KR20110010077A - Apparatus and method of decompressing distributed way coded video - Google Patents

Abstract

PURPOSE: A device and a method of decoding a dispersion video encoding image are provided to effectively remove noise error of assistance information as minimizing the waste of compression data, thereby remarkably increasing the compression performance. CONSTITUTION: A key feature restoring unit restores key feature information transmitted from a decoding device(20). An assistance information generating unit(22) generates assistance information by using the key feature restored in the key feature restoring unit. A multi noise model based noise model providing unit provides noise model to be use in a certain channel coding unit. A channel code decoding unit decodes certain channel coding units about a given decoding unit.

Description

Apparatus and method for decoding distributed video encoded image {APPARATUS AND METHOD OF DECOMPRESSING DISTRIBUTED WAY CODED VIDEO}

The present invention relates to an apparatus and method for decoding a distributed video encoded image that improves compression performance based on a multi-noise model.

In general, since digital image data before compression has a considerable amount of size, it is compressed by an efficient compression method such as MPEG and H.26X. These technologies have been used for applications such as DVDs and broadcasts. Recently, wireless technologies such as 2.5G / 3G have been used for wireless mobile video transmission.

Compression of digital image data is achieved by finding and reducing information that is not needed or can be found later from the image data before compression, firstly by reducing temporal redundancy and then by spatial redundancy. To reduce spatial redundancy, and finally to reduce statistical redundancy.

Conventional compression techniques represented by MPEG and H.26X achieve high compression efficiency by eliminating temporal redundancy based on motion prediction / compensation, but they put a lot of computational burden on the video encoding apparatus. Therefore, in applications where the amount of computation and / or power resources of the encoding device are limited, such as sensor networks or cellular video calls, which are recently increasing or will explode in the future, it is important to reduce the complexity of the encoding device because existing compression technologies are inadequate. It has been a problem.

Distributed source coding (DSC) technology has attracted attention as one of methods for solving the complexity problem of the encoding apparatus. Distributed source coding techniques are based on the Slepian-Wolf theory. According to the Slepian-Wolf theory, even if each of the sources having correlation is independently compressed and encoded, if the cross correlation is used when decoding, the same compression coding gains as those of the prediction coding are linked to each other can be obtained. Can be.

The Distributed Video Coding (DVC) technique based on the Wyner-Ziv theory is an extension of the lossy coding / decoding to the case of lossy compression coding. This means that in order to reduce the redundancy between pictures, a series of signal processing procedures, such as motion prediction / compensation, which have been performed in the video encoding apparatus, can be moved to the decoding apparatus without large compression coding gain / loss.

The most representative examples of distributed video encoding techniques are A. Aaron, S. Rane, R. Zhang, and B. Girod et al. Wyner-Ziv coding technology introduced by the paper, "Wyner-Ziv coding for video: Applications to compression and error resilience," published at the IEEE Data Compression Conference, 2003.

The distributed video encoding technique generates auxiliary information corresponding to the current coding unit by using similarity between before and after pictures of the current coding unit compressed by the decoding apparatus, and then, the difference between the auxiliary information and the image data corresponding to the current coding unit. Is regarded as noise generated in the virtual channel, and then the pre-compression image data corresponding to the current coding unit is reconstructed by removing the noise in the auxiliary information using the parity data transmitted from the encoding apparatus.

FIG. 1 is a diagram illustrating a configuration of a coding apparatus 110 and a decoding apparatus 120 of a conventional Wyner-Ziv coding technique.

As shown in FIG. 1, the encoding apparatus 110 of the conventional Wyner-Ziv coding technique includes a key picture encoding unit 111, a quantization unit 112, a block unitization unit 113, and a channel code encoding unit ( 114, and the corresponding decoding apparatus 120 includes a key picture reconstructor 121, an auxiliary information generator 122, a noise model generator 123, a channel code decoder 124, and an image reconstruction. The unit 125 is included.

The encoding apparatus 120 according to the Wyner-Ziv coding technique divides data to be compressed into two classes, one of which is image data encoded by distributed video encoding (hereinafter referred to as 'WZ picture'), and the other. Denotes image data (hereinafter, referred to as a "key picture") encoded by a conventional encoding method other than the distributed video encoding method. In general, the key pictures are encoded by the key picture encoder 111 in a predetermined method selected by a user and transmitted to the decoding apparatus 120, such as H.264 / AVC.

The key picture reconstructing unit 121 of the decoding device 120 corresponding to the encoding device 110 according to the conventional Wyner-Ziv coding technique may select a predetermined method previously selected by the user, such as the H.264 / AVC intra picture coding method. The auxiliary picture generator 122 restores the key picture encoded and transmitted, and generates side information using the key pictures restored by the key picture restorer 121. This auxiliary information is estimation information generated by the decoding apparatus, which is considered to be similar to the WZ picture to be reconstructed, and regards the difference between the pre-compression image data and the auxiliary information thereof as a noise error. Since this noise error is not the noise error caused by the transmission on the actual communication channel, the noise causing this error is called virtual channel noise.

In general, the generation of auxiliary information in the auxiliary information generator 122 is performed by using interpolation, which assumes linear movement between key pictures located before and after the WZ picture. Extrapolation may also be used. Since interpolation is superior to extrapolation in most respects, interpolation is mostly used.

Meanwhile, in order to encode WZ pictures, the quantization unit 112 of the encoding apparatus 110 performs quantization on the WZ picture information, and the block unitization unit 113 divides the quantized values into coding units. The channel code encoder 114 divides the quantized value into predetermined channel coding units, generates compressed data for the channel code, and stores the compressed data in a buffer (not shown). In this case, the compressed data is parity data of image data corresponding to a coding unit to be compressed.

Compressed data stored in the buffer, that is, parity data is sequentially transmitted according to a request of the decoding apparatus 120 through a reverse channel (ie, a feedback channel). The channel code decoder 124 of FIG. 1 uses the parity data transmitted from the encoding apparatus 110 based on one noise model for the virtual channel generated by the noise model generator 123 to provide auxiliary information. By correcting the existing noise error, the intermediate image data value corresponding to the coding unit of the quantized image data before compression is estimated. Here, the image data before compression is mixed with the same concept as the original image. The image reconstructor 125 of FIG. 1 receives an intermediate image data value, which is a set of predetermined channel coding units output by the channel code decoder 124, and inversely quantizes the intermediate image data value to reconstruct the WZ picture.

Ambiguity generated during inverse quantization in the above process is removed by referring to the auxiliary information generated by the auxiliary information generator 122 and the noise model estimated by the noise model generator 123.

The decoding method of the Wyner-Ziv coding technique basically corrects a noise error in auxiliary information using a channel code. However, since the decoding apparatus does not have information on the pre-compression image data necessary to correct the noise error, the decoding apparatus may not know the amount of parity data necessary to correct the noise error. Therefore, the decoding apparatus has a structure for gradually requesting parity data to the encoding apparatus through a reverse channel. For a detailed description, see A. Aaron, S. Rane, R. Zhang, and B. Girod et al. In Proc. See "Wyner-Ziv coding for video: Applications to compression and error resilience" published in the IEEE Data Compression Conference, 2003.

