KR100655546B1 - Method of predicting motion vector - Google Patents

Abstract

The present invention relates to a video coding method in which coding efficiency is improved through an inter mode using a long term reference picture and a direct mode of a B picture in a video coding system.

According to the present invention, the P picture is coded in the inter mode with reference to the long-term reference picture depending on whether a scene is changed. In addition, the B picture is calculated and coded in the B picture according to the type of the reference buffer indicated by the motion vector of the same position block of the specific picture.

Motion Picture, P Picture, B Picture, Long Term Reference Picture, Direct Mode, Motion Vector

Description

Motion vector prediction method {METHOD OF PREDICTING MOTION VECTOR}

Fig. 1 shows the order in which each picture is displayed when using two B pictures.

2 illustrates a sequence of coding a digital image using a B picture in a video coding system.

3 is a flowchart illustrating a method of coding a video sequence in a video system according to a preferred embodiment of the present invention.

4 is an exemplary diagram illustrating a method of coding a moving picture sequence having a scene transformation according to a preferred embodiment of the present invention.

5 is an exemplary diagram illustrating a method of coding a B picture in a direct mode according to a preferred embodiment of the present invention.

TECHNICAL FIELD The present invention relates to a video coding system, and in particular to motion vectors of the bi-predictive image block based on the type of reference picture associated with the image block in the same position relative to the bi-predictive image block. The present invention relates to a motion vector prediction method that improves coding efficiency by determining.

In order to optimally compress and code a moving picture picture sequence, it is desirable to be able to detect a scene change in the sequence. This is because close-up conversation scenes such as news, sports broadcasts, interviews, or video images such as multi-point video conferences include repetitive scene transformations. Such scene transformation may occur throughout the picture, or may occur in some areas within the picture.

Each time a scene change is detected, the coding method of the digital image is changed accordingly. For example, since the similarity is very low between the picture where the transition has occurred and the picture of the previous scene, the picture of the new scene is compared with the inter mode coding through motion compensation from the reference picture. Intra picture, which is coded through prediction in a picture, is selected and coding is performed.

In more detail, a picture in which a scene transformation occurs in the entire picture becomes an intra picture that is coded in the intra mode in every block, and a picture in which a scene change occurs in some regions of the picture indicates that all blocks in the area in which the scene change has occurred are in intra mode. Is coded. Since the intra mode generates a higher bit amount than the inter mode, a sequence in which scene change occurs frequently has a fatal problem in a low bit rate application.

In general, when using a B picture in a video coding system, the coding order is different from the display order.

1 shows the order in which each picture is displayed when using two B pictures. As shown in FIG. 1, when the intra picture I is displayed for the first time among pictures displayed in time, two B pictures B1 and B2 are sequentially displayed next to the intra picture I. After the above B pictures B1 and B2 are displayed, the P picture P3 is displayed third. In the same manner as above, the following steps are also made. That is, when the third P picture P3 is displayed, the fourth and fifth B pictures B4 and B5 are displayed, and then the P picture P6 is displayed.

However, the order of coding the digital image may not be the same as the display order. In other words, the P picture is coded in preference to the B picture.

2 illustrates a sequence of coding a digital image using a B picture in a video coding system. As shown in Fig. 2, when the intra picture I is coded, the P picture P3 is coded before the two displayed B pictures B1 and B2. Subsequently, P6, B4, B5, P9, B7, B8, P12, B10, and B11 are coded in turn.

Here, the B pictures B1, B2, B4, B5, B7, B8, B10, and B11 may be intra mode, forward mode, backward mode, biprediction mode, bi. There are five prediction modes, including -predictive mode and direct mode. In the bi-prediction mode, two reference pictures have two reference pictures, and the positions of the two reference pictures may be both located in front of each other in time or behind the B picture, or may exist back and forth around the B picture.

In particular, the direct mode takes advantage of the temporal redundancy characteristic that motion continuity remains constant between two adjacent pictures. That is, in the direct mode, the B picture is derived from the forward motion vector and the backward motion vector of the direct mode from the motion vector of the block at the same position of the subsequence picture immediately after the B picture on the display. Since the direct mode does not transmit overhead information such as motion information to the decoder, there is an advantage that the bit rate can be reduced.

