KR20050019876A - Method for error concealment in video sequences - Google Patents

Abstract

A method and apparatus for error concealment in a video sequence is disclosed. When there is a scene transition in a video sequence and there is an error in a picture belonging to the scene transition, an error concealment procedure is used to conceal the error based on the type of scene transition. The scene transition is provided to the video decoder in an additional enhancement information message with information about the type of the scene transition. If the scene transition is a gradual scene transition, a spatiotemporal error concealment algorithm is used to conceal the picture. If the transition is a scene cut and only part of the picture is lost or damaged, spatial error concealment is used to conceal the lost or damaged part of the picture. If the entire picture belonging to a scene cut is lost or damaged and the picture starts a new scene, it is not concealed.

Description

Method for error concealment in video sequences}

The present invention relates generally to video encoding, and more particularly to concealing artifacts of picture quality caused by errors.

The video sequence consists of still pictures or frames. The video compression method is based on reducing parts of redundant and perceptually irrelevant video sequences. Redundancy in a video sequence may be classified into spectral, spatial and temporal redundancy. Spectral redundancy refers to the similarity between different color components of the same picture. Spatial redundancy results from the similarity between adjacent pixels in a picture. Temporal redundancy exists because of the probability that objects that appear in the previous image frame will also appear in the current image frame. Compression can be achieved by using this temporal redundancy and predicting the current picture from another picture called the reference picture. Further compression may be achieved by generating motion compensation data that describes the motion between the current picture and the reference picture.

Video compression methods typically distinguish between pictures that use temporal redundancy reduction and those that do not. Compressed pictures that do not use the temporal redundancy reduction method are commonly called intra- (INTRA, or I-) frames or pictures. Temporally predicted images are generally forwardly predicted from a picture that occurred before the current picture, and are called inter- or P-frames. Compressed video clips typically consist of a series of pictures, which can be classified into intra-pictures that are approximately temporally independent and inter-coded differentially temporally. Intra pictures are typically used to stop the temporal propagation of transmission errors in a reconstructed video signal and to provide random access points to the video bitstream. Since the compression efficiency provided by intra pictures is generally lower than the compression efficiency provided by inter pictures, such intra pictures are rarely used, especially in low bit-rate applications.

The video sequence may consist of multiple camera scenes or shots. A shot is defined as a set of consecutive frames or pictures taken with one camera. In general, frames within one shot have a high correlation. However, in a typical video sequence, the picture content is quite different when transitioning from one scene to another, so the first picture of the scene is generally intra coded. Changes between different shots in a video sequence are referred to as "scene transitions". Scene transitions can take many different forms. For example, in a "scene cut," one shot may end and the other may begin abruptly. In other cases, the transition is gradual and occurs for more than one frame. Examples of gradual scene transitions include "dissoves", "fades (fade-in, fade-out)" and "wipes".

Compressed video is easily damaged by transmission errors, mainly for two reasons: First, due to the use of temporal predictive differential coding (inter frames), errors are conveyed both temporally and spatially. Second, the use of variable length codes increases the susceptibility of the video bitstream to errors. There are many ways for the receiver (video decoder) to indicate the damage caused on the transmission path. In general, upon reception of a signal, a transmission error is first detected and then corrected or concealed by the next decoder. The word "error correction" refers to the process of fully recovering erroneous data as if no error had occurred from the beginning, while "error concealment" is rarely seen in a reconstructed video sequence. It is the process of concealing the effect of transmission error so that it cannot be recognized.

Currently, the Joint Video Team (JVT) of the ISO / IEC Motion Picture Expert Group and the ITU-T Video Coding Expert Group (ITU-T) for the ITU-T H.264 / MPEG-4 part 10 AVC video codec. The video decoder developed by the Video Coding Expert Group lacks a method of determining how to conceal transmission errors of intra coded frames and scene change frames, and the present invention was developed to solve this problem.