On the other hand, accurately predicting the information on the virtual channel in the Wyner-Ziv decoding apparatus greatly affects the performance of channel decoding. However, since there is no information on the original image, the decoding apparatus cannot accurately predict a channel (noise), and accordingly, it is common to perform prediction through difference between front and rear key pictures assuming linear motion. For details, see C. Brites, and F. Pereira, Proc. See "Correlation Noise Modeling for Efficient Pixel and Transform Domain Wyner-Ziv Video Coding", a paper published in IEEE Transactions on Circuits and Systems for Video Technology, 2008.

However, since the noise model according to the conventional method is used to experimentally generate a noise model in advance, it does not guarantee good performance for each unit of compressed data input in a predetermined channel coding unit. As such, since it is not always possible to guarantee good performance, there is a lot of waste of parity data used for noise error correction during channel code decoding. In addition, it is not always possible to generate an accurate noise model, and even if one suitable noise model is generated, a channel code is generated when a noise error, which is determined to be relatively low when generating the noise model, is actually generated when the compressed data is actually recovered. There was a serious problem with decryption.

In order to solve the above problems, the present invention provides an apparatus and method for decoding a distributed video encoded image for improving decoding performance using a multiple noise model.

The present invention provides a decoding apparatus and method for decoding a distributed video encoded image. The decoding apparatus includes a key picture restoring unit for restoring key picture information transmitted from an encoding device, an auxiliary information generating unit for generating auxiliary information using the key picture restored by the key picture restoring unit, and the auxiliary information. A multiple noise model generator for generating a multiple noise model composed of a plurality of candidate noise models for estimating noise errors, and a plurality of candidate noise models generated by the multiple noise model generator for use in a predetermined channel coding unit A noise model determination unit for determining a noise model, and correcting a noise error of the auxiliary information based on the determined noise model using compressed data transmitted from the encoding apparatus, and correcting predetermined channel coding units for a given decoding unit. A channel code decoding unit to decode, and the predetermined decoded by the channel code decoding unit The board using the middle value and the image data, the noise model and the side information generated by the side information generation unit consisting of the determined coding unit comprises image restoring section for restoring the WZ picture.

In addition, the decoding method for decoding a distributed video encoded image comprises the steps of: reconstructing key picture information transmitted from an encoding device; Generating auxiliary information using the reconstructed key picture; Providing a noise model to be used for a predetermined channel coding unit based on a multiple noise model consisting of a plurality of candidate noise models for estimating noise errors in the auxiliary information; Correcting a noise error of the auxiliary information using the compressed data transmitted from the encoding apparatus based on the provided noise model; Decoding predetermined channel coding units for a predetermined decoding unit; And reconstructing a WZ picture using the decoded intermediate image data value composed of the predetermined channel coding units, the auxiliary information, and the provided noise model.

As described above, the decoding apparatus according to the present invention generates a plurality of candidate noise models that effectively reflect the characteristics of the virtual channel by using the identification information transmitted from the encoding apparatus and / or the information generated or reconstructed by the decoding apparatus. A noise model suitable for a predetermined channel coding unit input to the channel code decoding unit is selected by using information generated or restored during decoding and / or confirmation information received from the encoding apparatus. Through this, the channel code decoder significantly increases the compression performance by effectively eliminating the noise error of the auxiliary information while minimizing the waste of the compressed data.

FIG. 1 is a diagram illustrating a configuration of a coding apparatus corresponding to a conventional Wyner-Ziv coding technique and a decoding apparatus according to the same.
2 is a block diagram of a distributed video coded image which improves decoding efficiency by using a noise model adaptive to a predetermined channel coding unit using a coding apparatus of a Wyner-Ziv coding technique in a pixel region and a multiple noise model according to the present invention. It is a figure which shows the structure of an apparatus.
FIG. 3 is a diagram illustrating an example of intermediate image data and a predetermined channel coding unit that undergo a signal processing process in the encoding / decoding apparatus of FIG. 2.
4 is a diagram illustrating an example of a multiple noise model generated by a multiple noise model generator of the decoding apparatus of FIG. 2.
5 is a diagram illustrating a configuration of a channel code decoder of the decoding apparatus of FIG. 2.
FIG. 6 is a diagram illustrating a configuration of an image reconstruction unit of the decoding apparatus of FIG. 2.
7 is a block diagram of a distributed video coded image which improves decoding efficiency by using a noise model adaptive to a predetermined channel coding unit using a coding apparatus of a Wyner-Ziv coding technique and a multiple noise model according to the present invention. It is a figure which shows the structure of an apparatus.
8 is a flowchart illustrating a decoding method according to an embodiment of the present invention.
9 shows a flowchart for generating multiple noise models according to an embodiment of the present invention.
10 is a graph illustrating a performance evaluation result of a method of decoding a distributed video encoded image according to an embodiment of the present invention.

As the invention allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. Like reference numerals are used for like elements in describing each drawing.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when a component is referred to as being "directly connected" or "directly connected" to another component, it should be understood that there is no other component in between.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "having" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Hereinafter, with reference to the accompanying drawings, it will be described in detail a preferred embodiment of the present invention. Hereinafter, the same reference numerals are used for the same components in the drawings, and duplicate descriptions of the same components are omitted.

An apparatus and method for decoding a distributed video coded image according to the present invention identify a degree of noise that may occur for each of predetermined input channel coding units, generate a plurality of candidate noise models suitable for the adaptive channel coding, and adaptively select among them. By selecting and using the noise model, the compression performance of image data can be improved.

Specifically, the apparatus and method for decoding a distributed video encoded image according to the present invention generates a plurality of candidate noise models reflecting channel characteristics by using information received from an encoding apparatus and / or information generated or reconstructed by the decoding apparatus. Then, a noise model suitable for each of the predetermined channel coding units input to the channel code decoding unit is selected by using information generated or restored by the decoding apparatus and / or information received from the encoding apparatus from among a plurality of candidate noise models. By performing channel decoding, waste of the amount of compressed data transmitted is reduced.

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

2 is a diagram illustrating a configuration of a decoder 20 of a distributed video encoded image for improving decoding performance by using the Wyner-Ziv encoding apparatus 10 and the multiple noise model according to the present invention.

Referring to FIG. 2, the Wyner-Ziv encoding apparatus according to the present invention may include a key picture encoder 11 and a WZ picture encoder 12.

The encoding apparatus 120 divides the data to be compressed into two classes, one of which is a WZ picture encoded by a distributed video encoding scheme, and the other of which is a key picture encoded by a conventional encoding scheme other than the distributed video encoding scheme. to be. The key picture encoder 11 encodes the key pictures by a predetermined method selected by a user in advance, such as the H.264 / AVC intra picture encoding method, and transmits the encoded pictures to the decoding apparatus 120.

The WZ picture encoder 12 performs quantization on the WZ picture information, and divides the quantized values into coding units. The coding unit is a unit for encoding and decoding an image. The coding unit includes a frame or a block having various sizes subdivided into frames, and includes a pixel region or a transform region obtained by using a transform such as a discrete cosine transform (DCT). Include. In addition, the coding unit becomes a unit to be decoded by the decoding apparatus and is used interchangeably in the same meaning as the term decoding unit.

