JP2018032900A - Video encoder and video decoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve coding efficiency of video encoding.SOLUTION: A video encoder includes an orthogonal transformation unit 14 and an entropy coding unit 17, a waveform approximation unit 30 finds parameters by performing parametric waveform fitting for an orthogonal transformation factor string outputted from the orthogonal transformation unit 14, and the entropy coding section 17 encodes the encoding object when generating a bit string while adding the parameter or a quantized number value thereof. A video decoder decodes the parameters or quantized number value thereof from the bit string in an entropy decoding unit 50, and obtains decode value of orthogonal transformation factor string to be fed to an inverse orthogonal transformation unit 53, by reproducing the approximate waveform of the orthogonal transformation factor string in a factor restoration unit 51.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a video encoding device and decoding device.

  In recent years, ultra-high resolution images such as 4K (spatial resolution / frame frequency: 3840x2160 / 60p, etc.) and 8K (spatial resolution / frame frequency: 7680x4320 / 60p, 7680x4320 / 120p, etc.) have been widely used. The amount of data of ultra-high resolution video is enormous, and it is necessary to compress the amount of video data for transmission over broadcast waves and IP networks.

  As data compression of video and images, the image is divided into blocks, orthogonal transform such as discrete cosine transform (DCT) is performed for each block, the resulting transform coefficient is quantized, and the quantized transform coefficient A hybrid coding method for entropy coding is well known.

  In HEVC (MPEG-H Part 2 High Efficiency Video Coding) /H.265, in addition to discrete cosine transform, discrete sine transform (DST) is used to encode the residual signal of intra prediction of small luminance blocks. ) Is also used (Patent Document 1).

  Also, the resolution was obtained by converting the resolution of the original image signal and coding the structural component, which is a low-resolution image, and subtracting the signal obtained by encoding / decoding the structural component from the original image signal and performing the reverse resolution conversion. A technique for encoding histogram information of a texture component has been proposed (Patent Document 2).

  Conventionally, because of the statistical properties of natural image signals, discrete cosine transform is used for image coding and video coding based on the fact that the Kalunen-Loeve transform basis, which is the principal component vector, can be approximated by a discrete cosine transform basis. It has been used a lot. In particular, in moving picture encoding, prediction processing is performed using correlation between frames or within frames, and conversion processing is generally performed on the prediction residual.

  First, an example of a conventional video encoding device that performs data compression of a video signal using orthogonal transform and prediction processing will be described.

  FIG. 9 is a block diagram showing an example of a configuration of a conventional video encoding apparatus having orthogonal transform such as discrete cosine transform. The video encoding device includes a block division unit 10, a memory 11, a prediction unit 12, a subtraction unit 13, an orthogonal transformation unit 14, a scanning unit 15, a quantization unit 16, an entropy encoding unit 17, The inverse quantization unit 18, the scanning unit 19, the inverse orthogonal transform unit 20, and the addition unit 21 are configured.

  The block division unit 10 divides the frame of the input video into partial areas that are units for performing subsequent processing. For example, the block division may be changed according to the prediction unit 12 or the orthogonal transform unit 14 or may be the same. The block division shape is typically rectangular (square or rectangular), but is not limited thereto. The block dividing unit 10 uses processing parameters (for example, the prediction unit 12) of other processing units (prediction unit 12, orthogonal transform unit 14, scanning unit 15, quantization unit 18, etc.) by an optimization processing unit not shown in FIG. 9. A rate distortion function (bit rate and code) together with reference frames used for prediction, basis functions used in the orthogonal transform unit 14, coefficient scanning order in the scanning unit 15, quantization width in the quantization unit 18 and presence / absence of coefficient signaling, etc. The division shape and the division size may be adaptively controlled so that the evaluation value defined based on the conversion error takes an optimum value.

  At each stage of the operation of the video encoding device, the memory 11 stores a partial area of the video that has been encoded up to that stage (a frame that has already been encoded or has already been encoded. The local decoded image in the encoded partial area) is held. This local decoded image indicates a result of performing a decoding operation (processing of the inverse quantization unit 18, the scanning unit 19, the inverse orthogonal transform unit 20, and the addition unit 21) in the video encoding device.

  The prediction unit 12 tries to approximate (referred to as prediction) a pixel value sequence in a block (hereinafter referred to as a target block) output from the block division unit 10 from the locally decoded image stored in the memory 11, and uses the approximation result as a prediction block. Output. The prediction includes, for example, motion compensation prediction and intra-screen prediction.