In this case, in the conventional direct mode forward motion vector MV _f and backward motion vector MV _b , a co-located block in a subsequent picture uses a motion vector MV. If so, this motion vector MV is obtained by scaling using the time distance between pictures. That is, the forward motion vector MV _f and the backward motion vector MV _{b in the} direct mode are determined using Eq. 1 and Eq. 2 below.

MVf = (TRb * MV) / TRd ....... (Equation 1)

MVb = (TRb-TRd) * MV / TRd ....... (Expression 2)

Here, MV is a motion vector of a block of the same position in the subsequence picture, MVf is a forward motion vector of the direct mode for the B picture, and MVb represents a backward motion vector of the direct mode for the B picture, respectively. . In addition, TRd is the temporal distance between the reference picture referenced by the subsequence P picture and the subsequence P picture, and TRb is the temporal distance between the B picture and the reference picture referenced by the subsequence P picture. .

Finally, the direct mode obtains two motion-compensated blocks using two motion vectors MV _f and MV _b , respectively, and finally averages or interpolates the two motion-compensated blocks. Coding mode in which prediction values are obtained through calculation.

An object of the present invention is to solve the above problems.

Accordingly, one object of the present invention is to provide a motion vector prediction method for moving picture coding that can improve coding efficiency using a long-term reference picture for a B picture in a direct mode.

Another object of the present invention is to provide a motion vector prediction method for video coding that can reduce the amount of bits through an inter mode for a picture having a scene change.

According to a preferred embodiment of the present invention for achieving the above object, the motion vector determination method of the direct mode of the B picture is applied to each block of the B picture when coding is performed using the direct mode in each block of the B picture. The motion vector of the direct mode for the B picture is differently determined according to the type of the reference buffer in which the reference picture referenced by the same position block of the specific picture for the picture is stored.

Preferably, the particular picture is one of short-term reference pictures used for coding the B picture.

Here, the type of the reference buffer is determined using a reference picture index calculated in advance in the same position block of the specific picture. The reference picture index is stored in a system buffer.

When the motion vector calculated in the same position block of the specific picture indicates a long term reference buffer, the forward motion vector of the direct mode for the B picture is a motion vector calculated in the same position block of the specific picture, The backward motion vector for direct mode is determined to be zero.

When the motion vector calculated in the same position block of the specific picture points to a short-term reference buffer, the motion vector in the direct mode for the B picture scales the motion vector calculated in the same position block of the specific picture by the inter-picture time distance. Is determined.

The motion vector calculated from the same position block of the specific picture is stored in the system buffer.

According to another preferred embodiment of the present invention, when coding is performed using the direct mode in each block of the B picture, the B picture according to the type of the reference buffer in which the same position block of the specific picture for each block of the B picture is stored. The motion vectors of the direct modes for are determined differently.

The reference buffer includes a long term reference buffer and a short term reference buffer.

The specific picture is preferably one of a short term reference picture or a long term reference picture.

When the specific picture is a long term reference picture, the forward motion vector of the direct mode for the B picture is a motion vector calculated from the same position block of the specific picture, and the backward motion vector of the direct mode for the B picture is zero. Is determined.

When the specific picture is a short-term reference picture, the motion vector of the direct mode for the B picture is differently determined according to the type of the reference buffer in which the reference picture referenced by the same position block of the specific picture is stored.

According to another preferred embodiment of the present invention, an inter mode coding method of a video may include determining whether a scene change occurs in a P picture; And when the scene change occurs in the P picture, coding the P picture with reference to a long term reference picture.

The scene-transformed P picture is preferably one of a scene cut picture or a partial scene change picture.

When the scene-transformed P picture is a partial scene transformed picture, blocks included in an area in which the scene change is generated are coded using a long term reference picture. Here, the long term reference buffer for storing the long term reference picture is a buffer for storing a coded picture before a predetermined time.

When the scene-transformed P picture is a partial scene transformed picture, blocks included in an area where no scene change is generated are coded using a short term reference picture. Here, the short-term reference buffer for storing the short-term reference picture is a buffer for storing the coded picture after a predetermined time.

