KR101320611B1 - A method and apparatus for encoding/decoding of N bit video data with quantization parameter extension - Google Patents

Abstract

Disclosed is a method of encoding N bit image data in which a quantization parameter is extended. This N bit image data encoding method generates difference data from which predictive data of image data is removed from input image data, converts the generated difference data into a frequency domain, and is proportional to a bit depth N representing the precision of the image data. By quantizing the transformed differential data based on the extended quantization parameter (QP '), it is possible to encode image data having an extended bit depth by changing only a minimum of hardware specifications. Regardless, the same image quality can be obtained.

Description

A method and apparatus for encoding / decoding of N bit video data with quantization parameter extension}

1 is a block diagram illustrating a configuration of an audio and video coding standard (AVS) encoder according to an embodiment of the present invention.

2 is a block diagram illustrating a more detailed configuration of the transform and quantizer 113 and the inverse transform and inverse quantizer 114 according to the first embodiment of the present invention.

3 is a flowchart for describing an operation of the transform and quantization unit 113 illustrated in FIG. 2.

4 is a block diagram illustrating a more detailed configuration of the transform and quantizer 413 and the inverse transform and inverse quantizer 414 according to the second embodiment of the present invention.

FIG. 5 is a flowchart for describing an operation of the transform and quantization unit 413 illustrated in FIG. 4.

6 is a block diagram illustrating a configuration of an AVS decoder according to an embodiment of the present invention.

7 is a block diagram showing a more detailed configuration of the inverse transform and inverse quantization unit 614 according to the third embodiment of the present invention.

FIG. 8 is a flowchart for describing an operation of the inverse transform and inverse quantization unit 614 shown in FIG. 7.

 9 is a block diagram showing a more detailed configuration of the inverse transform and inverse quantization unit 914 according to the fourth embodiment of the present invention.

FIG. 10 is a flowchart for describing an operation of the inverse transform and inverse quantization unit 914 illustrated in FIG. 9.

11 is a graph illustrating a relationship between a peak signal to noise (PSNR) and a bit rate according to a first embodiment of the present invention.

12A and 12B are graphs showing the relationship between the PSNR and the bit rate according to the second embodiment of the present invention.

The present invention relates to image processing technology, and more particularly, to a method and apparatus for encoding and decoding image data having various bit depths of 8 bits or more.

In video communication systems and devices, there is a growing demand for large screens and hence high resolution. In the electronics market, consumer demand for large CRTs, LCDs, PDPs and high-definition projectors (TVs) is increasing. Moreover, the spread of image data processed and stored digitally such as MPEG, DVD, DV, and the like is accelerating this further. Accordingly, improving the image quality of images or dynamic effects displayed on large screens with high resolution has emerged as a very important problem.

In terms of the video encoding apparatus, the image quality of the displayed image is affected by the number of bits representing the image data value. This is because the bit depth representing the precision of the image data, that is, the number of bits representing the image data value increases, so that the data can be represented at various levels.

However, according to the conventional audio and video coding standard (AVS), the conventional encoding apparatus performs encoding only on data having a constant bit depth, such as 8 bit data. However, as the video compression technology develops, it may process data having a higher bit depth, such as 10 bit or 12 bit or more, depending on the type of video data input or the demand of a high-definition display application. The performance of the encoding device that can be required. However, increasing the specification of the hardware to process data with higher bit depths is undesirable because it causes a price increase. Therefore, in the coding apparatus and method according to the AVS, it is necessary to improve the image quality of the displayed image data by encoding N bit image data having various bit depths while maintaining the same hardware specification.

The technical problem to be solved by the present invention is in the audio and video coding standard (AVS), encoding method and apparatus that can encode 8-bit or more N-bit image data with a minimum change of hardware specifications, and at the same time improve the coding efficiency To provide. In addition, since the range of the quantization parameter is extended in proportion to the bit depth, quantization is performed based on the quantization step corresponding to the extended quantization parameter, so that a constant bit rate and PSNR can be obtained at the same quantization parameter regardless of the bit depth. A method and apparatus for encoding / decoding N bit image data are provided.

In order to solve the above technical problem, the N-bit image data encoding method of the quantization parameter is extended according to the present invention (a) generating the difference data from the prediction data of the image data from the input image data, (b) Converting the generated differential data into a frequency domain; and (c) converting the difference data based on a quantization parameter QP 'whose range is extended in proportion to a bit depth N representing the precision of the image data. Preferably, the method includes the step of quantizing, and the encoding method further comprises (d) generating a bitstream by encoding the information about the bit depth N and the quantized result, and outputting the bitstream.

In addition, the range of the extended quantization parameter QP 'is preferably extended in proportion to the bit depth N.

In addition, the step (c) preferably includes a step of quantizing the transformed differential data by performing a predetermined operation based on a quantization step corresponding to any one of the extended quantization parameter QP '. (c1) dividing the transformed difference data into a first predetermined value, (c2) multiplying a result of dividing by the first predetermined value by a predetermined coefficient s, (c3) multiplying a result of multiplying the predetermined coefficient s by a second Dividing by a predetermined value, (c4) multiplying a result of dividing by the second predetermined value by a predetermined coefficient q, and (c5) adding a predetermined value k to a result of multiplying the predetermined coefficient q, and adding the predetermined value k. And moving the predetermined bit to the right, wherein the first and second predetermined values are values determined corresponding to the bit depths, and the steps (c1) and (c3) are integer operations.

In order to solve the another technical problem, an N-bit image data encoding apparatus having an extended quantization parameter according to the present invention is a difference data generation unit for generating differential data from which prediction data of the input image data is removed from the input image data. And a transform unit for converting the generated difference data into a frequency domain, and quantizing the converted difference data based on a quantization parameter QP 'whose range is extended in proportion to a bit depth N representing the precision of the image data. Preferably, the encoding apparatus includes a quantization unit, and the encoding apparatus further includes an entropy encoding unit that generates and outputs a bitstream by encoding the information about the bit depth N and the quantized result.

