KR100413979B1 - Predictive coding method and device thereof - Google Patents

Abstract

PURPOSE: A predictive coding method and a device thereof are provided to perform a coding process by using a spatial redundancy of a neighboring quantized block when coding an intra block by a DCT(Discrete Cosine Transform) method, thereby reducing the number of bits used for DC coefficients. CONSTITUTION: A motion estimation portion(11) estimates an object motion. A motion compensation portion(12) compensates the motion. A subtractor(13) subtracts a compensation value from an inputted image sequence. A texture coding portion(14) codes texture information. An adder(15) adds output signals. A previous image memory(16) stores a detected previous image, and enables to estimate the object motion. A quantizer(17) quantizes an output signal of the coding portion(14). An intra block DC predictor(18) predicts a DC value of an intra block. A DPCM(Differential Pulse Code Modulation) portion(19) detects a difference value between the previous image and a current image. A multiplexer(20) multiplexes output signals of the quantizer(17) and the DPCN portion(19). A buffer(21) temporarily stores an output signal of the multiplexer(20).

Description

Prediction Coding Method and Apparatus

The present invention relates to a prediction encoding method and apparatus for encoding an intra block of a block-oriented coding scheme based on a discrete cosine transform (DCT).

Standardized methods such as Moving Picture Experts Group (MPEG-1), MPEG-2, Joint Photographic Coding Experts Group (JPEG), H261, and H263 using block-oriented coding are 8 × for intrablocks that do not predict in the time direction. DCT of 8 blocks is performed and DC (Discrete Cosine) coefficient DCT is performed by PCM conversion or transmitted by DPCM (Differential Pulse Code Modulation) using DC value of one neighbor block already transmitted.

However, the DC coefficient occupies a considerable portion of the total DCT coefficients, which requires a lot of bits in the case of simply transmitting to the PCM, and the transmission by DPCM cannot obtain a high prediction gain considering only one neighboring block. .

Accordingly, an object of the present invention is to provide a prediction encoding method and apparatus for predicting and encoding a DC value using three other blocks neighboring a current block.

In order to achieve the above object, the predictive encoding method and apparatus of the present invention set three neighboring blocks with respect to a current block to predict a DC value, and perform predictive encoding by using the configured three blocks.

Therefore, according to the present invention, when DCT encoding an intra block without time direction prediction, the number of bits required for DC coefficients can be reduced by performing encoding using spatial redundancy of neighboring quantized blocks. Improve the quality of transmission.

1 is a circuit diagram showing the configuration of a predictive encoding device of the present invention.

2 is a diagram showing a configuration state of a macro block.

3 and 4 are diagrams showing neighboring blocks used in predictive coding according to the present invention.

(A) (b) of FIG. 5 is a diagram showing the DC value setting for the prediction in the case of being located outside the edge or shape information of the VOP and the intra block in the method 1 of the present invention.

(A) (b) of FIG. 6 sets DC values for prediction when three blocks are located outside the boundary or shape information of the VOP and two blocks are intra blocks in the method 2 of the present invention. Shown.

(A) and (b) of FIG. 7 show setting of DC values of two other blocks when one block is an intra block in the method 3 of the present invention.

(A) and (b) of FIG. 8 show setting of DC values of two other blocks when one block is an inter block in the method 4 of the present invention.

9 is a view showing the setting of a quantized DC value of a block existing at the boundary of shape information among three blocks in the method 5 of the present invention.

FIG. 10 is a diagram illustrating setting of quantized DC values of one block existing outside the boundary or shape information of a VOP among three blocks according to the method of the present invention. FIG.

FIG. 11 is a diagram illustrating setting of a quantized DC value of one block existing outside of boundary or shape information of a VOP among three blocks according to the method of the present invention. FIG.

* Explanation of symbols for main parts of the drawings

11: motion estimation unit, 12: motion compensator, 13: subtractor, 14: internal information encoder, 15: adder, 16: previous image memory, 17: quantizer, 18: intra block DC predictor, 19: DPCM unit , 20: multiplexer, 21: buffer

Hereinafter, a prediction encoding method and apparatus of the present invention will be described in detail with reference to the accompanying drawings.

