F-I) * 5 + E + J is applied in Equation 2, when the integer pixel value of xmm5 is stored in the register 12, the operation of xmm6 = xmm6 + xmm5 is performed in the register 13. Data values corresponding to m00 to m07 are stored.

Subsequently, if ((G + H) * 4-F-I) * 5 + E + J + 16 is applied, when 16 data values are stored in the register 14, the operation of xmm6 = xmm6 + 16 is performed in the register 15 so that the corresponding m00 + 16. To m07 + 16 are stored. Subsequently, when ((G + H) * 4-F-I) * 5 + E + J + 16 + 16) / 32 is applied in Equation 2, the operation of xmm6 = (xmm6 + 16) / 32 is performed on the register 16, thereby finally the first half pixel. The data values of h00, h01, h02, h03, h04, h05, h06, and h07 are stored.

As such, the process of generating the first half pixel using 6-tap filtering according to Equation 2 uses only four-law operations and shifts, and the data processing speed is improved by processing these operations with SIMD instructions supported by the processor.

Subsequently, horizontal interpolation may include generating a first quarter pixel by obtaining an average value between the above-described first half pixel and an adjacent integer pixel of the first half pixel. Referring to FIG. 3C, the value of the first half pixel obtained in FIG. 3B is stored in the register 17. Subsequently, arbitrary integer pixel values are stored in registers 18 and 19. For example, the data values of xmm2 and xmm3 described in FIG. 3B are stored in the registers 18 and 19. At this time, the data stored in the registers 18 and 19 are integer pixel values located at both sides of the first quarter pixel position to be obtained. Therefore, two integer pixel values may be used in the registers 18 and 19 according to the position of the first quarter pixel to be obtained.

Subsequently, in the register 20, the first half pixel value stored in the register 17 and the integer pixel value stored in the register 18 are alternately positioned by using the PUNKxx instruction among the SIMD instructions to alternately place the first half stored in the register 17. The pixel value and the integer pixel value stored in the register 18 are alternately stored.

Subsequently, the register 21 alternately stores the integer pixel value stored in the register 19 and the first half pixel value stored in the register 17 in the opposite manner to the above-described method.

Accordingly, the value of the first quarter pixel is obtained and stored in the register 22 using a PAVGB SIMD instruction, which is a SIMD instruction that averages pixel information stored in the register 20 and pixel information stored in the register 21. The first quarter pixel is completed. As a result, the first quarter pixel is completed by obtaining an average value between the first half pixel and the neighboring integer pixels of the first half pixel.

4 is a structural diagram illustrating a process of completing longitudinal interpolation according to the present invention.

The longitudinal interpolation according to the present invention includes storing a first half pixel and a predetermined integer pixel in a second register, generating a second half pixel, and generating a second quarter pixel.

Referring to FIG. 4, in the vertical interpolation, the horizontal interpolation is first performed, and then, in the pixel information obtained from the horizontal interpolation, a first half pixel and an integer pixel neighboring the first half pixel described above are stored in a second register. Has The second register is composed of six registers having variables of xmm0 to xmm5, and each of the second registers alternately stores four integer pixel values and four first quarter pixel values generated after horizontal interpolation.

For example, an integer pixel value from y00 to y03 and a first half pixel value from h00 to h03 are alternately stored in the second register of xmm0. In addition, an integer pixel value from y10 to y13 and a first half pixel value from h10 to h13 are alternately stored in the second register of xmm1. In addition, an integer pixel value from y20 to y23 and a first half pixel value from h20 to h23 are alternately stored in the second register of xmm2.

In addition, an integer pixel value from y30 to y33 and a first half pixel value from h30 to h33 are alternately stored in the second register of xmm3. In addition, an integer pixel value from y40 to y43 and a first half pixel value from h40 to h43 are alternately stored in the second register of xmm4. In addition, an integer pixel value from y50 to y53 and a first half pixel value from h50 to h53 are alternately stored in the second register of xmm5.

Herein, the above-described integer pixel values are partially illustrated in FIG. 2. More specifically, for example, y00 is 0 rows and 0 columns on the y axis, y10 is 1 row and 0 columns on the y axis, and y20 is the y axis. 2 rows and 0 columns, and y30 means 3 rows and 0 columns on the y axis. In addition, a first half pixel value, for example, h03 is a first half pixel 0 row and 3 columns, h13 is a first row 3 columns, and h33 is a first half pixel 3 rows and 3 columns.

Accordingly, an integer pixel value and a first half pixel value at positions of all integer pixel pixels related to the second half pixel to be obtained in FIG. 2 are stored in the second register.

Thereafter, vertical interpolation generates the second half pixel by performing the six tap filtering operation on the data stored in the second register. Thereafter, an average value between the above-described second half pixel and the neighboring integer pixel of the second half pixel is obtained to generate a second quarter pixel. The method of obtaining the second half pixel and the second quarter pixel is omitted because the same method is described with reference to FIGS. 3B and 3C.