FIG. 1 is a flow chart illustrating an error concealment method according to the present invention, showing how a suitable error concealment method for pictures during scene change is selected according to the type of scene change.

2 is a block diagram illustrating a video encoder implemented to provide an encoded data stream that includes information indicative of a scene change for the purpose of error concealment, in accordance with the present invention.

3 is a block diagram illustrating a video decoder implemented according to the present invention and corresponding to the video encoder illustrated in FIG. 2.

It is an object of the present invention to provide a method in which an appropriate form of error concealment can be selected for frames belonging to a scene transition of a video sequence. The method is equally applicable to abrupt scene transitions (eg scene cuts) and gradual scene transitions, eg fade, wipes and dissolves and the like.

Two types of information are needed to perform effective error concealment of frames belonging to a scene change: 1) information on frames where the shot change starts and ends; And 2) information regarding the type of associated scene transition (cut, dissolve, fade, wipe, etc.). Since the two types of information described above are not necessary for accurate decoding of Video Coding Layer (VLC) data, the present invention provides information related to scene transition as Supplemental Enhancement Information (SEI). Proposed to be included in the encoded video bitstream as an Additional Enhancement Information (SEI) message. All the necessary information for the concealment of errors occurring in the frames belonging to the scene change can then be inferred from the SEI message.

According to the present invention, each scene of the video sequence is combined with a scene indentifier value. The scene identifier values of successive scenes are different, so the video decoder can conclude that a scene change occurred when receiving a different scene identifier than previously received. In the scene transition period the frames are combined with two scene identifier values, i.e., scene identifier values from both scenes at the time of transition. In addition, progressive scene transitions are associated with a particular transition type, such as dissolve, fade-in, fade-out, wipe, or other non-mentioned types (eg, other types of transitions). This coarse classification provides sufficient guidance to the decoder, allowing the decoder to select a suitable error concealment algorithm to conceal data loss or corruption during scene transitions.

Thus, according to a first aspect of the invention, a method is provided for concealing an error in a video sequence frame, wherein the video sequence comprises at least a first scene and a second scene, wherein the second scene is a scene from the first scene. The transitions, which are transitions, comprise a plurality of frames and the transitions are of many types. The method,

Identifying the type of scene transition; And

Applying an error concealment procedure to conceal an error of a frame belonging to the transition based on the identified type of scene transition.

The identified type of scene transition may be a scene cut or a gradual scene transition.

Preferably, when the entire picture belonging to the scene cut is lost, the lost picture is not concealed.

Advantageously, if a portion of the picture belonging to the scene cut is lost or damaged, a spatial error concealment algorithm is applied to conceal the lost or damaged portion of the picture.

Advantageously, a spatio-temporal error concealment algorithm is applied to conceal the lost or damaged portion of the picture if the entire picture belonging to the progressive transition is lost or damaged.

Preferably, if a portion of the picture belonging to the progressive transition is lost or damaged, a spatiotemporal error concealment algorithm is applied to conceal the lost or damaged portion of the picture.

Preferably, information indicative of the identified scene transition is provided to the decoder in a supplemental enhancement information message to cause the decoder to conceal the error based on the information.

Advantageously, the information indicative of the identified scene transition comprises an indication of a scene transition type, and the information indicative of the identified scene transition is provided for each frame belonging to the transition.

According to a second aspect of the present invention, there is provided a video encoding apparatus for encoding a video sequence into a data stream, the video sequence having at least a first scene and a second scene, and having a transition from the first scene. , The scene change has a plurality of frames and the scene change is any one of a plurality of types. The video encoding apparatus,

Means for identifying frames related to the transition;

Means for providing information about the type of the conversion.

According to a third aspect of the invention, there is provided a video decoding apparatus for decoding a video sequence into a data stream, said video sequence having at least a first scene and a second scene, and having a transition from said first scene. And the scene change has a plurality of frames and the scene change is any one of a plurality of types. The video decoding apparatus,

Means for receiving the data stream; And

An error concealment algorithm that conceals an error of a frame belonging to the transition based on the type of the scene transition.