According to another aspect of the present invention, there is provided a method of decoding an N bit image data having an extended quantization parameter according to the present invention. (A) A bit depth indicating a precision of image data included in the bit stream by decoding an input bit stream. Restoring information about (N) and quantization data, (b) inverse quantizing the reconstructed quantized data based on a quantization parameter QP 'whose range is extended in proportion to the bit depth N; and (c) restoring the difference data from which the inverse quantized result is inversely transformed into the time domain to remove the predictive data of the image data from the image data.

In addition, the step (b) is preferably the step of inverse quantization of the reconstructed quantization data based on the quantization step corresponding to any one of the extended quantization parameter (QP '), the step (b) (B1) preferably includes multiplying the reconstructed quantized data by a predetermined coefficient r and (b2) dividing a result of multiplying the predetermined coefficient r by a predetermined coefficient n.

In order to solve the still another technical problem, the N-bit data decoding apparatus in which the quantization parameter is extended according to the present invention decodes the input bitstream to a bit depth N representing the precision of the image data included in the bitstream. An entropy decoding unit for reconstructing information and quantization data, an inverse quantizer for dequantizing the reconstructed quantized data based on a quantization parameter QP 'whose range is extended in proportion to the bit depth N, and the inverse quantization It is preferable to include an inverse transform unit for restoring the difference data from which the predicted data of the image data is removed from the image data by inversely transforming the result into the time domain.

Preferably, the inverse quantization unit inverse quantizes the reconstructed quantization data based on a quantization step corresponding to any one of the extended quantization parameters QP ', and the inverse quantization unit is predetermined in the reconstructed quantization data. A third multiplier for multiplying the coefficient r and a third divider for dividing the result of multiplying the predetermined coefficient r by the predetermined coefficient n are preferable.

Another object of the present invention is to provide a computer-readable recording medium having recorded thereon a program for executing the encoding method and the decoding method according to the present invention.

Hereinafter, with reference to the accompanying drawings and embodiments will be described a preferred embodiment of the present invention. The present invention is not limited to the embodiments described below, and various changes can be made by those skilled in the art from the appended claims, and such conversions fall within the scope of the present invention.

Moving image data is encoded / decoded by a motion prediction technique. Motion prediction is performed with reference to past frames or both past and future frames. When encoding / decoding a current frame, the frame referred to is referred to as a reference frame. In block-based video encoding, a still image or still frame constituting a video is generally divided into macroblocks (MBs) of 16 × 16 units, and macroblocks are subblocks of smaller units. Divided into Thus, the motion between still images is predicted as block to block and encoded.

Hereinafter, an N bit image data encoding apparatus and method in which a quantization parameter is extended according to the present invention will be described with reference to FIG. 1.

1 is a block diagram illustrating a configuration of an N-bit image data encoding apparatus in which a quantization parameter is extended according to an embodiment of the present invention. The division unit 100, the prediction unit 110, the first adder 112, An encoding control unit 120, a transform and quantization unit 113, and an entropy encoding unit 130 are included.

When raw data having an 8-bit bit depth is input, the divider 100 selects a frame or field to be encoded according to the control of the encoding controller 120, and selects each selected angle. Split a frame or field into macro blocks. Each macro block consists of 16x16 pixels. The macroblock of 16 × 16 pixels is composed of a plurality of 16 × 16 blocks and / or a plurality of 8 × 8 blocks that contain luminance or color difference information of the image data.

The encoding controller 120 extracts information about the bit depth N representing the precision of the input image data, and controls the overall operation of the encoding apparatus.

The first adder 112 receives the raw data output from the divider 100 and the prediction data predicted by the intra prediction module 117 or the inter prediction module 118. The differential data (X, differential data) is output by subtracting the prediction data from the raw data.

The transform and quantization unit 113 transforms the differential data X into an integer DCT (Discrete Cosine Transform) and quantizes the transformed differential data X to generate quantized data. The generated quantized data is output to the predictor 110.

The prediction unit 110 includes an inverse transform and inverse quantizer 114, a second adder 111, a deblocking unit 115, a memory 116, an intra-frame prediction unit 117, and inter-frame prediction. An inter prediction module 118 and a motion estimation unit 119 are included.

The inverse transform and inverse quantization unit 114 performs inverse integer DCT transform and inverse quantization on the quantized data output from the transform and quantization unit 113 to generate and output the recovered differential data X.

The second adder 111 adds the predicted data predicted by the intra-frame predictor 117 or the inter-frame predictor 118 to the reconstructed difference data X of the current frame, and decodes the result. Will output

The deblocking unit 115 performs filtering on each restored macroblock in order to reduce block distortion, and outputs each result of the filtering to the memory 116.

The memory 116 stores the data output from the deblocking unit 110 for later use by the intra-frame predictor 117 or the inter-frame predictor 118.

The intra prediction unit 117 performs intra frame prediction, and the inter frame prediction unit 118 performs inter frame prediction. These predictions are also called motion compensation.

The motion estimation unit 119 retrieves a reference frame for the macro block from the memory 116 and expresses a position difference between the image in the macro block of the retrieved reference frame and the image in the macro block of the current frame as a motion vector, Output to the first adder 112.

The entropy encoder 130 receives the control signal of the encoding controller 120, the motion data of the motion estimation unit 119, and the quantization data of the transform and quantization unit 113, and combines them to form an output bitstream. Create

The transform and quantization unit 113 converts and quantizes the difference data X corresponding to the bit depth N of the image data.

Hereinafter, the operation of the transform and quantization unit 113 will be described with reference to FIGS. 2 and 3.

2 is a block diagram illustrating a configuration of a transform and quantization unit 113 of an encoding apparatus according to an embodiment of the present invention, wherein the transform and quantization unit 113 may include a DCT transform unit 210 and a first divider ( 220, a first multiplier 230, a second divider 240, a second multiplier 250, and a shifting unit 260.

FIG. 3 is a flowchart illustrating a process of converting light quantization according to an embodiment of the present invention, in accordance with hardware specification according to step 310 and bit depth N of converting differential data X into a frequency domain. Steps 320 through 350 for adjusting the size of the transformed difference data X and steps 360 and 370 for generating quantization data from the transformed difference data X are adjusted.