In addition, the WZ picture encoder 12 generates compressed data about a quantized value using a channel code in a predetermined channel coding unit and stores the compressed data in a buffer (not shown). In this case, the compressed data is parity data of image data corresponding to a coding unit to be compressed.

3 is a diagram illustrating an example of intermediate image data and a predetermined channel coding unit that undergo a signal processing in the encoding apparatus of FIG. 2.

Referring to FIG. 3, a frame is composed of one or more blocks of a predetermined size, wherein pixel values or transform region coefficient values of the same components in and / or between blocks are quantized.

The quantization does not need to be the same size between predetermined channel coding units and within a predetermined channel coding unit, and also includes a case where quantization is not performed. In addition, data enumeration is performed according to a predetermined scan method such as a zigzag scan method or an alternate scan for pixel values or transform region coefficient values of the same component in and between blocks, which is not essential.

A plurality of quantized pixel values or transform region coefficient values are listed in a symbol stream. At this time, each symbol value (quantized pixel value or transform region coefficient value) is represented as bits having different weights of 2 ^M when binarized, and the symbol stream is a bit stream having a same or different M value (hereinafter, Mixed with the term bit plane). In this case, the bit plane having the maximum value M is referred to as the most significant bit plane, and the bit plane having the minimum value M is referred to as the least significant bit plane for each symbol constituting the symbol stream. This also includes the case where no quantization is performed in which M of the least significant bit plane of each symbol constituting the symbol stream is zero.

In the present invention, the channel coding unit refers to a data unit for generating compressed data, that is, parity data, by the WZ picture encoder 12. The channel coding unit subdivides a block or a coding unit having various sizes, which are divided into frames or frames, which are coding units. One bitplane, or part of a bitplane, of various lengths. In addition, the image data generating the parity data includes a value in the pixel region or a coefficient value in the transform region obtained by using a transform such as a discrete cosine transform (DCT).

The intermediate image data is a data value before the image is restored to a set of predetermined channel coding units. According to the unit (the coding unit) in which the image restoration unit 25 performs image restoration, the image restoration unit 25 may be a frame having various quantized pixel values or blocks having various sizes, which are not only values in the pixel region but also DCT. Contains the coefficient values in the transform domain obtained using a transform such as

Further, the present invention further subdivides the channel coding unit to define the minimum structural unit in the channel coding unit. This means the minimum unit for calculating the cross-over probability in the channel coding unit. The minimum configuration unit includes a pixel value constituting a frame or a block having various sizes subdivided into frames according to a channel coding unit input to the channel code decoder, or a bit constituting a bit plane of various lengths. The minimum constituent unit also includes a value in the pixel region or a coefficient value in the transform region obtained by using a transform such as DCT.

Referring back to FIG. 2, the decoding apparatus 20 according to the present invention includes a key picture reconstructing unit 21, an auxiliary information generating unit 22, a multiple noise model based noise model providing unit 30, and a channel code decoding unit ( 24, and an image restoring unit 25.

The key picture reconstructor 21 reconstructs the key picture using the information received from the key picture encoder 11. The auxiliary information generator 22 generates auxiliary information that is regarded as a current WZ picture by using the restored key pictures. At this time, the generated auxiliary information is regarded as a noise error due to a virtual channel is added to the original image corresponding to a predetermined channel coding unit. Prediction of the noise present in the virtual channel is necessary for channel code decoding for the purpose of eliminating this noise error using error correction techniques.

The noise model provider 30 based on the multiple noise model provides a noise model to be used for a predetermined channel coding unit based on the multiple noise model including a plurality of candidate noise models for estimating noise errors in the auxiliary information.

To this end, the multi-noise model based noise model providing unit 30 includes a multi-noise model generation unit 23 and a noise model determination unit 26.

The multi-noise model generation unit 23 generates N candidate noise models for noise existing in a predetermined channel coding unit by using information generated or restored by the decoding apparatus and / or confirmation information transmitted by the encoding apparatus.

Here, the confirmation information refers to all or part of the general / statistical information of the original image (ie, image data) that can be estimated or obtained by the decoding apparatus, and a CRC (Cyclic Redundancy Check) value of the original image is one example. .

The decoding apparatus 20 of the distributed video encoded image may not know accurate noise information. Therefore, the multi-noise model generation unit 23 generates noise for each bit constituting the bit plane which is the minimum component unit in the predetermined channel coding unit, in order to generate a plurality of candidate noise models for estimating the noise error in the auxiliary information. A candidate noise model is generated through parameters for compensating for probability distribution (PMF), noise distribution characteristics (stationary / non-stationary), and actual noise information estimation error.

Some examples of noise distribution PMF for generating the noise models include Laplacian, Gaussian, and Two-sided Gamma distributions.

The noise models are estimated according to the noise distribution characteristics in the channel coding unit for each of the minimum component units in the channel coding unit to estimate the noise of the non-stationary noise model and the components in the channel coding unit. In addition, the stationary noise model includes a stationary noise model. In addition, there may exist distinctions based on other distribution characteristics, and it does not deviate from the essential meaning of the present patent which generates several noise models by using the same.

The parameter refers to various correction variables that are used to correct the noise prediction performed by the decoding apparatus. Hereinafter, the operation of the multi-noise model generation unit 23 will be described in more detail with reference to FIG. 4. For ease of explanation, it is assumed that parity data is generated for the image data of the transform region, and that bit planes of the same frequency band (same transform region coefficient value) quantized in one frame are used as predetermined channel coding units.

The multi-noise model generation unit 23 obtains the difference between the corresponding key picture blocks in order to predict the noise on the auxiliary information generated for the bit plane which is a predetermined channel coding unit. For example, the multi-noise model generation unit 23 finds a corresponding block on the front-back key picture indicated by the motion vector for the front-back key picture referred to when the auxiliary information generation unit 22 of the decoding apparatus generates the auxiliary information, and the same frequency of these blocks. The difference between the bands is predicted as noise (310 in FIG. 4). Here, one block may be a channel coding unit used in the encoding apparatus.

For example, the multi-noise model generation unit 23 generates an error block between blocks, for example, an i-th block, performs transformation on the error block, and obtains the variance thereof. This variance becomes the variance of the noise located in the i-th block. Subsequently, when generating the candidate noise models, the multi-noise model generation unit 23 considers at least a probability distribution function (PMF) of the channel noise and / or a noise distribution characteristic of the noise model.

Specifically, in relation to the probability distribution of the channel noise, the multi-noise model generation unit 23 generates a noise model according to the noise probability distribution (PMF) for each of the bits constituting the bit plane that is the minimum component unit within the predetermined channel coding unit. (320 in FIG. 4).

As described above, some examples of noise probability distributions (PMF) for generating the candidate noise models include Laplacian, Gaussian, and Two-sided Gamma distributions.