The invention will become apparent upon reading the description made with reference to FIGS.

As mentioned above, the additional enhancement information (SEI) included in the encoded video bitstream is not necessary to accurately decode the encoded video data, but is nonetheless helpful for decoding or presentation purposes. Thus, SEI information is an ideal vehicle for carrying information about the scene to which a particular frame of a video sequence belongs and providing information about the scene transition.

According to the ITU-T H.264 / MPEG-4 part 10 AVC Video Coding Standard, an SEI element includes one or more SEI messages. Each SEI message consists of an SEI header and an SEI payload. The type and size of the SEI payload is encoded using extensible syntax. The size of the SEI payload is expressed in bytes. Valid SEI payload types are listed in Appendix C of the draft JVT Committee (see document JVT_D015d5).

The SEI payload may be an SEI payload header. For example, the payload header may indicate a picture to which special data belongs. Payload headers are defined individually for each payload type. Definitions of SEI payloads are specified in Appendix C of the draft JVT Committee (see, again, document JVT_D015d5).

Transmission of the SEI unit takes place simultaneously with other Network Abstraction Layer (NAL) units. The SEI message may be associated with a slice, a portion of a picture, a picture, a group of certain pictures, a past sequence, a sequence currently decoded, or a sequence to be decoded in the future. The SEI message may also be associated with one or more NAL units before or after the transmission order.

Table 1 below defines the SEI payload syntax used in the ITU-T H.264 / MPEG-4 part 10 AVC video coding standard, and Table 2 lists signals related to scene information, as proposed in accordance with the present invention. Represents a special syntax used in association.

TABLE 1

: SEI Payload Syntax

sei_payload (Payload, PayloadSize) { Category Descriptor if (PayloadType == 1) temporal_reference (PayloadSize, PayloadSize) 7 else if (PayloadType == 2) clock_timestamp (PayloadSize, PayloadSize) 7 else if (PayloadType == 3) panscan_rect (PayloadSize, PayloadSize) 7 else if (PayloadType == 4) scene_information (PayloadSize, PayloadSize) 7 else reserved variable if (! byte_aligned ()) { bit_equal_to_one f (1) while (! byte_aligned ()) bit_equal_to_zero f (1) } }

TABLE 2

SEI scene information syntax

scene_information (PayloadType, PayloadSize) { Category Descriptor scene_identifier u (8) if (more_sei_payload_data ()) { second_scene_identifier u (8) if (more_sei_payload_data ()) scene_transition_type e (v) } }

The scene information parameters given in Table 2 above relate to the next NAL unit containing the macroblock data encoded in the transmission order.

scene_identifier : A scene is defined as a set of consecutive frames taken by a camera. In general, frames within a scene have a high correlation. According to the present invention, frames in a given scene share the same scene_identifier parameter value, and successive scenes in the encoding order should not have the same scene_identifier parameter value.

second_scene_identifier : If present, the second_scene_identifier parameter indicates the next NAL unit containing coded macroblock data belonging to one frame containing picture data from both scenes. In other words, the frame belongs to a gradual scene change. The second_scene_identifier parameter is a scene identifier of a scene behind in encoding order.

scene_transition_type : If the scene_transition_type parameter does not exist in the SEI scene information, this indicates that the scene change type is unknown, undefined or irrelevant. If present, the values given in Table 3 below are valid:

TABLE 3

: Scene transition types according to a preferred embodiment of the present invention

value Explanation 0 dissolve One Fade-out 2 Fade-in 3 Wipe Etc reservation

The manner in which the scene information described above is used in the decoder procedure to deal with data loss or corruption is now described.

If the entire picture is lost just before the current picture and the scene has changed since the previously received picture, the lost picture should not be concealed since the lost picture starts a new scene. If the entire picture is lost just before the current picture and there are no scene changes since the previously received picture, the decoder hides the lost picture. If the entire picture is lost during the transition period, the decoder uses the indicated scene transition type when concealing the lost picture.