According to another preferred embodiment of the present invention, a video coding method comprises the steps of: a) determining whether to convert a scene to a P picture; If there is a scene-converted P picture, coding the P picture in an inter mode with reference to a long term reference picture; b) determining the type of the reference buffer in which the same position block of the specific picture of the B picture is stored when coding is performed using the direct mode in each block of the B picture according to the coding order; And c) calculating a motion vector of the direct mode for the B picture according to the type of the reference buffer and coding the signal in the direct mode.

In the step c), when the specific picture is a long-term reference picture, the forward motion vector of the direct mode for the B picture is a motion vector calculated from the same position block of the specific picture, and the direct mode for the B picture. The backward motion vector is determined to be zero.

In the step c), when the specific picture is a short-term reference picture, determining a type of a reference buffer by using a reference picture index of the same location block of the specific picture; When the motion vector calculated in the same position block of the specific picture indicates a long term reference buffer, the forward motion vector of the direct mode for the B picture is a motion vector calculated in the same position block of the specific picture, Determining a backward motion vector for the direct mode to zero; And when the motion vector calculated in the same position block of the specific picture points to a short-term reference buffer, the motion vector in the direct mode for the B picture may use the motion vector calculated in the same position block of the specific picture as a time distance between pictures. Calculating by scaling.

Here, the short term reference buffer is composed of a FIFO.

Hereinafter, exemplary embodiments of the present invention will be described with reference to the accompanying drawings.

Before describing an embodiment of the present invention, a picture in which a scene change occurs in the whole picture in a video having a long side transform is a scene cut picture, and a picture in which a scene change occurs in a part of the picture is a partial scene change picture. Defined as (partial scene change picture).

3A and 3B are flowcharts illustrating a method of coding a video sequence in a video system according to a preferred embodiment of the present invention. 3A and 3B, pictures are sequentially received from a moving image sequence (S111).

A type of a picture is determined based on the pictures (S 114). That is, it distinguishes whether the input picture is a P picture or a B picture. Here, in the present invention, it is assumed that the intra picture has been coded in advance.

When one picture is a P picture, it is determined whether scene change has occurred in the P picture (S 117). Here, the scene change is determined by comparing the P picture with a picture (P picture or B picture) displayed immediately before the P picture.

If the scene is transformed in the entirety of the P pictures by the determination of S 117, the P picture is a scene cut picture. If the P picture is determined to be a scene cut picture by S 117, the coding is performed by referring to the long term reference picture (S 120).

If the P picture is not a scene cut picture, it is determined whether the P picture is a partial scene change picture (S 123).

When the P picture is a partial scene transformation picture, blocks included in the scene-transformed area are coded by referring to the long-term reference picture by moving to S 120.

Blocks included in the non-scene transformed region are coded with reference to the short-term reference picture (S 129 and S 132).

The long term reference picture refers to a picture stored in a long term reference buffer, and the short term reference picture refers to a picture stored in a short term reference buffer.

The short-term reference buffer is composed of First In First Out (FIFO) that first outputs a picture input first, and is a place where pictures coded before a relatively short time are stored.

The long term reference buffer is a place where pictures coded before a relatively long time are stored. The long term reference picture stores the first picture of each scene set, that is, an intra picture, a scene cut picture, a partial scene change picture, and the like.

If a scene cut picture or a partial scene change picture does not exist in the long term reference buffer, a picture in which the scene change occurs may be newly stored. Therefore, as illustrated in FIG. 4, the long-term reference picture includes an intra picture I0, which is the first scene cut picture of scene set A1, a first scene cut picture P50, a first partial scene change picture P120, and the like that are set in scene set B1. Can be stored in. Here, the scene set divides similar pictures into one set. For example, if a debate was followed by announcers in a row, panel A followed by announcer followed by panel A, then the announcer first appeared in scene set A, then panel A followed by scene set B. The announcer comes back to scene set A, then to scene set B.

As described above, when a scene transformation occurs in the P picture, the P picture is in an inter mode in which a short reference picture or a long term reference picture is coded instead of an intra mode that is conventionally coded through prediction for the P picture itself. Is performed.

As a result, the bit amount can be reduced, thereby improving coding efficiency.

S117 to S132 will be described with reference to FIG. 4.