라고 하면, 다중잡음모델 생성부(23)는 라플라시안 분포에 따라, 하기 [수학식 1]에 의하여 소정 채널코딩단위 내 최소구성단위에 대한 잡음모델을 생성할 수 있다.Among the plurality of noise probability distributions (PMFs), for example, the Laplacian distribution will be described. The variance of the noise located in the i-th block constituting the predetermined channel coding unit generated by the noise prediction

In this case, the multi-noise model generation unit 23 may generate a noise model for the smallest unit within a predetermined channel coding unit according to Equation 1 according to the Laplacian distribution.

는 소정 채널코딩단위 중i번째 블록의 라플라시안 계수값을 의미한다. In Equation 1, x means noise,

Denotes the Laplacian coefficient value of the i-th block of the predetermined channel coding unit.

을 구함으로써 잡음을 예측하고, 예측된 잡음 즉, 비트 각각에 대한 분산값들에 대해 라플라시안 분포에 따른 잡음모델을 생성한다. In this way, the multi-noise model generation section 23 has a variance value for each bit in the bit plane.

By estimating the noise, we generate a noise model based on the Laplacian distribution for the predicted noise, that is, the variance values for each bit.

In this manner, the multi-noise model generator 23 may generate noise models using a Gaussian distribution, a two-sided gamma distribution, and the like in addition to the Laplacian distribution.

The noise probability distribution (PMF) that is considered when generating the candidate noise models determines the distribution form of noise, and the probability distribution is used to determine the crossover probability for the smallest unit within a predetermined channel coding unit to be decoded. Calculate the cross-over probability. For example, when one bit plane constituting the auxiliary information is a predetermined channel coding unit, when decoding the bit plane, each bit constituting the bit plane becomes a minimum constituent unit in the predetermined channel coding unit and calculates an intersection probability thereof. .

Subsequently, the multiple noise model generator 23 further generates additional noise models according to the noise distribution characteristic of the noise model generated in relation to the noise distribution characteristic (330 of FIG. 4). In detail, the multi-noise model generation unit 23 determines whether the stationary / non-stationary characteristics of the generated noise model are based on the noise probability distribution. In addition, the multi-noise model generation unit 23 further generates a noise model for a characteristic that the noise model does not have among the stationary and non-stationary characteristics.

For example, the noise model generated in the above embodiment has a noise model configured by predicting noise for each of the minimum component units in the channel coding unit, and thus the generated noise model has the characteristics of a non-stationary noise model. The multiple noise model generator 23 further generates a noise model based on the generated noise model to have a stationary characteristic.

) 하나만을 사용한 잡음모델을 적용하면 stationary 특성을 갖는 잡음모델을 생성할 수 있다. 여기서 B는 비트 플레인을 구성하는 비트를 나타내기 위함이 아니며 해당 비트 플레인을 구성하는 각각의 블록에 위치한 잡음을 샘플로 하는 랜덤변수의 분산값을 나타내고 있음을 의미한다.For example, the multi-noise model generation unit 23 uses the noise estimation for the entire bit plane to determine the noise variance value of the block in the same channel coding unit for all the blocks in the bit plane.

By applying a noise model using only one, a noise model with stationary characteristics can be generated. Here, B is not intended to represent the bits constituting the bit plane, but means that the variance value of the random variable that samples the noise located in each block constituting the bit plane is sampled.

Stationary noise distribution by determining whether the model is stationary or non-stationary means that a single cross-over probability is used for the minimum component units within a given channel coding unit, and non-stationary noise Distribution means using the respective crossing probabilities for the smallest unit within a predetermined channel coding unit. In general, the performance of the non-stationary noise model is good, but if the noise prediction is very inaccurate, the use of the stationary noise model is rather beneficial.

Since the noise prediction is used to estimate the actual noise by using the difference of the inter-block frequency bands on the front and back key pictures corresponding to the estimated motion vector, the estimation of the motion vector is not accurate or the noise prediction method is not suitable. If not, there is a high possibility of false noise prediction, and it is desirable to correct this effectively.

Therefore, the multi-noise model generator 23 generates several noise models according to the noise probability distribution and the noise distribution characteristics, and then changes the generated noise models to compensate for the actual noise information estimation error. Create

In detail, candidate noise models generated through prediction may be different from actual noise models, and predetermined parameters may be defined and used to generate candidate noise models for the purpose of quantifying the difference and later compensating for the difference. For example, a scaling factor that may occur while generating a noise model may be defined and used.

As such, since the noise prediction is not accurate noise prediction, the multiple noise model generator 23 generates a plurality of noise generated to compensate for the actual noise information estimation error based on the noise prediction value (the statistical value of the noise). A plurality of candidate noise models are generated by changing the noise models.

For example, the multiple noise model generator 23 generates a plurality of candidate noise models by applying the generated noise models while changing a parameter for compensating an actual noise information estimation error (340 of FIG. 4).

As an example of such correction, there may be a case where a block having a different prediction value is found and corrected compared to the surrounding noise prediction value. As another embodiment, the correction may be performed based on reliability of each block of the generated auxiliary information, confirmation information transmitted from the encoding apparatus, and similarity information between neighboring blocks based on the motion vector estimated by the decoding apparatus.

)을 사용하여 계산할 경우, 다음의 세 가지의 경우가 존재할 수 있다. Hereinafter, an embodiment of correction through comparison with the above-described ambient noise prediction value will be described. The noise estimates of the neighboring blocks are estimated from the variance of the noise

), There are three possible cases:

)을 사용하여 계산할 경우, 다음과 같이 예측 잡음을 정정한다.The multi-noise model generation unit 23 calculates the noise estimates of the neighboring blocks by the variance value of the noise present in the bit plane.

), Correct the prediction noise as follows.

의 경우:

를

로 정정a)

In the case of:

To

Corrected by

의 경우:

를 임계범위(

) 내의 값으로 정정b)

In the case of:

Is the critical range (

To the value in

의 경우:

를

로 정정c)

In the case of:

To

Corrected by

)와 stationary의 특성을 갖는 잡음모델에서 사용하는 비트 플레인의 잡음 예측치(

)는 실제 원본영상과 보조정보의 차를 통해 얻어진 값이 아니므로, 이를 보상하기 위한 보상계수(

, Compensation factor)가 필요하다. At this time, the ambient noise prediction value used in the noise model having a non-stationary characteristic (in the case of the above embodiment)

) And noise prediction of the bit plane used in the noise model with stationary characteristics

) Is not a value obtained through the difference between the original image and the supplementary information, so a compensation coefficient (

, A compensation factor is required.

라 하고, 라플라시안(Laplacian) 분포를 따른다면 [수학식 2]의 조건을 만족하는 보상계수를 사용한다.Accordingly, the multi-noise model generation unit 23 calculates the variance of the actual noise present in the bit plane.

If the Laplacian distribution is followed, a compensation factor that satisfies the condition of [Equation 2] is used.

를 변화시키면서 생성된 후보 잡음 모델들에 적용함으로써 복수개의 최종 후보 잡음 모델들을 생성한다. 따라서, 라플라시안(Laplacian) 잡음모델에 대한 후보 잡음모델은 [수학식 2]를 만족하는

의 값의 범위에 대하여 생성된다. Therefore, the multi-noise model generation unit 23 is a compensation coefficient that satisfies the condition of [Equation 2] to compensate for the actual noise information estimation error

A plurality of final candidate noise models are generated by applying to candidate noise models generated while varying. Therefore, the candidate noise model for the Laplacian noise model satisfies [Equation 2].