  In motion compensated prediction, a partial area approximating the pixel value sequence in the target block is searched from a frame at another time (a time different from the frame in which the block dividing unit 10 performs block division) held in the memory 11. The partial region having the highest degree of approximation (for example, the smallest sum of squared errors) is obtained. A relative position indicating where the partial area with the highest degree of approximation is located with respect to the target block position is obtained as a motion vector and passed to the entropy encoding unit 17, and a pixel value sequence of the partial area with the highest degree of approximation is obtained. It passes to the subtraction means 13 as a prediction block.

  For example, the intra-screen prediction is performed based on the pixel value sequence in the encoded partial area of the current frame (the frame at the same time as the frame on which the block dividing unit 10 performs block division) held in the memory 11. Try to approximate the pixel value sequence. For example, based on a pixel value sequence of encoded pixels (reference pixels) existing around the target block, an average value thereof is obtained, and a prediction block is generated by setting all pixel values of the prediction block to the average value ( Average prediction). Or the pixel value of a prediction block is produced | generated by the bilinear calculation, for example from the reference pixel row | line which consists of 4 pixels (planar prediction, planar prediction). Further, for example, a prediction block is generated by extrapolating the reference pixel column in a predetermined direction (direction prediction). Further, for example, a partial region that most closely approximates the pixel value sequence of the target block is searched from the pixel value sequence of the encoded partial region of the current frame held in the memory 11, and the partial region having the highest degree of approximation is found. A relative position indicating where the target block position is located is obtained as a motion vector and passed to the entropy coding unit 17 and the pixel value sequence of the partial area having the highest degree of approximation is passed to the subtracting unit 13 (intra block copy). ).

  The subtraction unit 13 subtracts each pixel value of the prediction block output from the prediction unit 12 from each pixel value of the target block output from the block division unit 10 and outputs the result as a residual block.

  The orthogonal transform unit 14 applies orthogonal transform to the residual block and outputs a transform coefficient sequence (transform coefficient value sequence) as a result. As the orthogonal transform, for example, a discrete cosine transform, a discrete sine transform, a Hadamard transform, a wavelet transform, or a transform that approximates them can be used.

  The scanning unit 15 scans the two-dimensional transform coefficient sequence output from the orthogonal transform unit 14 in a predetermined order, rearranges it into a one-dimensional number sequence, and outputs the result. The scanning order of the scanning unit 15 is, for example, zigzag scanning from a direct current component to a low frequency component and a high frequency component (zigzag scanning), or main scanning (for example, ascending order of horizontal frequency) and sub scanning (for example, vertical frequency). Ascending order) raster scanning or Hilbert scanning can be applied.

  The quantization unit 16 quantizes the conversion coefficient sequence from the scanning unit 15 and converts it into a quantization index sequence. In the quantization, for example, each transform coefficient value is divided by a predetermined numerical value (quantization width), and the quotient is used as a quantization index. The quantization width may be constant regardless of the number of the transform coefficient sequence (which frequency is), or the quantization width may be different depending on the frequency. When the quantization width differs according to the frequency, for example, the quantization width for each frequency is defined as a table (quantization table).

  Further, the quantization unit 16 may terminate the transform coefficient sequence halfway. For example, the quantization unit 16 determines that the absolute value of the transform coefficient after the nth transform coefficient sequence (n is an integer value equal to or greater than 1 and equal to or less than the total number of transform coefficient sequences) is a predetermined threshold (for example, 0, 1 or 2) ) In the case of the following, it may be assumed that all the nth and subsequent coefficients are 0, and the subsequent numerical value transfer to the entropy encoding 17 may be terminated.

  The entropy encoding unit 17 applies entropy encoding to the quantization index sequence output from the quantization unit 16 and the motion vector output from the prediction unit 12, and outputs the resulting bit sequence. In addition, the entropy encoding unit 17 also includes an identifier (mode information) indicating what prediction the prediction unit 12 has performed, the type of basis function applied by the orthogonal transformation unit 14, the block division shape in the block division unit 10, and the quantization unit 16 Information regarding the quantization width in the above may be included in the encoding target.

  The inverse quantization unit 18 converts the quantization index sequence into a transform coefficient sequence. The inverse quantization unit 18 is an operation for obtaining a representative value in a range that the input value can take when the quantization unit 16 outputs a quantization index. This representative value can be, for example, an arithmetic average value of an upper limit value and a lower limit value of a range that the input value can take.