If a portion of the current picture is lost or corrupted and the scene information SEI message is combined with the picture, the decoder performs the following operations:

1) If the scene has changed since the previous picture was received, the decoder applies a spatial error concealment algorithm to conceal the lost or corrupted portion of the current picture.

2) If the scene has not changed since the previously received picture, the decoder uses a spatiotemporal error concealment algorithm.

3) If the current picture belongs to a transition, the decoder uses the displayed transition type when concealing the lost picture.

According to the present invention, the encoder should generate a scene information SEI message if it is necessary to generate an error prone transmission environment, or a video content description based on the encoded video signal. Even if there is no immediate need for a content description, the latter need may arise for some types of video content, such as entertainment video, to generate a video content description based on an encoded video signal. Thus, according to the invention, it is preferred that the encoder always generate a scene information SEI message whenever possible.

Therefore, the encoder generates a scene information SEI message for each scene cut and gradual transition. For each scene-cut picture, there is an associated scene information SEI message that can be repeated later for error resilience. For each progressive scene transition, the (preferably repeated) associated with the first transition picture (e.g., the first picture consisting of both the transit-from and the transit-to scene) There is a scene information SEI message. For each gradual scene transition, it is also related to the next first picture (the last transition picture refers to the last picture consisting of both the transit-from and the transit scene). Scene information SEI message (preferably repeated) is present. As mentioned above, the value of the scene_identifier parameter is different in successive scenes.

In a packet-oriented transmission environment, the transport packetizer copies each scene information SEI message in at least two packets, if possible, to ensure accurate reception of at least one message occurrence. In RTP transmission, the packetizer uses a composite packet to combine the scene information SEI message with the encoded picture content. In a byte-stream-oriented transmission environment, each scene information SEI message is at least copied.

According to the present invention, a method of error concealment in a video sequence by shot change signal is shown in flow chart 100 of FIG.

When the decoder encounters data loss or corruption during the decoding process, the decoder determines in step 110 whether the lost or corrupted picture is the entire picture or part of the picture. If the entire picture is lost, the decoder determines what type of loss situation has occurred (step 120). If the entire picture was lost just before the current picture, and the scene has changed since the previously received picture (eg, as indicated by the scene_identifier parameter value of the received SEI information), in this case, as described above The lost picture must not be concealed, because the lost picture marks the beginning of a new scene (step 124). If the entire picture is lost just before the current picture and no scene change has occurred since the previously received picture, the decoder conceals the lost picture, as shown in step 122. If the entire picture was lost during the transition period, the decoder uses the indicated scene transition type (obtained from the received SEI information) when concealing the lost picture, as shown in step 126.

If a portion of the picture is missing, the decoder determines what type of loss situation has occurred (step 130). If a portion of the current picture is lost or corrupted and a scene information SEI message is associated with the picture, the decoder should perform the following actions: If the scene has changed since the previously received picture, the decoder may As shown at 134, a spatial error concealment algorithm is applied to conceal lost or corrupted portions of the current picture. If the scene has not changed since the previously received picture, the decoder uses a spatiotemporal error concealment algorithm, as shown in step 132. If the current picture belongs to a scene change, the decoder uses the indicated scene change type when concealing the lost picture, as shown in step 136.