Generated for a range of values.

의 값을 알 수는 없으나, 적어도 [수학식 2]를 만족하는

의 값의 범위는 알 수 있다. 따라서, 다중잡음모델 생성부(23)는

값의 최소값에서 최대값 사이의 값을

에 대입함으로써 여러 개의 최종 후보 잡음모델들을 생성할 수 있으며, 이는 Non-stationary/Stationary 특성 어느 경우에 상관없이 두 경우 모두에 대하여 적용할 수 있다. 도 4에서는

은 0, 0.01 및 0.02 등에 대해 최종 후보 잡음 모델들이 생성되는 경우를 예시적으로 나타낸다.Accurate compensation factor that satisfies Equation 2 due to absence of original information in distributed video decoding apparatus

Is unknown, but at least satisfies Equation 2

The range of values is known. Therefore, the multi-noise model generation unit 23

Value between the minimum and maximum values

Multiple final candidate noise models can be generated by substituting for, which can be applied to both cases regardless of non-stationary / stationary characteristics. In Figure 4

Exemplarily illustrates a case where final candidate noise models are generated for 0, 0.01, 0.02, and the like.

는 라플라시안 잡음분포에 해당하는 하나의 예로써 다른 PDF에서도 이와 유사한 파라메터가 사용될 수 있으므로, 위의 보상계수

이외의 다른 파라메터를 이용하여 실제잡음모델과 생성된 잡음모델사이의 차이를 보상하는 것도 역시 가능하다. 따라서, 실제잡음모델과 생성된 잡음모델사이의 값의 보상을 수행하기 위하여 다른 파라메터 값을 사용하는 것은 본 특허 기본 사상에 어긋나는 것이 아니며, 본 특허의 기본 사상을 구현하는 개개의 특수한 예의 범위를 벗어나지 않는다.At this time, the compensation factor

Is an example of the Laplacian noise distribution, and similar compensation parameters can be used in other PDFs.

It is also possible to use other parameters to compensate for the difference between the actual noise model and the generated noise model. Therefore, the use of other parameter values to compensate for the value between the actual noise model and the generated noise model is not contrary to the basic idea of the present patent, and does not depart from the scope of each particular example of implementing the basic idea of the present patent. Do not.

In this case, the compensation factor is an example corresponding to the Laplacian noise distribution, and similar parameters may be used in other PDFs. Therefore, the compensation factor compensates for the difference between the actual noise model and the generated noise model by using a parameter other than the above compensation factor. It is also possible. Therefore, the use of other parameter values to compensate for the value between the actual noise model and the generated noise model is not contrary to the basic idea of the present patent, and does not depart from the scope of each particular example of implementing the basic idea of the present patent. Do not.

In this way, several candidate noise models can be generated for different PMFs, and a multiple noise model is generated by collecting all the candidate noise models. There may be a number of ways to implement a method for generating a multi-noise model, and thus the present invention is not limited to any range by the above embodiments.

As such, the multiple noise model generator 23 generates a plurality of candidate noise models for estimating noise error in the auxiliary information and provides these candidate noise models to the noise model determiner 26.

The noise model determiner 26 determines a noise model suitable for decoding a current channel coding unit among a plurality of generated candidate noise models or reduces the number of candidate noise models.

The channel code decoder 24 corrects a noise error of the auxiliary information generated by the auxiliary information generator based on the determined noise model by using the compressed data transmitted from the encoding apparatus, and determines a predetermined decoding unit for a given decoding unit. Decode the channel coding units.

The configuration and operation of the channel code decoder 24 will be described with reference to FIG.

5 is a diagram illustrating a configuration of a channel code decoder of the decoding apparatus of FIG. 2.

Referring to FIG. 5, the channel code decoder 24 gradually receives compressed data (ie, parity data) from the encoding apparatus 10 based on the noise model selected by the noise model determiner 26. Channel code decoding for measuring channel code decoding reliability using a channel code decoder 510 for removing noise of information and a soft decision output value output from the channel code decoder 510 and / or confirmation information transmitted from the encoder. The reliability calculator 520 is included. Here, the channel code decoding reliability calculator 520 measures the decoding reliability using the soft decision output value output from the channel code decoder and / or identification information such as cyclic redundancy code (CRC) information transmitted from the encoder.

과 비교하여 임계값 보다 크면 그 복호화 결과를 신뢰하고 낮으면 신뢰하지 않는 방법으로 이루어진다. 이는 연판정 출력값의 경우 신뢰도가 낮을 수 있으므로 출력값을 바로 이용하기 보다는 일정이상의 신뢰도를 보이는 경우만 그 값을 이용하는 것이며, 이를 통하여 복호화 결과에 대한 에러율 산출이 가능하다. 이러한 방법은 복호화 결과의 신뢰도를 측정하는 하나의 예로 상기와 같은 실시 예 이외에도 다양한 접근이 가능하다.Reliability measurements with soft decision outputs threshold the soft decision outputs.

If it is larger than the threshold, the decoding result is trusted and if it is low, it is not trusted. In the case of the soft decision output value, the reliability may be low. Therefore, the value of the soft decision output value is used only when a certain level of reliability is shown, rather than using the output value immediately. This method is one example of measuring the reliability of the decoding result, various approaches are possible in addition to the above-described embodiment.

In this case, the noise model determiner 26 determines a noise model suitable for a given channel coding unit among a plurality of candidate noise models generated by the multiple noise model generator 23. In detail, the noise model determiner 26 is suitable for decoding a current channel coding unit among a plurality of candidate noise models generated based on information generated or restored by the decoding apparatus and / or confirmation information transmitted by the encoding apparatus. Choose a noise model or reduce the number of candidate noise models.

An embodiment thereof is as follows.

Example  One

보다 신뢰도가 높은 값들은 신뢰하고 낮은 값들은 신뢰하지 않는 방법으로 채널 복호화 결과(연판정 출력값)의 에러율을 계산한다. 이 때 잡음모델 결정부(26)에서는 수신된 패리티데이터로 채널 복호화가 이루어진 후 추정된 에러율을 바탕으로 1~N번의 잡음모델 중 수렴속도가 빠른 순으로 잡음모델을 추려나간다. 최종적으로 먼저 목표 에러율 이하의 추정된 에러율을 달성한 잡음모델이 소정 채널코딩단위에 대한 최적의 잡음모델이 된다. 이 때 최적의 잡음모델은 하나이상이 될 수도 있으며 소정 채널코딩단위에 대한 잡음모델로 모두 결정이 된다.Since the channel code decoder 510 gradually receives the parity data, the channel code decoder 510 receives the first parity data and performs channel decoding using the multiple noise model. In this case, the first parity data means parity data received first when the parity data is gradually divided into several times and additional parity data is received at the request of the decoding apparatus. If the soft decision output value is obtained as a result of channel decoding after the parity data is received, the channel code decoding reliability calculation unit 520 measures the reliability to determine a threshold value.