1 is a circuit diagram showing the configuration of a predictive encoding device of the present invention. As shown therein, a motion estimation unit 11 for estimating the motion of the object based on the input image sequence and / or shape information of the object, and the motion estimation unit 11 A motion compensation unit 12 for compensating the movement of an object with motion and / or the shape information, and a subtractor 13 for subtracting a compensation value of the motion compensation unit 12 from the input image sequence. And an internal information coding unit 14 for encoding internal information using the output signal of the subtractor 13 and / or the shape information of the object, and the motion compensating unit 12 and the internal information encoding unit ( An adder 15 that adds an output signal of 14) and a previous image that detects and stores a previous image of the object as an output signal of the adder 15, and outputs to the motion estimator 11 to estimate the motion of the object. Image memory (16 ), The quantization unit 17 for quantizing the output signal of the internal information encoder 14, the shape information of the input object, and the output signals of the previous image memory 16 and the quantization unit 17. An intra block DC predictor 18 for predicting a DC value of the signal, a DPCN unit 19 for detecting a difference value between a previous image and a current image as an output signal of the intra block DC predictor 18, and the quantizer (17) and a multiplexer 20 for multiplexing the output signal of the DPCN unit 19, the multiplexer 20, and a buffer 21 for temporarily storing and outputting the output signal of the multiplexer 20.

In the predictive encoding apparatus of the present invention configured as described above, the motion estimator 11 estimates the movement of the object based on the input image sequence and / or the shape information of the object, and moves the estimated motion and / or the shape information. The compensator 12 compensates for the movement of the object and outputs it.

The subtractor 13 subtracts the output signal of the motion compensator 12 from the input image sequence, and the internal information encoder 14 internally outputs the output signal of the subtractor 13 or the shape information of the input object. Performs information coding and outputs it.

The output signal of the internal information encoder 14 is input to the adder 15 and added to the output signal of the motion compensator 12. The output image of the adder 15 causes the previous image memory 16 to transmit the previous image of the object. Is detected, stored, and outputted to the motion estimator 11 for use in estimating the motion of the object.

The output signal of the internal information encoder 14 is input to the quantization unit 17 and quantized, and an intra block is formed by the shape information of the input object and the output signals of the previous image memory 16 and the quantization unit 17. The DC predictor 18 predicts the DC value, and the DC value predicted by the intra block DC predictor 18 is input to the DPCN unit 19.

The DPCN unit 19 subtracts the image of the current object from the image of the previous object, and PCN outputs the subtracted value.

The output signal of the DPCN unit 19 and the output signal of the quantization unit 17 are multiplexed by the multiplexer 20 and output through the buffer 21.

That is, in the block-oriented coding scheme based on the DCT, the macroblock is composed of 16 pixels each horizontally and vertically as shown in FIG. 1, and is divided into four and composed of L1, L2, L3, and L4. It is divided into blocks of luminance components and blocks of two color components composed of C1 and C2.

Here, each pixel component of C1 and C2 corresponds to a position subsampled at 2: 1, respectively, horizontally and vertically of the luminance component block.

Here, a block using time direction prediction is called an inter block, and a block not using time direction prediction is called an intra block. In the case of an intra block, DCT is the above-mentioned six 8 × 8 blocks, that is, L1, It is performed on the brightness values in the L2, L3, L4, C1 and C2 blocks.

At this time, one DC coefficient per block is calculated. The calculated DC coefficient is quantized to 8 bits in the quantization unit 17 and predictively coded in the intra block DC prediction unit 18.

In the present invention, three blocks adjacent to the current block are used to encode DC coefficients.

That is, in FIG. 3, assuming that the L1 block is the current block B, B1 (L2 of the macro block located to the left of the current macro block), B2 (L3 of the macro block located above the current macro block), and B3 (current B1 (L1 of the current macro block), B2 (L4 of the macro block located on top of the current macro block), and assuming that L2 block is the current block B, B3 (L3 of the macro block located on top of the current macro block) is used, and B1 (L4 of the macro block located on the left side of the current macro block) and B2 (assuming that L3 block is the current block B) L1) and B3 (L2 of the current macro block) are used.