In order to implement the error concealment method as shown in flow chart 100 of FIG. 1, a video encoder implemented in accordance with the present invention monitors scene changes and delivers information indicating scene changes in the bitstream generated by the encoder. It should be possible. 2 shows a block diagram of such a video encoder 200. As shown, a video encoder implemented in accordance with the present invention includes a scene change monitor 210, an encoding engine 220, a controller 230, and a multiplexer / packetizer 240. A video coding engine 220 in which a video input signal representing the video sequence is applied to the input of the video encoder and via the scene change monitor 210, the individual frames of the video sequence are encoded, for example in an intra- or inter-format. Is supplied. The scene change monitor calculates a sum of accumulated absolute differences between pixels of successive frames of the sequence, for example, to identify the various scenes making up the video sequence, or known from the prior art. By applying another scene detection method, frames are examined. The scene change monitor 210 provides the controller 230 with an indication of a scene to which each frame belongs. When a transition between scenes (eg, scene cut, fade, dissolve, wipe, etc.) is detected, the scene transition monitor 210 also provides an indication of the transition type to the controller 230. The control unit assigns an identifier (e.g., a number) to each scene identified by the scene change monitor and associates the identifier to each frame identified as belonging to the scene in question. In addition, when a scene change is detected, the controller 230 instructs the video coding engine 220 to encode the first frame of the new scene in the intra encoding format. Advantageously, then all subsequent frames belonging to the new scene are encoded in an inter encoding format, unless for some reason a given frame has to be encoded in a different format. In a preferred embodiment of the present invention, the controller 230 combines the additional enhancement information (SEI information) with each frame belonging to a scene change, and delivers the SEI information to the multiplexer / packetizer 240. Advantageously, the SEI information for the frames forming part of the scene transition is formed as described in the preferred embodiment of the present invention. The multiplexer / packetizer also receives encoded video data from video encoding engine 220 and forms a single bitstream from the encoded video data and the SEI information. The bitstream is then transmitted, for example, via a transmission channel to the corresponding video decoder (see FIG. 3) or to a storage device (not shown) for later playback and viewing. 3 is a block diagram illustrating a video decoder 300 implemented in accordance with the present invention, corresponding to the video encoder described in connection with FIG. As can be seen in FIG. 3, a video decoder according to the present invention includes a depacketizer / demultiplexer 310, a controller 320, an error concealer 330, and a video decoding engine 340. do. The depacketizer / demultiplexer receives an encoded bitstream representing a video sequence from a transport channel in the form of a data packet. The depacketizer / demultiplexer reconstructs the encoded video bitstream from the received data packet and splits the video bitstream into its component parts (e.g., other types of information related to encoded video frames of the sequence). do. According to the present invention, the depacketizer / demultiplexer extracts, among other things, additional enhancement information (SEI information) including information related to scene transition from the encoded video bitstream, and controls the SEI information into the control unit 320. To pass. The data needed to decode the encoded video frames is passed from the depacketizer / demultiplexer 310 to the video encoding engine 340, where the individual frames of the video sequence are for example intra and inter decoding techniques. Is reconstructed using As each frame is decoded, video decoding engine 340 examines the received video data for errors that may have occurred during transmission of the encoded video bitstream on the transport channel from the encoder. If the video decoding engine detects that a particular frame is such an error, or that the frame is so severely damaged that it cannot be decoded at all (e.g., the frame is completely corrupted), the video decoding engine may then implement an appropriate error concealment algorithm. Attempt to conceal the error or the entire frame. According to the present invention, the selection of an appropriate error concealment method is performed by the error concealment unit 330. The error concealment unit receives information about scenes to which each frame belongs from the control unit 320. In the case of frames that are part of a scene change, the control unit 320 also passes information related to the type of scene change to the error concealment unit. Thus, when the decoding engine 340 detects an error affecting a frame that is part of a scene transition, the error concealment unit 330 is adapted to conceal the error by considering the scene to which the frame belongs and the type of scene transition. The method can be chosen. Preferably, in making this selection, the error concealment section 330 applies the selection method as described in the preferred embodiment of the present invention described above. The video decoding engine 340 then conceals the error of the frame using the selected error concealment algorithm and outputs the decoded frame for display on a display device (not shown), for example.

In order to confirm the effect of the error concealment method on scene cut and scene transition frames according to the invention, it is implemented according to the ITU-T H.264 / MPEG part 10 AVC video coding standard and operated by the method according to the invention. A series of simulation experiments were performed using video encoders and decoders modified to make this possible. These simulation experiments are described in detail below.