Referring to FIG. 3, look at the operation of the transform and quantization unit 113.

In general, since the difference data X is processed in a matrix form, the difference data X will be represented as a difference matrix X for convenience of explanation.

In operation 300, the transform and quantization unit 113 receives information about a bit depth N of an input image matrix.

In operation 310, the first adder 112 generates a difference matrix X by prediction of an input image matrix.

In operation 320, the DCT converter 210 generates a first matrix B by transforming the difference matrix X input from the first adder 112 to an integer DCT using Equation 1 below. For a block, the number and organization of the signal matrix is determined by the raw video format and the encoding mode of the macro block)

A = T

B = AT ^T

Here, "占" denotes a matrix product, T denotes an integer DCT transform matrix, T ^T denotes a transpose of the integer DCT transform matrix T, X denotes a differential matrix, and B denotes a DCT transformed first matrix. According to an embodiment of the present invention, the integer DCT transformation matrix T for 8 × 8 DCT transformation is as follows.

T = [8 8 8 8 8 8 8 8

10 9 6 2 -2 -6 -9 -10

10 4 -4 -10 -10 -4 4 10

9 -2 -10 -6 6 10 2 -9

8 -8 -8 8 8 -8 -8 8

6 -10 2 9 -9 -2 10 -6

4 -10 10 -4 -4 10 -10 4

2 -6 9 -10 10 -9 6 -2].

The DCT converter 210 outputs the integer DCT transformed first matrix B to the first divider 220 using the T matrix and the T ^T matrix.

In operation 330, the first divider 220 divides the converted first matrix B into a first predetermined value Shift Tab0 determined according to the bit depth N of the input image data, and divides the quotient. A second matrix C is generated by rounding to the nearest integer value. The generated second matrix C is transmitted to the first multiplier 230. Operation 330 may be expressed by Equation 2 below.

C = B // Shift Tab0,

Where a // b divides a by b and rounds (approximates to the nearest integer).

 When the first multiplier 230 is implemented by a 16-bit multiplier, when the bit depth N of the input image data is 10 bits, the first predetermined value Shift Tab0 is set to 7. When the bit depth N is 12 bits, the first predetermined value Shift Tab0 is preferably set to 9. However, the first predetermined value Shift Tab0 may be set differently according to the specification of the multiplier implementing the first multiplier 230.

In operation 340, the first multiplier 230 coordinates the second matrix C with a predetermined matrix S having a predetermined coefficient s (i) and a domain-based array multiplication. D) is generated, and the third matrix D is transmitted to the third divider 240. Operation 340 may be represented by Equation 3.

D = S. * C

Here, C is a second matrix, s (i) is a predetermined coefficient determined by the following equation (4), ". *" Operation means a coordinate-based array product. P. * Q is an operation of multiplying each element of P with each element of Q at the same position as that element. According to an embodiment of the present invention, when the difference matrix X is an 8x8 matrix, the predetermined matrix S may be as follows.

S =

Each element s (i) of the predetermined matrix S is as follows.

/v(i), s (i) ≒

/ v (i),

/v(i)와 오차 범위내에 있는 정수이다.Where i is an integer between 0 and 5, and "≒" means almost the same value. That is, s (i) is

/ v (i) and an integer within the margin of error.

According to an embodiment of the present invention,

v (i) has the value of v = [512 × 512, 442 × 442, 464 × 454, 512 × 442, 512 × 464, 442 × 464],

s (i) has the value s = [32768, 43969 39898 37958 36158, 41884].

 We will now discuss the coordinate-based array product in detail. According to an embodiment of the present invention, a process of performing coordinate-based array multiplication of the second matrix C and the predetermined matrix S is as follows.

First, the data of the 0th or 4th row and the 0th or 4th column is multiplied by a preset coefficient s (0), and is any of the first, 3, 5, and 7th rows and is the first, 3, 5, 7th. Data in any one of the columns is multiplied by the predetermined coefficient s (1), and data in the second or sixth row and the second or sixth column is multiplied by the predetermined coefficient s (2). Moreover, data which is 0th or 4th row and is any one of 1st, 3, 5, and 7th row, and data which is 1st, 3rd, 5th, and 7th row and 0th or 4th column is predetermined coefficient s (3 Multiply by), and the data of the second or sixth row in the 0th or 4th row and the data of the 0th or 4th row in the second or 6th row are multiplied by the predetermined coefficient s (4), and the remaining data is multiplied by the predetermined coefficient s (5). Multiply by

Thereafter, the first multiplier 230 transmits the third matrix D generated by the coordinate-based array product to the second divider 240.

In operation 350, the second divider 240 divides the third matrix D by a second predetermined value Shift Tab1 determined corresponding to the bit depth N of the input image data, and divides the quotient of the nearest integer value. Approximation is performed to generate a fourth matrix (E). The generated fourth matrix E is transmitted to the second multiplier 250. Operation in operation 350 may be expressed as Equation 4 below.

E = D // Shift Tab1,

Here, "//" is the same as in Equation 2.

When the second multiplier 250 is implemented by a 16-bit multiplier, when the bit depth N of the input image data is 10 bits, the second predetermined value Shift Tab1 is set to 17. When the bit depth N is 12 bits, the second predetermined value Shift Tab1 is preferably set to 15. However, the second predetermined value Shift Tab1 may be set differently according to the specification of the multiplier implementing the second multiplier 230.

When DCT conversion of the difference matrix X generated in operation 310 is performed in operation 320, important values of image data are collected into a specific portion. For example, meaningful values are collected in reconstructing image data in the upper left portion of the first matrix B generated by DCT conversion. In quantizing these values, in order to improve coding efficiency under limited constraints, more important values may be weighted to quantize important values at finer intervals, and other values may be quantized at slightly less detailed intervals. In step 340, multiplying the predetermined coefficient s by multiplying a different scale factor for each data before extracting the quantized data from the transformed difference data (X), so that important data can be quantized more precisely.