As shown in FIG. 4, if the current P picture to be coded is the scene cut picture P200 included in the scene set B2, the reference pictures stored in the shotham reference buffer cannot be referenced. This is because the scene cut picture P200 is the first picture of scene set B2, and the scene cut picture P200 is different from the pictures included in scene set A2 (eg, P199, P198, P197, etc.). Since the similarity is significantly reduced from the pictures included in the scene set A2, the coding may not be performed correctly by referring to these pictures.

In this case, the P picture is coded in the inter mode with reference to other reference pictures P50 and P120 included in the scene set B1 which is the same scene set as the scene set B2 including the P picture.

On the other hand, when the partial scene transformation (P 250) occurs in the P picture, coding is performed differently according to two cases. That is, blocks included in the region where the partial scene transformation exists in the P picture are coded in the inter mode with reference to the long term reference pictures P50 and P120 stored in the long term reference buffer. The blocks included in the region where the partial scene transformation does not exist are coded in the inter mode with reference to the short-term reference pictures P249, P248, and P247 stored in the short-term reference buffer.

As described above, when coding is performed on one P picture, it is determined whether a next picture exists and the corresponding picture is input (S 159).

When the picture is a B picture, predictability is obtained using five prediction modes (intra mode, forward mode, backward mode, bi-prediction mode, and direct mode) of the B picture (S 135, S138). Based on this predictability, one prediction mode is selected. In the present invention, a description will be given focusing on the direct mode among them.

First, one block of a plurality of blocks constituting the B picture is read (S 141). Of course, other blocks can be read in sequence.

When one block is read from the B picture, the type of the reference buffer in which the same block of the specific picture is stored is determined (S142). Here, the block of the same position means a block of a specific picture existing at a position corresponding to one block of the B picture.

The specific picture is determined among pictures that are faster than the B picture in coding order regardless of the display order. That is, the specific picture is one of the reference pictures used for the B picture coding. This particular picture includes a short term reference picture and a long term reference picture. The short term reference picture may be a picture immediately before or after the B picture in the display order, or may be located before or after a predetermined interval in which other pictures exist based on the B picture. Here, the long term reference picture is a picture stored in a long term reference buffer, and the short term reference picture refers to a picture stored in a short term reference picture. If the specific picture is a long term reference picture, a forward for the direct mode of the B picture is performed. The motion vector is a motion vector calculated from the same position block of the specific picture, and the backward motion vector for the direct mode of the B picture is determined to be zero and coded (S 150). However, when the specific picture is a short-term reference picture, the reference picture index and the motion vector calculated in the co-located block of the specific picture are read (S 144). The reference picture index and the motion vector are calculated in advance and stored in the system buffer.

The reference picture index is used to determine whether the reference buffer indicated by the motion vector for the specific picture is a long term reference buffer (S 147). As described above, the reference buffers store reference pictures to be referred for coding for B pictures. The reference buffer includes a short term reference buffer and a long term reference buffer.

If the motion vector for the specific picture points to the long term reference buffer, the B picture is coded using Eq. 3 and Eq. 4 below (S 150).

MVf = MV ......... Eq.3

MVb = 0 ......... Eq.4

Here, MV is a motion vector calculated from a subsequence picture, MVf is a forward motion vector of a direct mode for a B picture, and MVb represents a backward motion vector of a direct mode for a B picture, respectively.

That is, when the motion vector for the specific picture points to the long term reference buffer, the forward motion vector of the direct mode of the B picture is a motion vector for the specific picture as it is, and the backward motion vector becomes zero (0). .

As shown in FIG. 5, when the motion vector for the specific picture P200 indicates the long term reference picture P50 in S 150, TRd and TRb in the existing equations Eq. 1 and Eq. 2 mean. Will disappear. That is, since TRd and TRb represent a distance in time between the specific picture P200 included in scene set B2 and the long term reference picture P50 included in scene set B1, up to another scene set A2. The forward motion vector and the backward motion vector in the direct mode may not be calculated using the TRb. Accordingly, when the motion vector for the specific picture P 200 indicates a long-term reference picture, the forward motion vector and the backward motion vector of the direct mode for the B picture are approximated to obtain TRd and TRb representing distances in time. Can be removed.