Assuming that the L4 block is the current block B, B1 (L3 of the current macro block), B2 (L1 of the current macro block) and B3 (L2 of the current macro block) are used as shown in FIG.

In addition, as shown in FIG. 3, when the C1 block is assumed to be the current block B, B1 (C2 of the macro block located to the left of the current macro block), B2 (C1 of the macro block located above the current macro block) and Using B3 (C2 of the macro block located on top of the current macro block), and assuming that C2 block is the current block B, B1 (C1 of the current macro block), B2 ( C2) and B3 (C1 of the macro block located on top of the current macro block).

In this way, the intra block DC predictor 18 determines the block to be used and calculates the DC predictive value as shown in Equation (I) below when all of the B1, B2 and B3 blocks are intra blocks.

Here, DC_P is a calculated DC prediction value, and DC_B1, DC_B2 and DC_B3 are DC coefficients of quantized luminance or color of B1, B2 and B3 blocks.

Here, in the case of color, it is a macro block of 16x16 size.

Assuming that the DC coefficient for the current B block is DC_, the predicted value to be transmitted is expressed by the following equation (II).

Here, DC_T is a prediction value to be transmitted.

By encoding and transmitting DC_T calculated by Equation (II), the decoder can calculate DC_B in the reverse process as in Equation (III) below.

When at least one of the three neighboring blocks is not an intra block, any one of the following seven methods is used.

Here, one selected method is determined in consideration of implementation complexity and encoding performance.

Method 1

As shown in (a) (b) of FIG. 5, the intra blocks among the B1, B2, and B3 blocks use quantized DC, and blocks located outside the image, intra blocks, and uncoded blocks (that is, The median filtering calculates DC_P using 128 as the quantized coefficient of the block which is very small and does not need to be coded as the same thing in the past.

For example, when B1 = 127, B2 = 155, and B3 = 120, DC_P is 12701.

Method 2

As shown in (a) (b) of FIG. 6, the intra block among the B1, B2, and B3 blocks uses quantized DC, the quantized DC coefficient of the block located outside the image is 128, and the intra block and The quantized coefficients of the uncoded blocks are obtained by directly calculating the values of the pixels that have already been handed down, and then the DCP is calculated by median filtering as shown in Equation (IV) below.

When the DC value of the inter block is R_DC_B in Equation (IV), R_DC_B is an average value of pixel values in the reproduced inter block, which is represented by Equation (V).

Here, N and M are the number of pixels in the shape information when the shape information is used, or the number of 64 pixels present in one block.

Method 3

Follow Method 1, wherein any one of B1, B2, and B3 blocks is located in the shape information, as shown in (a) (b) of FIG. 7, and the other two are quantized DCs of the intra block if it is an intra block. Replace DC in block and calculate DC_ with median filtering.

Here, DC_P is equal to the quantized DC of the intra block.

Method 4

Following Method 2, quantization of the block when any one of B1, B2, and B3 blocks is located in the shape information and is an intra block or an unencoded block, as shown in Fig. 8A. The DC of the other two blocks is replaced with the given DC, and then DC_P is calculated by median filtering.

Here, DC_P is equal to the quantized DC of an intra block or an uncoded block.

Method 5-

According to the above-described method 1 and method 2, the prediction is performed using two quantized DC values in which padding is performed on blocks located at the boundary of VOP or shape information among B1, B2, and B3 blocks in FIG. 9.

There are several paddings. For example, average replacement padding that fills an area where no image information exists with an average value of image information in a block, and zero padding that fills an area where no image information exists with a value of '0'. For example, repetitive padding that fills an area where image information does not exist with object boundary information in the horizontal and vertical directions and overlaps the padding with an average value or extends to the end of an area without image information in a block while preserving the spatial frequency in the image information. Extension interpolation padding to interpolate.

Method 6

Follow the method 1, method 2 and method 5 described above, and the image or shape information when any one of the blocks B1, B2 and B3 is located outside the image or shape information as shown in FIGS. 10 and 11 The quantized DC of the block located outside of is replaced with the quantized DC value of the block not located outside, and then selected as the quantized DC value for prediction of the block.