Before and after step 340, steps 330 and 350 divide data into first and second predetermined values that are determined corresponding to bit depths. This is a step of adjusting the data level in order to process the data with the extended bit depth under such a constraint because the specification of the hardware is limited in quantizing the difference data X through a predetermined calculation process. . Therefore, when the bit depth of the image data is increased, in order to process it in a system having the same hardware specification, it is preferable that the first predetermined value is increased in proportion to the bit depth, and conversely, as the first predetermined value is increased, It is preferable that the predetermined value decreases in proportion to the bit depth. As described above, it is preferable that the first and second predetermined values are appropriately adjusted according to the hardware specification so that the results of multiplying the predetermined value or predetermined coefficients in a predetermined calculation process for quantization are not overflowed.

Taking a specific numerical value as an example, in the conventional AVS encoding apparatus, when encoding conventional 8-bit video data, the first predetermined value is set to 5, but in the case of 10-bit or 12-bit video data, the first predetermined value. It is preferable to set this to 7 or 9. That is, in the case of using a 16-bit memory for input and output of the difference matrix (X), a 16-bit far plier for quantization, and a 32-bit Arithmetic Logic Unit (ALU) for a predetermined operation, a conventional 8-bit image is used. In the case of encoding data, the first and second predetermined values are set to 5 and 19, but when the bit depth is increased to 10 bits or 12 bits, the first and second predetermined values are set to 7, 17, or 9, respectively. Can be set to 15.

In operation 360, the second multiplier 250 multiplies the fourth matrix E by a predetermined coefficient q [QP] to generate a fifth matrix F. FIG. The fifth matrix F is transmitted to the shift unit 260. The operation in operation 360 may be expressed as Equation 5 below.

F = q [QP] .E

Here, "." Means a scalar product.

The predetermined coefficient q [QP] is determined as follows by the quantization parameter QP.

/

]q [QP] = round [

Of

]

Here, the quantization parameter QP is an integer between 0 ≦ QP ≦ 63,

round [X] is the integer closest to X.

The symbol "≒" means that the value is approximately the same.

That is, q [QP] is determined according to the quantization parameter QP.

q [64] = {32768,29775,27554,25268,23170,21247,19369,17770,

16302,15024,13777,12634,11626,10624,9742,8958,

8192,7512,6889,6305,5793,5303,4878,4467,

4091,3756,3444,3161,2894,2654,2435,2235,

2048,1878,1722,1579,1449,1329,1218,1117,

1024,939,861,790,724,664,609,558,

512,470,430,395,362,332,304,279,

256,235,215,197,181,166,152,140}

Has the value of.

In operation 370, the shift unit 260 adds a predetermined value k to the fifth matrix F, and shifts the result to the right by 15 bits to generate the sixth matrix G. FIG. The sixth matrix G is transmitted to the entropy encoder 130 as quantized data. The predetermined value k is defined as follows.

*10/31For blocks in a frame, k = (1 << 15) * 10/31 =

* 10/31

*10/62For blocks between frames, k = (1 << 15) * 10/62 =

* 10/62

Here, << means left shift.

The reason for setting k differently according to intra-frame prediction and inter-frame prediction is that it is necessary to set different dead zones of the quantization unit because the characteristics of the difference matrix X are different for each prediction reason.

The whole process described above is expressed by the following equation.

A = T

B = AT ^T

C = B // Shift Tab 0

D = S (i). * C

E = D // Shift Tab1

F = q [QP] .E

G = (F + k) >> 15 (Intra: k = (1 << 15) * 10/31; Inter: k = (1 << 15) * 10/62)

Here, X is an 8x8 matrix of original data, T is an integer DCT transformation matrix, and T ^T is a transpose matrix of the integer DCT transformation matrix T.

The quantization parameter QP is an integer between 0≤QP≤63,

/v(i)이고(i는 0≤i≤5인 정수), s (i)

/ v (i) (i is an integer of 0≤i≤5),

v = [512 × 512, 442 × 442, 464 × 454, 512 × 442, 512 × 464, 442 × 464],

/

이고,q [QP] ≒

Of

ego,

이다.q [QP] * r [QP] ≒

to be.

Where a // b divides a by b, approximates the quotient divided by the nearest integer, PQ is a matrix multiplication of matrix P and matrix Q, aP is a scalar or array product of a and matrix P (array multiplication), P. * Q, is a domain-based array multiplication of matrix P and matrix Q, and a ≒ b means that a and b are approximately equal values.

At this time, the value of v (i) is determined according to the integer DCT transformation matrix T. If v (i) is determined, the value of s (i) is also determined. In addition, the first predetermined value Shift Tab0 and the second predetermined value Shift Tab1 may be set according to the specification of the multiplier for implementing the first multiplier 230 and the second multiplier 250, respectively. As a value that can be set differently, it is preferably set differently according to the bit depth.

The operation of the inverse transform and inverse quantization unit 114 shown in FIG. 2 is the same as the inverse transform and inverse quantization unit 614 shown in FIG. 7 to be described later.

As described above, according to the first embodiment of the present invention, a hardware specification of an existing encoding apparatus, for example, an arithmetic logic unit (ALU) that supports 16 bit memory, 16 bit multiplier, and 32 bit operation It is possible to encode raw data with high fidelity of N bits (10 bits, 12 bits or even 14 bits) by using only the predetermined coefficients and changing only predetermined coefficients, and at the same time, the coding efficiency can be improved.

However, according to the first embodiment, although the N-bit image data encoding apparatus can encode image data having various bit depths, it is transmitted as the bit depths of the data vary under the same quantization parameter (QP) conditions. There is still a problem that the picture quality represented by the bit rate and PSNR cannot be kept constant. For example, when encoding 8 bit image data, the bit rate is 10 Mbps and PSNR is 35 dB at QP = 0, whereas when encoding 10 bit image data, the bit rate is 20 Mbps and QNR is 45 dB at QP = 0. Therefore, even when encoding 10-bit image data, an appropriate QP must be set in order to obtain a 10Mbps bit rate and a PSNR of 35dB, which causes inconvenience because it requires a large amount of computation.

Hereinafter, a second embodiment of the present invention will be described in detail with reference to FIGS. 4 and 5.