A. Error Concealment Simulation in Random Access Frames and Scene Cuts

In this simulation, the sequences and bit rate proposed in document VCEG-N79r1 were used, and the common conditions for the packet-loss environment (as defined in VCEG-N79r1) were applied. Additionally, to simulate the effect of error concealment on scene-cut frames, the well-known sequences "News", "Foreman", "Coastguard", "Carphone" ), And "silent" an artificial sequence of 30 frames with typical scene cuts was constructed. In the following, the artificial sequences are referred to as "MixedSeq". In all cases, an intra-frame period of about 1 second was used to enable frequent random access. In the mixed sequence, the intra period allowed all scene cuts to be intra coded. Loss-Aware R / D Optimization (LA-RDO) was also used. Other encoding parameters used to encode the sequences are shown in Table 4 below.

TABLE 4

: Encoder Parameters Used in Error Concealment Simulation in Random Access Frames and Scene Cuts

Bitstream mode RTP Motion vector resolution 1/4 pel Hadamard Transformation Used Search range 16 The number of previous frames used for inter motion search 5 Possible block type all Slice mode Fixed size, 1400 bytes / slice B-frames and SP-frames Not used Symbol mode UVLC Data partitioning 1 split per slice Sequence header No sequence header Search scope limit none Limited intra prediction Used Limited reference frames Used Number of decoders for LA-RDO 30

Error Concealment

The generation of the scene information SEI message in the encoder according to the invention was simulated according to the guidelines indicated in the best embodiment of the invention, as described above, and the following two decoder procedures are compared:

1. Standard Joint Model decoder, including the error concealment method described in Appendix D of the JVT Working Draft (see document JVT-C039).

2. A combined model decoder improved by the decoder procedure according to the invention.

Bit rate and PSNR calculation

As mentioned in the common conditions embodied in document VCEG-N79r1, encoding parameters such as quantization parameters may be close to the channel bit rate where the resulting bit rate is possible, taking into account IP / UDP / RTP headers of 40 bytes per packet. Is chosen. The PSNR value is calculated using each frame of the source sequence, including skipped and lost frames. In order to reduce the effect of affecting the overall result by the first frames (the first encoded frames have a larger average size than the average size of the entire sequence), the bit rate and average PSNR value are the sixth coded frame. Is calculated from them. This method allows short sequences to be encoded with fair results. Instead of encoding 4000 frames, 300-400 frames of each designated sequence are used to ensure that at least 100 frames are encoded and at least 300 frames are used.

Packet loss simulation

In the simulation, a packet containing a set of parameters (Table 4) is delivered reliably (eg by out-of-band during session setup) and therefore no error pattern for the packet from the error pattern file. It is also assumed that no is read. At least one packet of the first frame should be received to avoid decoder collisions. To satisfy this condition, the first packet of the first frame is always received regardless of the corresponding error pattern.

Representative Decoding Run

The coded bitstream is decoded several times (each called a decoding run). The start loss position of the run in n + 1 order follows the last loss position of the nth run. The number of decoding runs is chosen such that there are at least 8000 packets overall. The overall average PSNR is obtained by averaging the average PSNR values of all decoding runs. The representative decoding run is chosen such that its average PSNR is closest to the overall average PSNR. The instantaneous PSNR values and the decoded sequence of representative runs are obtained for instantaneous PSNR plots and stored for subjective quality assessment.

result

As shown in the simulation results for MixedSeq at 144 kbps, shown in Table 11 below, the use of intra error concealment for scene cuts is more than using inter error concealment in terms of objective and subjective quality. Provide better performance.

Conversely, as can be seen from the other six coding cases shown in Tables 5-10, using inter error concealment for frames that are not scene cuts is consistently better than using intra error concealment. This demonstrates the usefulness of the present invention.