For example, in FIG. 10, the DC of B3 is replaced by the DC of the block located on the left, and in FIG. 11, the DC of B1 is replaced by the DC of the block located above B1.

Method 7

Follow Method 1, Method 2 and Method 5 above, but are not located outside when any one of blocks B1, B2 and B3 is located outside of the image or shape information as shown in FIGS. 10 and 11 DC_ is replaced with the average value of DC_B of two blocks.

In this case, the average may use rounding or truncation rather than an integer.

As described above, the present invention performs encoding by using spatial redundancy of quantized DC values of three neighboring blocks when DCT encoding an intra block that does not make prediction in the temporal direction. It can reduce the number, thereby contributing to the improvement of the image quality in the transmission of a fixed rate.

Claims (27)

Three luminance blocks and / or color blocks B1, B2, and B3 neighboring the current luminance block and / or the color block B to be predictively coded are set, and the DC predictive value is calculated using the three blocks. Predictive encoding method. 2. The method of claim 1, wherein the current block B is L1 which is a luminance block of the macro block, and in the case of the block, the luminance block L2 of the macro block located to the left of the current macro block, and the luminance block L3 of the macro block located above the current macro block. And calculating the DC prediction value by setting the luminance block L4 of the macroblock located above the current macroblock to B1, B2, and B3. The method of claim 1, wherein the current block B is L2 which is a luminance block of the macro block, and when the block is a luminance block L1 of the current macro block, a luminance block L4 of a macro block located above the current macro block, and an upper portion of the current macro block. And predicting the DC predictive value by setting the luminance block L3 of the macroblock located on the right side to B1, B2 and B3. 2. The method of claim 1, wherein when the current block B is an L3 block that is a luminance block of the macro block, the luminance block L4 of the macro block located on the left side of the current macro block, and the luminance blocks L1 and L2 of the current macro block are defined by B1, B2 and A prediction encoding method, characterized by setting to B3 to calculate the DC prediction value. The method of claim 1, wherein when the current block B is an L4 block that is a luminance block of the macro block, the DC prediction value is calculated by setting luminance blocks L3, L1, and L2 of the current macro block to B1, B2, and B3. Predictive encoding method. The method according to claim 1, wherein when the current block B is a C1 block that is a color block of the macro block, the color block C2 of the macro block located on the left side of the current macro block, and the color block C1 of the macro block located on the top of the current macro block; And calculating the DC prediction value by setting the color blocks C2 of the macroblock located above the current macroblock to B1, B2, and B3. The method of claim 1, wherein when the current block B is a C2 block that is a color block of the macro block, the color block C1 of the current macro block, the color block C2 of the macro block located above the current macro block, and the top of the current macro block. And predicting the DC prediction value by setting the color blocks C1 of the macroblock located on the right side to B1, B2, and B3. The prediction encoding method according to any one of claims 1 to 7, wherein the DC prediction value is calculated by the following equation (I) when all three neighboring blocks B1, B2, and B3 are intra blocks. . Here, DC_P is a calculated DC prediction value, and DC_B1, DC_B2, and DC_B3 are DC coefficients of quantized luminance or color of B1, B2, and B3 blocks. The quantized coefficients of any one of claims 1 to 7, wherein the intra blocks among the B1, B2, and B3 blocks use quantized DC, and the quantized coefficients of blocks located outside of the image, intra blocks, and unencoded blocks are 128. The prediction encoding method according to claim 1, wherein the DC prediction value is calculated by median filtering. 10. The method of claim 9, wherein among blocks B1, B2, and B3, a block having a boundary of shape information is padded and predicted by a quantized DC value. 12. The method of claim 10, wherein the padding is an average value padding that fills an area where the image information does not exist with the average value of the image information in the block. 12. The method of claim 10, wherein the padding performs zero padding that fills a region having no image information with a value of '0'. The prediction encoding method of claim 10, wherein the padding performs repetitive padding that fills an area in which no image information exists with the hardness information of the object in the horizontal and vertical directions and fills an overlapping portion with an average value. 12. The method of claim 10, wherein the padding performs extended interflation padding that extends and interpolates to the end of an area without image information in the block while preserving the spatial frequency in the image information. 