According to the second embodiment of the present invention, the extension of the quantization parameter QP is performed along with the extension of the bit depth N bit. That is, the range of the quantization parameter is extended in proportion to the bit depth N. For example, it may be extended as follows.

QP '= QP + (bit depth N-8) * 8,

Here, QP is a conventionally used quantization parameter, and QP 'means a newly extended quantization parameter. According to an embodiment of the present invention,

For 8-bit video data, QP '= QP + (8-8) * 8 = [0, 63],

For 10bit image data, QP '= QP + (10-8) * 8 = [0, 79],

In the case of 12-bit image data, it may have a range of QP '= QP + (12-8) * 8 = [0, 95].

4 is a block diagram showing a more detailed configuration of the transform and quantization unit 413 and the inverse transform and quantization unit 414 according to the second embodiment of the present invention, wherein the transform and quantization unit 413 is a DCT transform unit. 210, a first divider 220, a first multiplier 230, a second divider 240, a second multiplier 450, and a shifting unit 460, and an inverse transform and quantization unit 414. ) Includes a third multiplier 470, a third divider 480, and an IDCT converter 290.

FIG. 5 is a flowchart for describing an operation of the transform and quantization unit 413 illustrated in FIG. 4. Hereinafter, the transformation and quantization processes will be described in detail with reference to FIGS. 4 and 5. Except for step 560 of FIG. 5, steps 500 to 550 and step 570 are the same as those of FIG. 3, and will be described briefly.

In operation 500, the transform and quantization unit 413 receives information about a bit depth N of an input image matrix.

In operation 510, the first adder 112 generates a difference matrix X by prediction of an input image matrix.

In operation 520, the DCT converter 410 generates a first matrix B by converting the difference matrix X input from the first adder 112 into an integer DCT using Equation 1 below.

In operation 530, the first divider 420 divides the converted first matrix B into a first predetermined value Shift Tab0 determined according to the bit depth N of the input image data, and divides the quotient. A second matrix C is generated by rounding to the nearest integer value.

In operation 540, the first multiplier 430 uses the second matrix C as a domain-based array multiplication with a predetermined matrix S having a predetermined coefficient s (i) as an element. D) is generated, and the third matrix D is transmitted to the third divider 440.

In operation 550, the second division unit 440 divides the third matrix D by the second predetermined value Shift Tab1 determined according to the bit depth N of the input image data, and divides the quotient of the nearest integer value. Approximation is performed to generate a fourth matrix (E).

In operation 560, the second multiplier 250 multiplies the fourth matrix E by a predetermined coefficient q [QP ′] to generate a fifth matrix F.

Here, QP 'is an integer of 0≤QP'≤63 + (bit depth N-8) * 8,

 When N = 10, q [QP '] is as follows.

q [80] = {32768,29775,27554,25268,23170,21247,19369,17770,

16302,15024,13777,12634,11626,10624,9742,8958,

8192,7512,6889,6305,5793,5303,4878,4467,

4091,3756,3444,3161,2894,2654,2435,2235,

2048,1878,1722,1579,1449,1329,1218,1117,

1024,939,861,790,724,664,609,558,

512,470,430,395,362,332,304,279,

256,235,215,197,181,166,152,140,

128, 117, 108, 99, 91, 83, 76, 70,

64, 59, 54, 49, 45, 41, 38, 35}

In operation 570, the shift unit 460 adds a predetermined value k to the fifth matrix F, and shifts the result to the right by 15 bits to generate the sixth matrix G. FIG.

The process of the first embodiment is represented by the following equation.

A = T

B = AT ^T

C = B // Shift Tab 0

D = S (i). * C

E = D // Shift Tab1

F = q [QP ']. E

G = (F + k) >> 15 (Intra: k = (1 << 15) * 10/31; Inter: k = (1 << 15) * 10/62)

Here, X is an 8x8 matrix of original data, T is an integer DCT transformation matrix, and T ^T is a transpose matrix of the integer DCT transformation matrix T.

The range of the extended quantization parameter QP 'is 0≤QP'≤63 + 8 * (N-8),

N is the bit depth N of the video data.

/v(i)이고(i는 0≤i≤5인 정수), s (i) ≒

/ v (i) (i is an integer of 0≤i≤5),

v = [512 × 512, 442 × 442, 464 × 454, 512 × 442, 512 × 464, 442 × 464],

/

이고,q [QP '] ≒

Of

ego,

이다.q [QP '] * r [QP'] ≒

to be.

Where a // b divides a by b, approximates the quotient divided by the nearest integer, PQ is a matrix multiplication of matrix P and matrix Q, aP is a scalar or array product of a and matrix P (array multiplication), P. * Q, is a domain-based array multiplication of matrix P and matrix Q, and a ≒ b means that a and b are approximately equal values.

At this time, the value of v (i) is determined according to the integer DCT transformation matrix T. If v (i) is determined, the value of s (i) is also determined. In addition, the first predetermined value Shift Tab0 and the second predetermined value Shift Tab1 may be set differently according to the specification of the multiplier for implementing the first multiplier 430 and the second multiplier 450, respectively. It is preferable that the value be set differently according to the bit depth.

As described above, according to the second embodiment of the present invention, even when image data having different bit depths is encoded, a constant bit rate and PSNR may be obtained for the same quantization parameter. For example, in 8-bit and 10-bit data coding, the bit rate is equal to 10 Mbps with respect to QP = 0, and the PSNR is also 35 dB. Therefore, in order to keep the bit rate and PSNR constant according to the bit depth, an appropriate QP must be calculated separately, so that system setting and the like can be prevented from becoming complicated. In addition, it is possible to obtain an image quality that is invariant to the QP regardless of the bit depth.

In the case where the QP is extended, the QP for the chroma block is also mapped with the extended Luma QP range. The mapping table specified in the conventional image data encoding apparatus does not change and may be maintained.

6 is a block diagram illustrating a configuration of an apparatus for decoding N-bit image data according to an embodiment of the present invention. The decoding control unit 620, the entropy decoding unit 630, and the inverse transform and inverse quantization unit 614 are illustrated. And a third adder 611, a deblocking unit 615, a memory 616, an intra-frame predictor 617, and an inter-frame predictor 618.