TABLE 5

: Sequence "Foreman" encoded at 64 kbps, 7.5 frames / s

TABLE 6

: Sequence "Foreman" encoded at 144kbps, 7.5 frames / s

TABLE 7

: Sequence "Hall" encoded at 32kbps, 10 frames / s

TABLE 8

: Sequence "Irene" encoded at 384kbps, 30 frames / s

TABLE 9

: Sequence "Paris" encoded at 144kbps, 15 frames / s

TABLE 10

: "Paris" sequence encoded at 384kbps, 15 frames / s

TABLE 11

: Sequence "MixedSeq" encoded at 144kbps, 15 frames / s

B. Simulation of error concealment of fades

In order to simulate the effect of error concealment on fade-out and fade-in frames, two artificial sequences were created with 10 fade-out frames, 10 fade-in frames and 10 general frames. . One is made from a combination of "News" and "Akiyo" (with low movement), and the other is made from "Carphone" and "Foreman" (with proper movement). lost. After encoding using the JVT joint model encoder and the I-P-P-P coding pattern, the loss of some of the fading frames was simulated and the lost bitstream was fed to the decoder. Two different error concealment methods have been used to conceal errors caused by the loss of fading frames:

1. Conventional error concealment method in JVT codec (as described in Appendix D of JVT Working Draft (JVT-C039)); And

2. Special error concealment method for fades according to the present invention as described below.

Error concealment method

The frames lost upon fade are concealed by copying and scaling the pixel values of the previous frame. However, if there is only one previous frame in the scene change period, scaling is not performed. If Mn ' is the average Y (luminance) pixel value of the previous frame and Mn'' is the average Y pixel value of the frame before the previous frame, the scaling factor f can be calculated as follows:

The concealed Y, U and V values (Ys, Us, Vs) of the pixel are calculated from the spatially corresponding values (Y, U, V) of the previous picture as follows:

As shown in the simulation results, the use of a special error concealment method for fade according to the present invention is considerably better in both objective and subjective quality compared to the conventional error concealment method defined in Appendix D of the JVT video coding recommendation. Performance. It may be argued that the visual quality of the gradual transition is not important. However, poor error concealment during scene transitions not only results in poor quality transition frames, but also because of temporal error propagation, which results in poor quality in general frames after scene transitions.

Error Concealment of Dissolves

The following two methods for error concealment of a lost frame during dissolve are shown below. If the decoder can buffer a sufficient number of frames before decoding, algorithm A should be used. If not, algorithm B should be used.

Algorithm A

If the pre-decoder buffer contains any intra-coded frame of the second scene (frame after the transition period), the intra frame is used as the second anchor frame upon error concealment. If such intra frame is not available, algorithm B should be used. The first anchor frame is the last reconstructed frame. When presenting dt1 the temporal distance between the first anchor frame and the missing frame, dt2 the temporal distance between the second anchor frame and the missing frame, (y1, u1, v1) is the pixel of the first anchor frame and (y2 If, u2, v2 is the spatially corresponding pixel of the second anchor frame, then the hidden pixels (y, u, v) are given by:

Where u and v are calculated similarly, but their signs should be taken into account:

Here, the mathematical function "clip1" is defined as:

Algorithm B

General Spatio-temporal Error Concealment.

Error concealment of Wipes

The decoder should detect:

1. the shape of the boundary between two scenes included in a wipe; And

2. The rate defining how quickly the last scene is covered by the start scene.

The detection can be performed, for example, by comparing the reconstructed pictures and calculating the correlation between the blocks. If two spatially corresponding blocks are correlated from consecutive pictures in time, the blocks are from the same scene. Otherwise, the blocks are from other scenes.

Based on the guessed shape and ratio, the decoder can calculate a guess of the position and appearance of the boundary in the missing picture or region. Lost areas that belonged to the last scene of the previous picture and are believed to belong to the last scene of the missing picture / area can be concealed by copying the area from the previous picture. Similarly, lost areas that belong to the start scene of the previous picture and are supposed to belong to the start scene of the lost picture / area may be concealed by copying the area from the previous picture. Lost areas that belong to the last scene of the previous picture and are supposed to belong to the start scene of the lost picture / region are spatially concealed from the content of the adjacent start scene. When concealing the lost area, a boundary matching the adjacent correctly reconstructed blocks can be used as is often done in error concealment.