The error rate of the channel decoding result (soft decision output value) is calculated in such a way that the higher reliability values are reliable and the lower reliability values are not. At this time, the noise model determiner 26 deducts the noise model in the order of convergence speed among the 1 ~ N noise models based on the estimated error rate after channel decoding is performed on the received parity data. Finally, the noise model that first achieves the estimated error rate below the target error rate is the optimum noise model for a given channel coding unit. At this time, there may be more than one optimal noise model, and all of them are determined by the noise model for a given channel coding unit.

Example  2

보다 신뢰도가 높은 값들을 신뢰하고 낮은 값들은 신뢰하지 않는 방법으로 채널 복호화 결과의 에러율을 계산한다. 이 때 추정된 에러율이 일정 에러율 이하를 갖게 되면 부호화기에서 전송된 CRC정보를 사용하여 CRC검사를 통하여 채널 복호화 결과의 오류 여부를 결정하게 된다. 이때 잡음모델 결정부(26)에서는 1~N번의 복수개의 후보 잡음모델들 각각을 사용하여 독립적으로 복호화하여 최소의 패리티데이터를 사용하여 CRC 검사를 통과한 잡음모델을 소정 채널코딩대상에 대한 최적의 잡음모델로 선택한다. 이 때 최적의 잡음모델은 하나이상이 될 수도 있으며 소정 채널코딩단위에 대한 잡음모델로 모두 결정이 된다.Since the channel code decoder 510 gradually receives the parity data, the channel code decoder 510 receives the first parity data and performs channel decoding using the multiple noise model. In this case, the first parity data means parity data received first when the parity data is gradually divided into several times and additional parity data is received at the request of the decoding apparatus. After receiving the parity data, if the soft decision output value is obtained as a result of channel decoding, the channel code decoding reliability calculation unit 520 measures the reliability to determine the threshold value.

The error rate of the channel decoding result is calculated by trusting more reliable values and not trusting lower values. At this time, if the estimated error rate has a predetermined error rate or less, it is determined whether an error of the channel decoding result is performed through the CRC check using the CRC information transmitted from the encoder. At this time, the noise model determiner 26 independently decodes each of a plurality of candidate noise models from 1 to N times, and then optimizes the noise model that has passed the CRC check using the minimum parity data for a predetermined channel coding object. Select to noise model. At this time, there may be more than one optimal noise model, and all of them are determined by the noise model for a given channel coding unit.

Example  3

)를 사용 하였을 때 이를 실제 잡음의 분산값()과 동일한 수준으로 만들기 위한 보상 계수이다. 따라서, 잡음모델 결정부(26)는 부호화 장치(10)로부터 부호화 단위에 해당하는 원본영상의 통계적인 데이터(

)를 전송 받은 후, [수학식 3]에 의한 실제의 분산값 추정이 가능하며, 이를 이용하여 사용할 다중잡음모델의 범위를 줄이거나 잡음모델을 선택 할 수 있다. 이 때, 사용 예에 따라 [수학식3]에서 사용한 [가정1], [가정2]는 바뀔 수 있으며 부호화 장치에서 전송된 확인정보를 사용한다는 점에서 본 특허의 본질적인 내용을 벗어나지 않는다.The noise model determiner 26 selects a noise model suitable for decoding a current channel coding unit from among a plurality of candidate noise models generated based on the confirmation information transmitted from the encoding apparatus 10, or selects the number of candidate noise models. Reduce For example, in the multi-noise model generation embodiment of the multi-noise model generation unit 23, it is used in the generation of candidate noise models based on the Laplacian distribution. The value is the noise prediction of the bit plane (

) Is a compensation factor to make it equal to the variance of the actual noise (). Accordingly, the noise model determiner 26 receives statistical data (the original image corresponding to the coding unit) from the encoding apparatus 10 (

After receiving), it is possible to estimate the actual variance value by [Equation 3], and use it to reduce the range of multiple noise model or select the noise model. At this time, [Assumption 1], [Assumption 2] used in [Equation 3] according to the use example can be changed and does not deviate from the essential contents of the present patent in that it uses the confirmation information transmitted from the encoding device.

E [x]: expected value, mean of x values

X: Original information of WZ picture corresponding to decoding unit

Y: information of auxiliary information corresponding to decoding unit

COV (X, Y): Covariance of X and Y

[Home 1]

E [N] = 0: In the Laplacian hypothesis, the average for noise is 0, and information generated or restored by the decoding device, such as the average value of the noise prediction value of the predicted bit plane, and / or other confirmation information transmitted from the encoding device. May be used and this patent is not limited to the present assumption.

[Home 2]

COV (N, X) = 0: In the auxiliary information image, there is no correlation between noise and the original image, so the value of covariance is 0, and information generated or restored by the decoding device and / or other information transmitted by the encoding device according to use. The identification information may use other identification information, and this patent is not limited to the present assumption.

In addition to the Laplacian noise distribution, it is also possible to use the confirmation information transmitted from the coding device to select and range-limit multiple noise models of other noise distributions, such as Gaussian and Two-sided Gamma distributions.

Other embodiments may exist and do not depart from the object of the present invention of selecting an optimal noise model for a given channel coding unit among multiple noise models.

As described above, the channel code decoder 24 decodes the predetermined channel coding units for the given decoding unit and outputs the decoded channel coding units to the image reconstruction unit 25.

The image reconstructor 25 uses a WZ picture by using an intermediate image data value composed of the predetermined channel coding units decoded by the channel code decoder, the auxiliary information generated by the auxiliary information generator, and the determined noise models. Restore it.

The configuration and operation of the image restoration unit 25 will be described with reference to FIG. 6.

FIG. 6 is a diagram illustrating a configuration of an image reconstruction unit of the decoding apparatus of FIG. 2.

Referring to FIG. 6, the image reconstructor 25 stores a predetermined channel coding unit output from the channel code decoder 24 and outputs intermediate image data including the data buffer 610 and the noise model. The decision unit 26 includes a noise model buffer 620 that stores the noise model determined when the predetermined channel coding unit is decoded. The image restoration unit 25 may further include an image restoration noise model determination unit 630 for determining a noise model most suitable for image restoration among the determined noise models stored in the noise model buffer, and the intermediate image data output from the data buffer. And an image restorer 640 for restoring the image using the auxiliary information and the noise model determined by the image restoration noise model determiner.

Here, the image restoration noise model determiner 630 determines a noise model that is most advantageous when reconstructing a coding unit among the noise models stored in the noise model buffer. There are many ways to make a decision. In order to explain this in more detail, there are four different embodiments of the following determination method for the case where a bit plane of the same frequency band in a conversion region is used as a predetermined channel coding unit. There may also exist methods that deviate from the following embodiments, and the decision made by other methods does not deviate from the basic idea of the present patent, and does not depart from the scope of individual specific examples for implementing the basic idea of the present patent.

1. Since the most significant bit plane occupies the symbol stream of FIG. 3, the noise model determined when decoding the most significant bit plane among the noise models stored in the noise model buffer is determined.