The entropy decoder 630 decodes the input bitstream to restore the control signal, the motion data, and the quantized data, and generates the control signal to the decoding controller 620 and the motion data to the interframe predictor 618. Are output to the inverse transform and inverse quantizer 614, respectively. The control signal includes information on a bit depth indicating the precision of the video data included in the input bitstream.

The decoding control unit 620 receives a control signal and controls the overall operation of the decoding apparatus.

The inverse transform and inverse quantizer 614 restores the differential data X by performing inverse integer DCT conversion on the quantized data received from the entropy decoder 630 and converts the restored difference data X 'into a third adder ( 611).

The third adder 611 adds the reconstructed differential data X 'and the predictive data predicted by the intra-frame predictor 617 or the inter-frame predictor 618, generates image data in units of blocks, and as a result, Is output to the deblocking unit 615.

The deblocking unit 615 performs filtering on each reconstructed macroblock to reduce block distortion, and outputs each result of the filtering to the memory 616.

The memory 616 stores the data output from the deblocking unit 610 as reference data for later use by the intra-frame predictor 617 or the inter-frame predictor 618.

The intraframe predictor 617 performs intraframe prediction, and the interframe predictor 618 performs interframe prediction. These predictions are also called motion compensation. The inter-frame prediction unit 617 extracts a prediction macro block corresponding to the motion vector of the reference frame of the memory 616, compensates for the motion of the prediction macro block of the reference frame, and outputs prediction data.

7 is a block diagram illustrating a more detailed configuration of the inverse transform and inverse quantization unit 614 shown in FIG. 7 according to an embodiment of the present invention. Inverse transform and inverse quantization unit 614 is a third multiplier. An inverse quantizer and an inverse transform unit 790 including a 770 and a third divider 780.

8 is a flowchart illustrating an inverse transform and inverse quantization process according to an embodiment of the present invention, and includes an inverse quantization process of steps 800 and 810 and an inverse transform process of step 830.

Hereinafter, a detailed operation of the inverse transform and inverse quantization unit 614 will be described with reference to FIGS. 7 and 8.

In operation 800, the input bitstream is decoded to restore quantized data, that is, the sixth matrix G.

In operation 810, the third multiplier 770 generates a seventh matrix H by multiplying r [QP] by the sixth matrix G received from the entropy decoder 630. Operation in step 810 may be expressed as Equation 6 below.

H = r [QP] .G

Here, the quantization parameter QP is an integer between 0 ≦ QP ≦ 63,

이고,q [QP] * r [QP] =

ego,

"." Means array product.

When q [QP] is the value mentioned in Equation 5 above, r [QP] has the following value.

r [64] = {32768,36061,38968,42495,46341,50535,55437,60424,

32932,35734,38968,42495,46177,50535,55109,59933,

65535,35734,38968,42577,46341,50617,55027,60097,

32809,35734,38968,42454,46382,50576,55109,60056,

65535,35734,38968,42495,46320,50515,55109,60076,

65535,35744,38968,42495,46341,50535,55099,60087,

65535,35734,38973,42500,46341,50535,55109,60097,

32771,35734,38965,42497,46341,50535,55109,60099}

n [64] = {14,14,14,14,14,14,14,14,

13,13,13,13,13,13,13,13,

12,12,12,12,12,12,12,12,

11,11,11,11,11,11,11,11,

10,10,10,10,10,10,10,10,

9,9,9,9,9,9,9,9,

8,8,8,8,8,8,8,8,

7,7,7,7,7,7,7,7}

In operation 820, the third divider 780 divides the seventh matrix H by n [QP], approximates the divided quotient to the nearest integer value, and generates the eighth matrix I. FIG. Operation 820 may be expressed as Equation 7 below.

I = H // n [QP]

 Here, "//" is the same as in Equation 2.

n [QP] is equal to the value mentioned in Equation 6.

In operation 830, the inverse DCT converter 790 generates an twelfth matrix M by performing inverse DCT transformation on the eighth matrix I. According to the third embodiment of the present invention, a process of inverse DCT conversion of the eighth matrix I is as follows.

In operation 832, the ninth matrix J is generated by multiplying the eighth matrix I by the transformation matrix T.

J = IT

Here, T is an integer DCT conversion matrix used in the DCT converter 210 of FIG. 2.

In operation 834, the ninth matrix J is divided by three, and the dividing quotient is approximated to the nearest integer value to generate a tenth matrix K. FIG.

K = J // 3

Here, "//" is the same as Equation 2.

In operation 836, an eleventh matrix L is generated by multiplying the ninth matrix K by T ^T. Operation of step 636 may be expressed as Equation 10 below.

L = ^T T · K

Where T ^T is the transpose of the integer DCT transformation matrix T.

In operation 838, the eleventh matrix L is divided by 7, and the quotient divided by the nearest integer value is generated to generate the twelfth matrix M, that is, the restored difference matrix X '. The operation of step 638 may be expressed as the following equation.

M = L // 7

The process in the third embodiment is summarized by the following equation.

H = r [QP] G

I = H // n [QP]

J = IT

K = J // 3

L = ^T T · K

M = L // 7

Here, G is quantized data, that is, a sixth matrix G, T is an integer DCT transform matrix, and T ^T is a transpose matrix of the integer DCT transform matrix T. Here, a // b is an operation of dividing a by b, approximating the quotient divided by the nearest integer value, and P · Q is a matrix multiplication of the matrix P and the matrix Q.

As described above, according to the third embodiment of the present invention, a hardware specification of an existing encoding apparatus, for example, an arithmetic logic unit (ALU) supporting 16 bit memory, 16 bit multiplier, and 32 bit operation; Arithmetic Logic Unit) may be used as is, and only predetermined coefficients may be changed to decode raw data having high fidelity of N bits (10 bits, 12 bits or even 14 bits). However, according to the third embodiment, although the N-bit image data encoding apparatus can encode image data having various bit depths, the bit rate of the data varies according to the same quantization parameter (QP) conditions. There is still a problem in that the image quality represented by (bit rate) vs. PSNR cannot be kept constant. For example, when encoding 8 bit image data, the bit rate is 10 Mbps and PSNR is 35 dB at QP = 0, whereas when encoding 10 bit image data, the bit rate is 20 Mbps and QNR is 45 dB at QP = 0. Therefore, even when encoding 10-bit image data, an appropriate QP must be set in order to obtain a 10Mbps bit rate and a PSNR of 35dB, which causes inconvenience because it requires a large amount of computation.