Error concealment of other conversion types

Normal spatiotemporal error concealment methods should be used.

As mentioned above, although preferred embodiments of the present invention have been described, those skilled in the art to which the present invention pertains may realize that the present invention may be embodied in a modified and deleted form without departing from the essential characteristics of the present invention. I can understand.

Claims (22)

A method of concealing errors in a video sequence frame, wherein the video sequence includes at least a first scene and a second scene, the second scene is a scene transition from the first scene, and the scene transition is performed by a plurality of frames. Wherein the scene transition is any one of a number of types, Identifying the type of scene transition; And Applying an error concealment procedure to conceal an error of a frame belonging to the transition based on the identified type of scene transition. The method of claim 1, And the identified type of scene transition is a scene cut. The method of claim 2, And when the entire picture belonging to the scene cut is lost, the lost picture is not concealed. The method of claim 2, And when a part of the picture belonging to the scene cut is lost or damaged, a spatial error concealment algorithm is applied to conceal the lost or damaged part of the picture. The method of claim 1, And the identified type of scene transition is a gradual scene change. The method of claim 5, And the scene change is a fade. The method of claim 5, And the scene change is a dissolve. The method of claim 5, And the scene change is a wipe. The method of claim 5, And a spatio-temporal error concealment algorithm is applied to conceal the lost or damaged portion of the picture if the entire picture belonging to the progressive transition is lost or damaged. The method of claim 5, If a part of the picture belonging to the progressive conversion is lost or damaged, a spatiotemporal error concealment algorithm is applied to conceal the lost or damaged part of the picture. The method of claim 1, And the information indicative of the identified scene change is provided to the decoder in a supplemental enhancement information message to cause the decoder to conceal the error based on the information. The method of claim 11, And wherein the information indicative of the identified scene change comprises an indication of a scene change type. The method of claim 11, Information indicative of the identified scene transition is provided for each frame belonging to the transition. A video encoding apparatus for encoding a video sequence into a data stream, the video sequence having at least a first scene and a second scene, having a scene transition from the first scene, the scene transition having a plurality of frames. The video encoding apparatus wherein the scene change is any one of a plurality of types, Means for identifying frames related to the transition; And Means for providing information on the type of transition. The method of claim 14, And the information is provided in an additional enhancement information message. The method of claim 15, The information is provided for each frame belonging to the transition. A video decoding apparatus for decoding a video sequence into a data stream, the video sequence having at least a first scene and a second scene, having a transition from the first scene, wherein the scene transition comprises a plurality of frames. The video decoding apparatus having the scene change is any one of a plurality of types, Means for receiving the data stream; And And an error concealment algorithm that conceals an error of a frame belonging to the transition based on the type of the scene transition. The method of claim 17, And the type of scene change is indicated in additional enhancement information provided to the encoding device. The method of claim 17, The type of scene transition is a gradual scene transition, the entire picture belonging to the gradual scene transition is lost or corrupted, and the error concealment algorithm comprises a spatiotemporal error concealment algorithm for concealing the lost or corrupted picture. Video decoding device. The method of claim 17, The type of scene transition is a gradual scene transition, wherein some of the pictures belonging to the gradual scene transition have been lost or corrupted, and the error concealment algorithm includes a spatiotemporal error concealment algorithm for concealing a portion of the lost or corrupted picture. Video decoding apparatus. The method of claim 17, The type of scene transition is a scene cut, a portion of a picture belonging to the scene cut is lost or corrupted, and the error concealment algorithm includes a spatial error concealment algorithm for concealing an error of the picture. Device. The method of claim 17, The type of scene transition is a scene cut, the entire picture belonging to the scene cut is lost or corrupted, and the error concealment algorithm is adapted to ignore the lost or corrupted picture.