  The scanning unit 19 converts the one-dimensional conversion coefficient sequence output from the inverse quantization unit 18 into a two-dimensional conversion coefficient sequence by the reverse operation of the scanning unit 16.

  The inverse orthogonal transform unit 20 performs the inverse transform of the orthogonal transform unit 14 and converts the two-dimensional transform coefficient sequence from the scanning unit 19 into a two-dimensional residual pixel value sequence (decoded residual block). For example, when the orthogonal transform unit 14 performs discrete cosine transform, the inverse orthogonal transform unit 20 performs inverse discrete cosine transform. For example, when the orthogonal transform unit 14 performs discrete sine transform, the inverse orthogonal transform unit 20 performs inverse discrete sine transform. Note that the two-dimensional residual pixel value sequence obtained by the inverse orthogonal transform unit 20 corresponds to the residual block of the output of the subtraction unit 13, but includes information errors due to intermediate orthogonal transform, quantization, and the like. For this reason, it is called “decoded residual block”.

  The addition unit 21 adds the pixel value sequence of the decoded residual block from the inverse orthogonal transform unit and the pixel value sequence of the prediction block from the prediction unit 21 for each pixel, and outputs the resulting pixel value sequence as a decoded block To do. The obtained decoded block is recorded as a pixel value sequence at the block position in the memory 11.

  The above is an example of the configuration and operation of a conventional video encoding device.

Japanese Patent No. 5302256 Japanese Patent No. 5700666

  In a video encoding apparatus using a prediction process, a prediction residual signal has a statistical property different from that of a natural image, and thus a discrete cosine transform base is not necessarily an optimal base. Therefore, in intra prediction by direction prediction, focusing on the fact that the residual value in the target block close to the reference pixel is close to zero, it is proposed to use the discrete sine transform that is the basis of the odd function from the boundary condition, Used in HEVC / H.265, it is effective. As described above, it has been conventionally performed that an orthogonal transformation such as a discrete cosine transformation or a discrete sine transformation is used to encode a video with a transformation coefficient sequence as a transformation result and transmit the video at a high compression coding rate.

  However, in the encoding process that uses orthogonal transform and encodes a video with a transform coefficient sequence, the transform coefficient sequence that is the transform result is a low frequency band, regardless of whether discrete cosine transform or discrete sine transform is used. The component, particularly the direct current component has a large power, and the power tends to attenuate as the frequency becomes higher. The appearance of such a statistical property means that the signal correlation has not been fully utilized in discrete cosine transform or discrete sine transform, and further information compression is possible.

  Accordingly, an object of the present invention made in view of the above problems is to use a statistical tendency still remaining in an orthogonal transform coefficient sequence in a video encoding method having an orthogonal transform such as a discrete cosine transform. Accordingly, an object of the present invention is to provide an encoding device and a decoding device that can improve the encoding efficiency.

  In order to solve the above problems, a video encoding device according to the present invention is a video encoding device having an orthogonal transform unit and an entropy coding unit, and is a parametric method for the orthogonal transform coefficient sequence output from the orthogonal transform unit. A waveform approximating unit for performing appropriate waveform fitting, wherein the waveform approximating unit outputs a parameter as a result of the waveform fitting, and the entropy encoding unit includes the parameter in an encoding target when generating a bit string It is characterized by.

  Further, in the video encoding device, the waveform approximation unit further outputs residual information based on a residual sequence obtained by subtracting an approximate waveform approximated using the parameter from the orthogonal transform coefficient sequence, and the entropy It is desirable that the encoding unit further includes the residual information in an encoding target when generating a bit string.

  In order to solve the above problem, a video decoding apparatus according to the present invention is a video decoding apparatus that decodes a bit string output from the video encoding apparatus, and corresponds to an entropy encoding unit of the video encoding apparatus. An entropy decoding unit, wherein the entropy decoding unit decodes the parameter from the bit sequence, and obtains a decoded value of the orthogonal transform coefficient sequence by reproducing an approximate waveform of the orthogonal transform coefficient sequence based on the parameter. To do.

  Further, in the video decoding device, the entropy decoding unit further decodes or creates the residual information from the bit sequence, reproduces an approximate waveform of an orthogonal transform coefficient sequence based on the parameters, and adds the residual information to the residual information. It is desirable to obtain the decoded value of the orthogonal transform coefficient sequence by adding the residual sequences obtained from the above.