10. The method of claim 9, wherein when one of the B1, B2, and B3 blocks is outside the VOP or shape information, the quantized DC of the outer block is replaced with the quantized DC value of the non-outside block. And then selecting the quantized DC value for prediction of the block. 10. The prediction encoding method according to claim 9, wherein when one of the blocks B1, B2, and B3 is outside the image or shape information, the DC prediction value is replaced by the DC_B average value of two blocks not located outside. Way. 8. The intrablock of claim 1, wherein the intra block among the B1, B2 and B3 blocks uses quantized DC, the quantized DC coefficient of the block located outside of the image is 128, and the intra block and The quantized coefficients of the unencoded block are predictive encoded by calculating median prediction by performing median filtering, as shown in Equation (IV) below. Way. Here, DC_ is a calculated DC prediction value, and DC_B1, DC_2 and DC_B3 are DC coefficients of quantized luminance or color of B1, B2 and B3 blocks. Here, R_DC_B is an average value of pixel values in the reproduced inter block, and N and M are the number of pixels in the shape information when the shape information is used or the number of 64 pixels present in one block. 18. The method of claim 17, wherein when one of the blocks B1, B2, and B3 is located in the shape information and is an inter block or an unencoded block, the DC of the other two blocks is replaced by the quantized DC of the one block. A method of predictive encoding, characterized by calculating median DC prediction by performing median filtering. 18. The predictive encoding method of claim 17, wherein a block having a boundary of shape information among B1, B2, and B3 blocks is padded and then predicted by a quantized DC value. 20. The method of claim 19, wherein the padding performs average value padding that fills an area in which no image information exists with an average value of image information in a block. 20. The method of claim 19, wherein the padding performs zero padding to fill a region where image information does not exist with a value of '0'. 20. The method of claim 19, wherein the padding performs repetitive padding that fills an area in which no image information exists with the hardness information of the object in a horizontal and vertical direction and fills an overlapping portion with an average value. 20. The method of claim 19, wherein the padding performs extended interflation padding that extends and interpolates to an end of an area without image information in the block while preserving the spatial frequency in the image information. 18. The method of claim 17, wherein when one of the blocks B1, B2, and B3 is outside the VOP or shape information, the quantized DC of the outer block is replaced with the quantized DC value of the block not sitting outside. And then selecting the quantized DC value for prediction of the block. 18. The prediction encoding method according to claim 17, wherein when one of the blocks B1, B2, and B3 is outside the image or shape information, the DC prediction value is replaced by the DC_B average value of two blocks that are not located outside. Way. The DC according to any one of claims 1 to 7, wherein any one of the B1, B2, and B3 blocks is located in shape information and is an intra block, the DC of the other two blocks is quantized by the intra block. Predicting encoding method for calculating a DC prediction value by median filtering after replacing. A motion estimator 11 for estimating the motion of the object based on the input image sequence and / or shape information of the object, and the motion and / or the shape information of the object estimated by the motion estimator 11 A motion compensator 12 for compensating for motion, a subtractor 13 for subtracting a compensation value of the motion compensator 12 from the input image sequence, an output signal of the subtractor 13 and / or the object An internal information encoding unit 14 for encoding internal information with shape information of an image, an adder 15 for adding output signals of the motion compensation unit 12 and the internal information encoding unit 14, and the adder 15 The output signal of the previous image memory 16 and the internal information encoder 14 to detect and store the previous image of the object, and output to the motion estimator 11 to estimate the motion of the object. Quantization An intra block DC predictor 18 for predicting a DC value of an intra block based on an input signal of the object 17, shape information of an object to be input, and output signals of the previous image memory 16 and the quantization unit 17, and the intra block The DPCN unit 19 which detects a difference value between the previous image and the current image as the output signal of the block DC predictor 18, and the multiplexer 20 which multiplexes the output signals of the quantizer 17 and the DPCN unit 19. And a multiplexer (20) and a buffer (21) for temporarily storing and outputting the output signal of the multiplexer (20).