2. After estimating the distribution of noise of the supplementary information by comparing the supplementary information corresponding to the intermediate image data and the intermediate image data, the noise corresponding to the difference between the intermediate image data and the supplementary information among the noise models stored in the noise model buffer is estimated. By well estimating, we determine the noise model that is expected to reduce the mean squared error (MSE) during image restoration. At this time, the MSE is an example of evaluating the image quality of the reconstructed image and can be evaluated by other methods, which does not deviate from the basic idea of the present patent.

3. The stationary noise model is not suitable for reconstructing the image of each block because the noise prediction for each of the minimum constituent units within a given channel coding unit is not made. Therefore, the noise model determined when decoding the most significant bit plane of the non-stationary noise model is used.

4. When referring to FIG. 3, the least significant bit plane means a bit plane in which M has a minimum value in each symbol constituting the symbol stream or the quantized original image, and the least significant bit plane of the least significant bit plane (Example 1) and (Example 2). Includes cases. Here, when M is a symbol value (quantized pixel value or transform region coefficient value) quantized to ^2M , M means the least significant bit of the quantized value of the bit located from the lowest bit of the symbol. Therefore, since the data represent the smallest value in the symbols of the intermediate reconstructed image, the noise model determined when decoding the least significant bit plane that does not fail channel decoding among the noise models stored in the noise model buffer is used.

Other embodiments may exist and do not depart from the object of the present invention of selecting an optimal noise model for reconstructing a decoding unit among multiple noise models.

In the exemplary embodiment of the present invention, examples of the pixel region and the conversion region are given, which are different from those of FIG. 2, but are illustrated by adding the transform unit 27 and the inverse transform unit 28 as shown in FIG. 7. This is not to be described separately without departing from the spirit of the invention. In addition, various configurations are possible.

Therefore, the coding unit described in the description of the present invention includes not only the pixel region but also the transform coefficients in the integer transform, DCT transform or wavelet transform according to the implementation of the present invention. In this case, the transform unit 27 and the inverse transform unit in FIG. This can be implemented as (28).

In the embodiment according to the present invention, a general WZ decoding process is dealt with, but a decoding apparatus of a distributed coded video image of various methods may also apply a multi-noise model to decode the original image during the decoding process. Do not escape.

As described above, noise estimation and optimal noise selection based on a multiple noise model in the Wyner-Ziv coding technique according to the preferred embodiment of the present invention can be made.

8 is a flowchart illustrating a decoding method according to an embodiment of the present invention.

In operation 810, the decoding apparatus restores key picture information transmitted from the encoding apparatus. In operation 820, the decoding apparatus generates auxiliary information using the restored key picture.

In operation 830, the decoding apparatus generates a multiple noise model including a plurality of candidate noise models for estimating noise errors in the auxiliary information. In operation 840, the decoding apparatus determines a noise model to be used for a predetermined channel coding unit among the plurality of candidate noise models. In one embodiment, after decoding the channel with the parity data received from the encoding apparatus, the decoding apparatus selects the noise models in the order of fast convergence speed from 1 to N candidate noise models based on the estimated error rate.

In another embodiment, the decoding apparatus independently decodes each of the plurality of candidate noise models to select a noise model that passes the CRC check using the minimum parity data as an optimal noise model for a predetermined channel coding object. . In another embodiment, the decoding apparatus receives the identification information transmitted by the encoding apparatus, for example, statistical data of the original image corresponding to the coding unit, and then performs an actual variance estimation and uses the multiple noise to use the same. You can narrow the model or select a noise model.

After selecting the noise model to be used, the decoding apparatus corrects the noise error of the auxiliary information by using the compressed data transmitted from the encoding apparatus based on the noise model determined in operation 850, and in step 860, the predetermined channel coding unit for the decoding unit. Decrypt them.

Subsequently, the decoding apparatus reconstructs the WZ picture by using the intermediate image data value composed of the predetermined channel coding units decoded in operation 870, the auxiliary information generated by the auxiliary information generator, and the determined noise models.

The generation of the multi-noise model composed of the candidate noise models will be described with reference to FIG.

9 shows a flowchart for generating multiple noise models according to an embodiment of the present invention.

In order to generate a multiple noise model, the decoding apparatus predicts noise using a difference between key pictures before and after the auxiliary information in step 910. Specifically, the decoding apparatus obtains the difference between the corresponding key picture blocks in order to predict noise on the auxiliary information generated for the bit plane which is a predetermined channel coding unit. For example, the decoding apparatus searches for a corresponding block on the front and back key pictures indicated by the motion vector for the front and back key pictures referenced in generating the auxiliary information, and predicts the difference between the same frequency bands of the blocks as noise.

In operation 920, the decoding apparatus generates a noise model based on the noise probability distribution based on the predicted noise. In one embodiment, the decoding apparatus generates a noise model according to the noise probability distribution (PMF) for each of the bits constituting the bit plane that is the smallest constituent unit in the predetermined channel coding unit. As described above, some examples of noise probability distributions (PMF) for generating the candidate noise models include Laplacian, Gaussian, and Two-sided Gamma distributions.

Subsequently, the decoding apparatus further generates additional noise models according to the noise distribution characteristic of the noise model generated in step 930. In detail, the decoding apparatus determines whether the stationary / non-stationary characteristics of the generated noise model are based on the noise probability distribution. The decoding apparatus further generates a noise model for a characteristic that the noise model does not have among the stationary and non-stationary characteristics.

For example, the decoding apparatus predicts noise for each of the minimum constituent units in the channel coding unit, and if the noise model is configured, it has a non-stationary characteristic. Configure.

Stationary noise distribution by determining whether the model is stationary or non-stationary means that a single cross-over probability is used for the minimum component units within a given channel coding unit, and non-stationary noise Distribution means using the respective crossing probabilities for the smallest unit within a predetermined channel coding unit. In general, the performance of the non-stationary noise model is good, but if the noise prediction is very inaccurate, the use of the stationary noise model is rather beneficial.

Subsequently, the decoding apparatus generates final candidate noise models by changing the generated noise models to compensate for the actual noise information estimation error in step 940.

As described above, the noise prediction is used to estimate the actual noise by using the difference between the inter-block frequency bands on the front and back key pictures corresponding to the estimated motion vector, so the estimation of the motion vector is not accurate or such noise prediction If the method does not fit well, there is a high possibility of false noise prediction, and it is desirable to correct it effectively. Therefore, the decoding apparatus may define the predetermined parameters for the purpose of quantifying the difference between the candidate noise models generated through the prediction and the actual noise model, and later compensate for this, and use them in generating the candidate noise model. For example, a scaling factor that may occur while generating a noise model may be defined and used.

For example, the decoding apparatus generates a plurality of candidate noise models by applying the generated noise models while changing a parameter value for compensating an actual noise information estimation error.

Therefore, the use of other parameter values to compensate for the value between the actual noise model and the generated noise model is not contrary to the basic idea of the present patent, and does not depart from the scope of each particular example of implementing the basic idea of the present patent. Do not.

In this way, several candidate noise models can be generated for different PMFs, and a multiple noise model is generated by collecting all the candidate noise models. There may be a number of ways to implement a method for generating a multi-noise model, and thus the present invention is not limited to any range by the above embodiments.