Hereinafter, a fourth embodiment of the present invention will be described in detail with reference to FIGS. 9 and 10.

According to the second embodiment of the present invention, the extension of the quantization parameter QP is performed along with the extension of the bit depth N bit. That is, the range of the quantization parameter is extended in proportion to the bit depth N. For example, it may be extended as follows.

QP '= QP + (bit depth N-8) * 8,

Here, QP is a conventionally used quantization parameter, and QP 'means a newly extended quantization parameter.

For 8 bit data, QP '= QP + (8-8) * 8 = [0, 63],

For raw data with 10 bits, QP '= QP + (10-8) * 8 = [0, 79],

In the case of raw data with 12 bits, it may have a range of QP '= QP + (12-8) * 8 = [0, 95].

9 is a block diagram illustrating a more detailed configuration of the inverse transform and quantization unit 914 according to the fourth embodiment of the present invention. The inverse transform and quantization unit 914 includes a third multiplier 970 and a third unit. A divider 980 and an inverse DCT converter 990.

FIG. 10 is a flowchart for describing an operation of the inverse transform and quantization unit 941 illustrated in FIG. 9. Hereinafter, an operation of the inverse transform and quantization unit 914 will be described with reference to FIGS. 9 and 10. Steps 1000 to 1030 are similar to steps 800 to 830 shown in FIG. 8, and thus will be described briefly.

In operation 1000, the input bitstream is decoded to restore quantized data, that is, the sixth matrix G.

In operation 1010, the third multiplier 970 generates a seventh matrix H by multiplying r [QP ′] by a sixth matrix G received from the entropy decoder 630.

H = r [QP ']. G

Here, QP 'is an integer of 0≤QP'≤63 + (bit depth N-8) * 8,

이고,q [QP] * r [QP] =

ego,

"." Means array product.

When n = 10, if q [QP '] is the same as in the second embodiment, r [QP'] and n [QP '] are as follows.

r [80] = {32768,36061,38968,42495,46341,50535,55437,60424,

32932,35734,38968,42495,46177,50535,55109,59933,

65535,35734,38968,42577,46341,50617,55027,60097,

32809,35734,38968,42454,46382,50576,55109,60056,

65535,35734,38968,42495,46320,50515,55109,60076,

65535,35744,38968,42495,46341,50535,55099,60087,

65535,35734,38973,42500,46341,50535,55109,60097,

32771,35734,38965,42497,46341,50535,55109,60099,

65535,35852,38832,42380,46086,50535,55193,59933,

65535,35541,38843,42795,46603,51160,55182,59933}

n [80] = {14,14,14,14,14,14,14,14,

13,13,13,13,13,13,13,13,

12,12,12,12,12,12,12,12,

11,11,11,11,11,11,11,11,

10,10,10,10,10,10,10,10,

9,9,9,9,9,9,9,9,

8,8,8,8,8,8,8,8,

7,7,7,7,7,7,7,7,7,

6,6,6,6,6,6,6,6,

5,5,5,5,5,5,5,5}

In operation 1020, the third divider 780 generates the eighth matrix I by dividing the seventh matrix H by n [QP ′] and approximating the divided quotient to the nearest integer value.

I = H // n [QP ']

 Here, "//" is the same as in Equation 2.

n [QP '] is the same as the value mentioned in step 1010.

In operation 1030, the inverse DCT converter 790 generates an twelfth matrix M by performing inverse DCT transformation on the eighth matrix I.

The entire process of the fourth embodiment is summarized as follows.

H = r [QP '] G

I = H // n [QP ']

J = IT

K = J // 3

L = ^T T · K

M = L // 7

Here, G is quantized data, that is, a sixth matrix G, T is an integer DCT transform matrix, and T ^T is a transpose matrix of the integer DCT transform matrix T. Here, a // b is an operation of dividing a by b, approximating the quotient divided by the nearest integer value, and P · Q is a matrix multiplication of the matrix P and the matrix Q.

As described above, according to the fourth embodiment of the present invention, even when image data having different bit depths is encoded, a constant bit rate and PSNR may be obtained for the same quantization parameter QP. For example, in 8-bit and 10-bit data coding, the bit rate is equal to 10 Mbps with respect to QP = 0, and the PSNR is also 35 dB. Therefore, in order to keep the bit rate and PSNR constant according to the bit depth, an appropriate QP must be calculated separately, so that system setting and the like can be prevented from becoming complicated. In addition, it is possible to obtain an image quality that is invariant to the QP regardless of the bit depth.

In the case where the QP is extended, the QP for the chroma block is also mapped with the extended Luma QP range. The mapping table specified in the conventional image data encoding apparatus does not change and may be maintained.

FIG. 11 is a graph illustrating a relationship between PSNR and a bit rate when only the bit depth is extended without expanding the quantization parameter according to the third embodiment of the present invention, and FIGS. 12A and 12B are a fourth embodiment of the present invention. Is a graph showing the relationship between the PSNR and the bit rate in the case where the quantization parameter is extended in proportion to the bit depth and the bit depth is also extended. Referring to FIG. 11, in the case of encoding N-bit image data according to the present invention, encoding efficiency is improved as compared with the prior art, and it can be seen that the improvement ratio is increased as the PSNR is higher. 12A and 12B, the same quantization parameter is extended since the range of the quantization parameter QP is extended in proportion to the N bit extension and the image data is processed based on the quantization step corresponding to the extended quantization parameter QP. In the case of (QP), it can be confirmed that the same bit rate and PSNR can be maintained even if the bit depth is changed.

Although a preferred embodiment of the present invention has been described in detail above, those skilled in the art to which the present invention pertains can make various changes without departing from the spirit and scope of the invention as defined in the appended claims. It will be appreciated that modifications or variations may be made. Accordingly, modifications of the embodiments of the present invention will not depart from the scope of the present invention.