  According to the present invention, in a video encoding system having orthogonal transform such as discrete cosine transform, encoding efficiency can be improved by utilizing a statistical tendency still remaining in an orthogonal transform coefficient sequence. When transmitting orthogonal transform coefficient sequences with parametric approximation parameters and residual information to compensate for errors that occur during approximation, the parameters can be set to be fewer than when all transform coefficient sequences are transmitted. In addition, since the residual information (residual value sequence) can be concentrated in the vicinity of 0, it is possible to enhance the compression effect by entropy coding.

It is the block diagram which showed an example of the structure of the video coding apparatus which concerns on this invention. It is the block diagram which showed an example of the structure of the waveform approximation part. It is the figure which illustrated the process of approximating a conversion coefficient sequence by a parametric function. It is the block diagram which showed an example of the structure of a waveform fitting part. It is the block diagram which showed an example of the structure of the video decoding apparatus which concerns on this invention. It is the block diagram which showed an example of the structure of a coefficient decompression | restoration part. It is the block diagram which showed an example of the structure of the video coding apparatus (what does not have a prediction process part) which concerns on this invention. It is the block diagram which showed an example of the structure of the video decoding apparatus (what does not have a prediction process part) which concerns on this invention. It is the block diagram which showed an example of the structure of the conventional video coding apparatus.

  Embodiments of the present invention will be described below with reference to the drawings.

(Embodiment 1)
FIG. 1 is a block diagram showing an example of the configuration of a video encoding apparatus according to Embodiment 1 of the present invention. The video encoding apparatus according to Embodiment 1 uses a prediction process. Hereinafter, the configuration and operation of a video (moving image) encoding apparatus will be described with reference to the conventional video encoding apparatus shown in FIG. The explanation will be made while paying attention to the difference.

  In this configuration, the video encoding apparatus 100 includes a block division unit 10, a memory 11, a prediction unit 12, a subtraction unit 13, an orthogonal transformation unit 14, a scanning unit 15, a waveform approximation unit 30, and entropy encoding. The unit 17, the scanning unit 19, the inverse orthogonal transform unit 20, and the addition unit 21 are configured.

  About the block division unit 10, the memory 11, the prediction unit 12, the subtraction unit 13, the orthogonal transformation unit 14, the scanning unit 15, the scanning unit 19, the inverse orthogonal transformation unit 20, and the addition unit 21 Since this is the same as the operation of the conventional video encoding apparatus in FIG.

  The waveform approximating unit 30 receives parameters of a one-dimensional transform coefficient sequence from the scanning unit 15 and approximates the transform coefficient sequence with a parametric function, and waveform components (residuals that cannot be approximated by the approximation). Difference information) is output to the entropy encoding unit 17, and a transform coefficient sequence approximated by the approximation and residual information is output to the scanning unit 19 as a decoded coefficient vector (decoded transform coefficient sequence).

FIG. 2 is a block diagram illustrating an example of the configuration of the waveform approximating unit 30. In FIG. 2, the waveform approximation unit 30 includes a waveform fitting unit 31 (31 ₁ , 31 ₂ , 31 ₃ ), a quantization unit 32, an inverse quantization unit 33, and an addition unit 34. As will be described later, the quantization unit 32, the inverse quantization unit 33, and the addition unit 34 are not essential components. The waveform fitting unit 31 may be configured by only one, or may be configured by multistage connection. In the example of FIG. 2, the waveform fitting units 31 _{1 to} 31 ₃ are configured by connecting in three stages. That performs a first waveform approximated by the waveform fitting portion 31 _1, waveform approximates a component which can not be approximated by the waveform approximated by a waveform fitting portion 31 _2. Further approximates the component which has not been waveform approximating the waveform fitting 31 ₁ and 31 ₂ by the waveform fitting portion 31 _3.

A component (residual component) that could not be approximated up to the waveform fitting unit 31 _m (m∈ {1, 2, 3}) is set as s ^(m) (where s ^(m) is a vector or a numerical sequence. is there). Incidentally, i is set to 1 or more transform coefficients sequence of the total number less ^integral, placing the i-th component value of ^{s (m)} and ^{s (m) (i).} However, s ⁽⁰⁾ means a conversion coefficient sequence from the scanning unit 15, and s ⁽⁰⁾ (i) is the i-th component.

The waveform fitting unit 31 _m approximates the waveform of the input vector s ^(m−1) with a parametric function (basic vector b ^(m) ), the parameter c ^(m) for the approximation, and the residual component s. ^(M) is output. Furthermore, the waveform fitting unit 31 _m includes a transform coefficient sequence r ^(m) approximated by parameters c ^{(1) to} c ^(m) up to the waveform fitting unit 31 _m (its i-th component is represented by r ^(m) (i) Output).