10 is a graph illustrating a performance evaluation result of a method of decoding a distributed video encoded image according to an embodiment of the present invention.

In the present invention, a simulation was conducted using Foreman, Stefan, Salesman, and Mobile & calendar sequences to evaluate the performance. The images used in the experiment are 150 frames each, except that the Salesman sequence is 225 frames, the size is QCIF, and the frame rate is 15 Hz. Through the simulation, the noise model generation method according to an embodiment of the present invention and the noise model are compared with each other in the rate-distortion performance exemption and the results are analyzed.

Referring to FIG. 10, the rate-distortion (Rate-PSNR) graph results show that the decrease of the relative rate is larger than that of the high QP (low QP), which is a function of the noise model selection method. Rather than focusing on the bit plane, we reduce the rate slightly over the entire bit plane, so we can see a greater effect on the upper bit plane, which consumes less parity information for noise reduction. For the same reason, the proposed noise model selection method shows better performance in the sequence with many linear motions that are easy to encode and decode.

Claims (20)

A decoding apparatus for decoding a distributed video encoded image,
A key picture reconstruction unit for reconstructing key picture information transmitted from the encoding apparatus;
An auxiliary information generating unit generating auxiliary information using the key picture restored by the key picture restoring unit;
A multi-noise model-based noise model providing unit for providing a noise model for use in a predetermined channel coding unit based on the multiple-noise model composed of a plurality of candidate noise models for estimating noise errors in the auxiliary information;
A channel code decoder for correcting a noise error of the auxiliary information based on the provided noise model using compressed data transmitted from the encoding apparatus, and decoding predetermined channel coding units for a given decoding unit;
And an image reconstruction unit for reconstructing a WZ picture by using the intermediate image data value composed of the predetermined channel coding units decoded by the channel code decoder, the auxiliary information generated by the auxiliary information generator, and the provided noise models. Decoding apparatus, characterized in that. The method of claim 1, wherein the multiple noise model based noise model providing unit,
A multiple noise model generator for generating a multiple noise model composed of a plurality of candidate noise models for estimating noise errors in the auxiliary information;
And a noise model determiner configured to determine a noise model to be used for a predetermined channel coding unit among a plurality of candidate noise models generated by the multiple noise model generator. The noise model of claim 1, wherein the multiple noise model generator predicts noise by using differences between key pictures before and after the auxiliary information, generates a noise model based on a noise probability distribution based on prediction noise, and generates the noise. Decoding apparatus, characterized in that for generating additional noise models further in accordance with the noise distribution characteristics of the model, 3. The method of claim 2, wherein the multiple noise model generator determines whether the generated noise model is a stationary / non-stationary characteristic according to the noise distribution characteristic of the generated noise model, and wherein the multiple noise model generator determines whether the generated noise model is a stationary / non-stationary characteristic. And a noise model is further generated for a characteristic that the generated noise model does not have. The method of claim 3, And the multiple noise model generator generates a plurality of candidate noise models by applying the generated noise models while changing a parameter value for compensating an actual noise information estimation error. 3. The decoding apparatus of claim 2, wherein the noise probability distribution comprises at least one of Laplacian, Gaussian, generalized gaussian, and Two-sided Gamma distributions. The decoding method of claim 1, wherein the noise model determiner selects the noise models in the order of convergence speed among candidate noise models based on the estimated error rate after channel decoding is performed on the parity data received from the encoding apparatus. Device. The decoding method of claim 1, wherein the noise model determiner selects a noise model that has passed a CRC test using minimum parity data after being independently decoded using each of the plurality of candidate noise models. Device. The method of claim 1, wherein the channel code decoding unit
A channel code decoder for correcting a noise error in auxiliary information corresponding to the predetermined channel coding unit by using the compressed data transmitted from the encoding apparatus and the noise model selected by the noise model determiner;
And a channel code decoding reliability measuring unit for measuring decoding reliability of a predetermined channel coding unit decoded by the channel code decoder. 10. The apparatus of claim 9, wherein the channel code decoding reliability measuring unit measures a decoding reliability of a channel code based on a soft output of the channel code decoder and / or a cyclic redundancy check (CRC) received from an encoding device. Decoding apparatus characterized in that for measuring. The image restoring apparatus of claim 1, wherein the image restoration unit is used.
A data buffer for storing predetermined channel coding units output from the channel code decoder and outputting intermediate image data including the predetermined channel coding units;
A noise model buffer configured to store a noise model determined by the noise model determiner when decoding the predetermined channel coding unit;
An image restoration noise model determination unit for determining a noise model most suitable for image restoration among the determined noise models stored in the noise model buffer;
And an image restorer for reconstructing an image by using intermediate image data including auxiliary channel coding units stored in a data buffer, auxiliary information, and noise models determined by the image restoration noise model determiner. A decoding method for decoding a distributed video encoded image,
Restoring key picture information transmitted from the encoding apparatus;
Generating auxiliary information using the reconstructed key picture;
Providing a noise model to be used for a predetermined channel coding unit based on a multiple noise model consisting of a plurality of candidate noise models for estimating noise errors in the auxiliary information;
Correcting a noise error of the auxiliary information using the compressed data transmitted from the encoding apparatus based on the provided noise model;
Decoding predetermined channel coding units for a predetermined decoding unit;
And reconstructing a WZ picture by using the decoded intermediate image data value composed of the predetermined channel coding units, the auxiliary information, and the provided noise model. The method of claim 12, wherein providing the noise model comprises:
Generating a multiple noise model composed of a plurality of candidate noise models for estimating noise errors in the auxiliary information;
And determining a noise model to be used for a predetermined channel coding unit among the plurality of candidate noise models. The method of claim 12, wherein generating the multiple noise model
Predicting noise using a difference between key pictures before and after the auxiliary information;
Generating a noise model according to a noise probability distribution based on the predicted noise, and further generating additional noise models according to the noise distribution characteristic of the generated noise model. 15. The method of claim 14, wherein generating the additional noise models further comprises:
Determining whether the generated noise model is a stationary / non-stationary characteristic according to the noise distribution characteristic of the generated noise model;
And generating a noise model for a characteristic of the stationary and non-stationary characteristics that the generated noise model does not have. The method of claim 14, The generating of the multiple noise model may further include generating a plurality of candidate noise models by applying the generated noise models while changing a value of a parameter for compensating an actual noise information estimation error. Decoding device. 13. The decoding method of claim 12, wherein the noise probability distribution comprises at least one of Laplacian, Gaussian, generalized gaussian, and Two-sided Gamma distributions. The method of claim 12, wherein the determining of the noise model includes selecting a noise model in order of convergence speed among candidate noise models based on an estimated error rate after channel decoding is performed on the parity data received from the encoding apparatus. Decoding method comprising a. The method of claim 12, wherein the determining of the noise model comprises selecting a noise model that has passed a CRC test using minimum parity data after being independently decoded using each of the plurality of candidate noise models. Decoding method comprising a. The method of claim 12, wherein the decoding step
And measuring decoding reliability of the channel code based on a soft output according to the channel code decoding and / or a cyclic redundancy check (CRC) received from an encoding device.

Images