The N-bit image data encoding / decoding apparatus according to the present invention is a coding apparatus according to the Audio and Video Coding Standard (AVS), which is a modification of a coding / decoding apparatus and method capable of processing only 8-bit image data. An encoding / decoding apparatus and method capable of processing N bit image data having various bit depths can be provided, and encoding efficiency can be improved.

In addition, since the range of the quantization parameter QP is extended in proportion to the N bit extension, and the image data is processed based on the quantization step corresponding to the extended quantization parameter QP, the bit depth in the case of the same quantization parameter QP The same bit rate and PSNR can be maintained even if.

Claims (21)

(a) generating difference data from prediction data of the image data, which is removed from the input image data; (b) converting the generated difference data into a frequency domain; And (c) quantizing the transformed differential data based on a quantization parameter QP 'whose range is extended in proportion to a bit depth N representing the precision of the image data, Step (c) is And quantizing the transformed differential data by performing a predetermined operation using predetermined coefficients determined according to the bit depth (N). The method of claim 1, wherein the extended quantization parameter QP 'range is And an extended quantization parameter, which is extended in proportion to the bit depth (N). 2. The method of claim 1, wherein step (c) And performing quantization of the transformed differential data by performing a predetermined operation based on a quantization step corresponding to any one of the extended quantization parameter QP '. Way. N bit image data encoding method. The method of claim 3, wherein step (c) (c1) dividing the transformed difference data into a first predetermined value; (c2) multiplying a result of dividing by the first predetermined value by a predetermined coefficient s; (c3) dividing a result of multiplying the predetermined coefficient s by a second predetermined value; (c4) multiplying a result of dividing by the second predetermined value by a predetermined coefficient q; And (c5) adding a predetermined value k to a result of multiplying the predetermined coefficient q, and shifting the result of adding the predetermined value k to the right of a predetermined bit, The first and second predetermined values are values determined corresponding to the bit depths, and steps (c1) and (c3) are integer operations. The method of claim 4, wherein the coefficient q is q [QP '] = round [

Of

Video data encoding method having an extended quantization parameter, wherein QP 'is an integer of 0≤QP'≤63 + 8 * (N-8), and round [X] is the X and the Nearest integer) The method of claim 1, wherein the encoding method and (d) encoding the information about the bit depth (N) and the quantized result to generate and output a bitstream. A difference data generator for generating difference data from which prediction data of the input image data is removed from the input image data; A converter for converting the generated difference data into a frequency domain; And A quantizer configured to quantize the transformed differential data based on a quantization parameter QP 'whose range is extended in proportion to a bit depth N representing the precision of the image data, The quantization unit quantizes the transformed differential data by performing a predetermined operation using predetermined coefficients determined according to the bit depth (N). 8. The method of claim 7, wherein the range of extended quantization parameter QP 'is N bit image data encoding apparatus, characterized in that it is extended in proportion to the bit depth (N). The method of claim 7, wherein the quantization unit And performing a predetermined operation based on a quantization step corresponding to any one of the extended quantization parameter (QP '), thereby quantizing the transformed differential data. The method of claim 9, wherein the quantization unit A first divider dividing the converted difference data into a first predetermined value; A first multiplier for multiplying a result of dividing by the first predetermined value by a predetermined coefficient s; A second divider dividing a result of multiplying the predetermined coefficient s by a second predetermined value; A second multiplier for multiplying a result of dividing by the second predetermined value by a predetermined coefficient q; And A shifting unit for adding a constant k to a result of multiplying the predetermined coefficient q and shifting the result of adding the constant k to the right of a predetermined bit, The first and second predetermined values are values determined corresponding to the bit depth N, And the operation performed by the first and second dividers is an integer operation. The method of claim 10, wherein the coefficient q is q [QP '] = round [

Of

N bit image data encoding apparatus, wherein QP 'is an integer of 0≤QP'≤63 + 8 * (N-8), and round [X] is an integer closest to X. The method of claim 1, wherein the encoding device And an entropy encoder configured to generate and output a bitstream by encoding the information about the bit depth (N) and the quantized result. (a) decoding the input bitstream and restoring quantization data and information about a bit depth N representing the precision of the image data included in the bitstream; (b) dequantizing the reconstructed quantized data based on a quantization parameter QP 'whose range is extended in proportion to the bit depth N; And (c) inversely transforming the inverse quantized result into a time domain to restore differential data from which prediction data of the image data is removed from the image data; Step (b) is (b1) multiplying the reconstructed quantized data by a predetermined coefficient r; And (b2) dividing the result of multiplying the predetermined coefficient r by a predetermined coefficient n. The method of claim 13, wherein step (b) And dequantizing the reconstructed quantized data based on a quantization step corresponding to any one of the extended quantization parameter (QP ′). delete The method of claim 13, wherein the coefficient r is r [QP '] = round [

]ego, Wherein the coefficient n is n [QP '] = 14-[QP' / 8], where QP is 0≤QP'≤63 + 8 * (N-8) Integer, round [X] is the closest integer to X, and [X] is the largest integer less than or equal to X). An entropy decoder configured to decode the input bitstream to restore information about a bit depth N representing the precision of the image data included in the bitstream and quantized data; An inverse quantizer for inversely quantizing the reconstructed quantized data based on a quantization parameter QP 'whose range is extended in proportion to the bit depth N; And An inverse transform unit configured to inversely transform the inverse quantized result into a time domain to restore difference data from which prediction data of the image data is removed from the image data, The inverse quantization unit A third multiplier that multiplies the reconstructed quantized data by a predetermined coefficient r; And And a third divider for dividing a result of multiplying the predetermined coefficient r by a predetermined coefficient n. The method of claim 17, wherein the inverse quantization unit And dequantizing the reconstructed quantized data based on a quantization step corresponding to any one of the extended quantization parameter (QP '). delete A computer-readable recording medium having recorded thereon a program for executing the method of any one of claims 1 to 6. A computer-readable recording medium having recorded thereon a program for executing the method of any one of claims 13 to 14 or 16.

Images