Meanwhile, the structures of the first and second half pixels and the first and second quarter pixels of the horizontal interpolation and the vertical interpolation generated by the above-described method will be described below.

5 is a structural diagram showing pixel information generated by horizontal interpolation and vertical interpolation in detail according to an exemplary embodiment of the present invention.

In FIG. 5, only some of the pixels generated by the horizontal interpolation 500 and 510 and the vertical interpolation 520 are illustrated. That is, the first row and the fifth row of FIG. 5 represent the first half pixel h and the first quarter pixel q generated by horizontal interpolation including integer pixels, and the second through fourth rows are vertical directions. A second half pixel h and a second quarter pixel q generated by interpolation are shown.

As an example, a generated process is described. In the horizontal interpolation 500, the first half pixel 503 is formed using a six-tap filtering operation method using the integer pixel 501 and the integer pixel 502. . Accordingly, the horizontal interpolation 500 may make the first quarter pixel 504 using the integer pixel 501 and the first quarter pixel 502.

In addition, as an example of the vertical interpolation 520 method, the vertical interpolation 520 is a six-tap filtering of the integer pixel 502 and the integer pixel 512 among the data stored in the second register described with reference to FIG. 4. The second half pixel 523 is produced by the calculation method. In addition, the vertical interpolation 520 performs a six-tap filtering operation on the first quarter pixel 503 and the first quarter pixel 513 among the data stored in the second register described with reference to FIG. 4. Make it up Accordingly, the vertical interpolation 520 may generate the second quarter pixel 526 by obtaining an average value of the integer pixel 502 and the first half pixel 523.

In this way, the horizontal interpolation and the vertical interpolation are completed, but the interpolation is not performed in some of the vertical interpolation operations (530), and the diagonal longitudinal interpolation generated in the part will be described below.

6 is a structural diagram illustrating a method of obtaining a third quarter pixel through diagonal longitudinal interpolation of the present invention.

6 is described in connection with FIG. Referring to FIG. 6, the present invention refers to a third half pixel in which the first and second half pixels generated by the horizontal interpolation and the vertical interpolation are combined, and the third half pixel positioned in the horizontal direction and the diagonally positioned third half pixel. Storing a value of three half pixels in a third register.

For example, a third half pixel value from h10 to h17 is stored in the third register 23, and a third half pixel value from h11 to h18 delayed by one byte is stored in the third register 24. do. The register 25 stores a third half pixel value having a diagonal relationship among the third half pixel information stored in the third register 23 and the third register 24. The register 26 calculates an average of the data value stored in the third register 23 and the data value stored in the third register 24 to produce a third quarter pixel value from q10 to q17. At this time, for example, one of the third quarter pixels to be made is a third quarter pixel existing between the third half pixel 522 and the third half pixel 523 in FIG. 5. This third quarter pixel can be obtained by performing a SIMD instruction of PAVGB.

In the register 27, third half pixel values of h00, h00, h01, h01, h02, h02, h03, and h03 are stored, and the third half pixel refers to a half pixel located in the first row of FIG. . At this time, two randomly arranged. In the register 28, third half pixel values of h20, h20, h21, h21, h22, h22, h23, and h23 are stored, and the third half pixel is the value of the half pixel in the fifth row of FIG. At this time, two randomly arranged.

Accordingly, the register 29 obtains an average of the third half pixel value stored in the register 25 and the third half pixel value stored in the register 27 to generate a third quarter pixel value from q00 to q07. In this case, the created third quarter pixel is a value of a pixel generated between the second quarter pixel q of the second row in FIG. 5.

In addition, the register 30 obtains an average of the third half pixel value stored in the register 25 and the third half pixel value stored in the register 28 to produce a third quarter pixel value from q20 to q27. In this case, the created third quarter pixel is a value of a pixel generated between the second quarter pixel q of the fourth row in FIG. 5.

As a result, the third quarter pixel value stored in the register 26 is obtained by using a third half pixel positioned in the horizontal direction, and the third quarter pixel values existing in the register 29 and the register 30 are diagonally formed. This value is obtained using the third half pixel information located.

The third quarter pixel is generated by performing PAVGB and PUNCKxx instructions among SIMD instructions supported by the processor.

Accordingly, the quarter quarter pixel interpolation method of the present invention is completed by generating the first and second quarter pixels by the horizontal interpolation and the vertical interpolation and the third quarter pixels by the diagonal longitudinal interpolation. In addition, the processor for image processing of the present invention includes a first register, a second register, a third register, and other registers for storing data of pixel information by a quarter-quarter pixel interpolation method. In addition, the processor for image processing of the present invention can provide a SIMD instruction to efficiently perform pixel processing.

7A to 7C are graphs and tables for comparing the quarter-quarter pixel interpolation using the SIMD instruction of the present invention and the execution time of the JM Reference.