Referring to FIG. 5, when the two B pictures are inserted and coded in the moving picture sequence, the P picture P200 having a faster coding order than the B1 picture and the B2 picture is coded first. Since P200 is a scene cut picture in which a scene change occurs, inter-mode coding is performed from the long term reference picture P50 stored in the long term reference buffer before the B picture.

A picture to be coded next according to a coding order becomes a B1 picture, and since the B1 picture corresponds to scene set A2, most blocks are predicted from short-term reference pictures included in the same scene set A2. Alternatively, coding is performed by the bi-prediction mode in which both reference pictures are in scene set A2. However, a motion vector of a direct mode is obtained from an intra mode, a backward mode or a pair prediction mode predicted from the P picture P200 included in another scene set B2, and a block at the same position in the P picture P200. It is very unlikely that a predictive direct mode or the like will be used as the coding mode for the block of the B1 picture.

In contrast, since P200, which is a specific picture used to calculate the motion vectors of the direct mode as well as the B2 picture and the B2 picture, corresponds to the same scene set B2, the direct mode with high coding efficiency in most blocks of the B2 picture is used. It will be used as the coding mode.

That is, the motion vector for the specific picture P200 is calculated from the long term reference picture P50 corresponding to the same scene set B2 in advance through the inter mode, and each block of the B2 picture is located at the same position in the specific picture P200. The forward motion vector MVf and the backward motion vector MVb in the direct mode are respectively calculated from the motion vector MV calculated from the long term reference picture P50 of the block, thereby performing direct mode coding. This is included in the same scene set B2 as the B2 picture and the specific picture P200, the long term reference picture P50 is also included in the scene set B1, and the similarity between the scene set B1 and the scene set B2 becomes very high. Direct mode can be used as the coding mode. Accordingly, the B2 picture improves coding efficiency by using a plurality of direct modes that do not require overhead information such as motion information.

On the other hand, if the motion vector for the particular picture points to a short-term reference buffer, the B picture is coded using the existing equations Eq. 1 and Eq. 2.

At this time, the short-term reference picture stored in the short-term reference buffer is included in the scene set including the B picture, and since the other scene set does not exist between the specific picture and the short-term reference picture, TRd and TRb representing the distance in time are reflected. Equations Eq. 1 and Eq. 2 determine the forward motion vector and the backward motion vector of the direct mode.

If coding is performed on one block for the B picture, the next block for the B picture is read by checking whether the next block exists (S 156). This process is performed for all blocks included in the B picture.

When all the coding for the B picture is performed, it is again checked whether the next picture exists and then inputs the corresponding picture to perform video coding (S 159).

As described above, according to the video coding method according to the present invention, the B picture is determined by differently determining the forward motion vector and the backward motion vector of the direct mode of the B picture according to the reference buffer indicated by the motion vector calculated from the specific picture. By coding a large number of direct mode is used as the coding mode, it is possible to improve the overall coding efficiency.

In addition, according to the video coding method according to the present invention, coding efficiency can be improved by reducing the bit amount by coding a P picture in which a scene transformation has occurred, in an inter mode through motion compensation using a long-term reference picture. have.

Claims (7)

A method of performing motion vector prediction for a current block in image signal processing, Obtaining reference picture index information of an image block providing a motion vector used to perform motion vector prediction for the current block; Performing motion vector prediction for a current block based on the reference picture index information; The reference picture index information is an index indicating a reference picture to which the image block refers. The performing of the motion vector prediction includes setting a motion vector provided by the image block to be the same as the motion vector for the current block. The method of claim 1, wherein the motion vector for the current block and the motion vector provided by the image block are forward motion vectors, respectively. The method of claim 2, wherein the reference picture index information is an index indicating a forward reference picture referenced by the image block. A method of predicting a motion vector for a current block in a video decoder, Obtaining a reference picture index of an image block that provides a motion vector used to predict a motion vector for the current block; Predicting a motion vector for a current block based on the reference picture index; The motion vector prediction method determines the motion vector for the current block to the same value as the motion vector provided by the image block. The method of claim 4, wherein the reference picture index is an index indicating a forward reference picture referenced by the image block. 5. The method of claim 4, wherein the motion vector for the current block is a forward motion vector. The method of claim 4, wherein the motion vector provided by the image block is a forward motion vector.

Images