FIG. 7A illustrates the average execution time of the 1/4 pixel interpolation method 710 and the JM Reference 700 using the SIMD instruction in the CIF (352 × 288) image, and FIG. 7B illustrates the SIMD instruction in the QVGA (320 × 240) image. Shows the average execution time of the 1/4 pixel interpolation method 730 and the JM Reference 720.

 As shown in the graphs of FIGS. 7A and 7B, the 1 / 4-pixel interpolation methods 710 and 730 using the SIMD instruction are faster in processing time in terms of average execution time than the JM Reference 700 and 720, which is a general method. have. This can be summarized as a table of FIG. 7C. That is, as shown in FIG. 7C, Referenc has an average execution time of 0.2030 in the CIF image, and 1/4 quarter pixel interpolation using the SIMD instruction has an average execution time of 0.0115. In addition, Referenc has an average execution time of 0.1728 in the QVGA image, and quarter-quarter pixel interpolation using the SIMD instruction has an average execution time of 0.0090. Therefore, quarter-quarter pixel interpolation using SIMD instructions can be performed 17 times faster than the method used in the JM Reference.

As described above, preferred embodiments of the invention are disclosed for purposes of illustration, and any person skilled in the art may make various modifications, changes, additions, etc. within the spirit and scope of the present invention. Modifications, changes and additions should be considered to be within the scope of the claims.

As described above, the quarter quarter pixel interpolation method of the present invention performs horizontal interpolation and vertical interpolation using a SIMD instruction, thereby reducing the amount of computation and increasing the data processing speed.

In addition, the quarter-quarter pixel interpolation using the SIMD instruction of the present invention has the effect of performing about 17 times faster than the method used in the JM Reference.

In addition, the 1/4 quarter pixel interpolation method of the present invention can perform encoding and decoding of H.264 / AVC quickly, thereby serving a service in various environments supporting SIMD instructions.

Claims (10)

In a 1/4 pixel interpolation method of image processing for removing temporal redundant data of an image expressed in pixel units, A horizontal interpolation process of obtaining a first half pixel and a first quarter pixel using neighboring integer pixel information among the pixels; A longitudinal interpolation process of obtaining a second half pixel and a second quarter pixel by using the horizontally interpolated pixel information after the horizontal interpolation; A third quarter by using the third half pixel positioned in the horizontal direction and the third half pixel positioned diagonally, among the third half pixel and the third quarter pixel generated after the horizontal interpolation and the vertical interpolation A diagonal longitudinal interpolation process for obtaining pixels, wherein the horizontal interpolation, the vertical interpolation, and the diagonal longitudinal interpolation are performed by a single instruction multiple data (SIMD) instruction supported by a processor. 1/4 pixel interpolation method. The method of claim 1, The equation for obtaining the first half pixel and the second half pixel is b = (((G + H) * 4-F-I) * 5 + E + J + 16) / 32, h = ((G + M) * 4-C-R) * 5 + A + T + 16) / 32. b and h are arbitrary positions of the first half pixel and the second half pixel, and A, C, E, F, G, H, I, J, M, R, and T are neighbors of b and h. 1 / 4-pixel interpolation method of image processing, characterized in that the position of the integer pixel. The method of claim 1, The horizontal interpolation is Storing a predetermined integer pixel in a first register to enable six-tap filtering; Generating a first half pixel by performing the six tap filtering operation on data stored in the first register; And generating a first quarter pixel by obtaining an average value between any integer pixel among the neighboring integer pixels of the first half pixel and the neighboring integer pixels of the first half pixel. . The method of claim 1, The longitudinal interpolation is After the horizontal interpolation, storing a first half pixel and an integer pixel neighboring the first half pixel among second pixel information obtained from the horizontal interpolation in a second register; Generating a second half pixel by performing the six tap filtering operation on data stored in the second register; And generating a second quarter pixel by obtaining an average value between the second half pixel and a neighboring integer pixel of the second half pixel. The method of claim 1, The diagonal longitudinal interpolation is Storing a value of the third half pixel positioned in the horizontal direction and the third half pixel disposed in a diagonal line in a third register; Obtaining a center pixel value of the third quarter pixel using values of two third half pixels positioned in the horizontal direction; And calculating a value of the remaining pixels excluding the center value of the third quarter pixel by using the value of the third half pixel positioned on the diagonal line. delete delete A processor for image processing performed by a 1/4 pixel interpolation method of removing temporal redundant data of an image expressed in pixel units, A horizontal interpolation for obtaining a first half pixel and a first quarter pixel using neighboring integer pixel information among the pixels; And After the horizontal interpolation, vertical interpolation is performed to obtain a second half pixel and a second quarter pixel by using the pixel information on which the horizontal interpolation is completed, and the horizontal interpolation and the vertical interpolation are performed using a SIMD supported by a processor. Processor for image processing, characterized in that performed by a Single Instruction Multiple Data) command. delete The method of claim 8, Between the third half pixel generated after performing the horizontal interpolation and the vertical interpolation, between the third half pixels positioned in the horizontal direction and between the third half pixels positioned diagonally. And further obtains a third quarter pixel existing in the processor.

Images