Processing performed by the waveform approximating unit 30 (particularly, the waveform fitting unit 31 ₁ ) in FIG. 2 will be described with reference to FIG. FIG. 3 is a diagram illustrating a process of approximating the transform coefficient sequence by a parametric function.

In waveform approximation processing, base vectors b ^{(1) to} b ^(m) used in each waveform fitting unit 31 (31 _{1 to} 31 _m ) are prepared in advance. The basis vector b can be set as a numerical sequence based on a predetermined function. For example, b ⁽¹⁾ is based on an exponential function, b ⁽²⁾ is based on a low-frequency oscillation function, and b ⁽³⁾ The basis vector b can be determined based on the parametric nature of the transform coefficient sequence input from the scanning unit 15 such that is based on a high-frequency oscillation function. As the basis vector, a numerical sequence based on a representative function may be set as a basis vector in advance, or the characteristics of similar images are experimentally examined, and, for example, a principal component extraction analysis is performed to perform a Karhunen-Labe transformation basis. For example, an optimal basis vector may be created.

In FIG. 3, s ⁽⁰⁾ is an example of a conversion coefficient sequence from the scanning unit 15. In contrast, the waveform fitting unit 31 _1, for example, an exponential function exp as basis vectors ^{b (1) (-Ax):} [A constant] Prepare the numerical sequence based on, on the basis vectors ^{b (1)} The most appropriate fitting coefficient (scalar coefficient) to be multiplied and fitted to the transform coefficient sequence of s ⁽⁰⁾ is obtained, and the coefficient is quantized to obtain c ⁽¹⁾ (scalar in this case), which is used as a parameter. And

Note that a numerical sequence based on the exponential function αexp (−βx) can be used as the basis vector b ⁽¹⁾ , and the fitting of the exponential function can be performed with two coefficients α and β. When such a function including a plurality of coefficients is used, the parameter c ⁽¹⁾ is handled as a parameter or a vector including a plurality of scalars based on α and β.

The waveform fitting portion 31 _1, basis vectors b ⁽¹⁾ and outputs a transform coefficient sequence r approximated by multiplying the fitting coefficients ^(1), the transform coefficient sequence s ⁽⁰⁾ transform coefficients to approximate the sequence r ^{(1 )} Is output as a residual component s ⁽¹⁾ . This approximated transform coefficients sequence r ^(1), the residual component s ⁽¹⁾ is a next input signal waveform fitting portion 31 _2.

In the next waveform fitting portion 31 _2, and a basis vector ^{b (2)} for example the sine function _{sin (ω 1 x): [} ω 1 is constant] Prepare the numerical sequence based on, multiplies this basis vector ^{b (2)} Thus, the most appropriate fitting coefficient for fitting to the residual component s ⁽¹⁾ is obtained, and the coefficient is quantized to obtain a parameter c ⁽²⁾ . In the waveform fitting portion 31 _2, basis vector ^b so far ^{(1), b (2)} a parameter ^{c (1),} and ^{c (2)} transform coefficient sequence ^r approximated by ^(2), further remains residual component ^{s (2)} and outputs (not shown), the next waveform fitting portion 31 ₃ of the input data. In general, each waveform fitting unit 31 outputs the transform coefficient sequence r reconstructed by the previous fitting and the remaining residual component s.

FIG. 4 is a block diagram illustrating an example of the configuration of the waveform fitting unit 31 (31 _m ). In FIG. 4, the waveform fitting unit 31 includes an inner product calculation unit 40, a quantization unit 41, an inverse quantization unit 42, a scalar multiplication unit 43, a subtraction unit 44, and an addition unit 45.

The inner product calculation unit 40 calculates an inner product of a vector in which all input residual components s ^(m−1) are arranged and a base vector b ^(m), and outputs an inner product value p ^(m) as a result. . Preferably, the basis vector b ^(m) is a unit vector. The inner product value p ^(m) is calculated by the following equation (1). N is the total number of transform coefficient sequences.

Coefficient value sequence (s ^(m−1) (i)) _{i = 1, 2,..., N} is approximated by a sequence (t ^(m) · b ^(m) _(i) ) _{i = 1, 2} ,. The optimum fitting coefficient (scalar coefficient in this case) t ^{(m) in this case} is obtained by the following equation (2).

The waveform fitting unit 31 _{m in} FIG. 4 operates to output a numerical value c ^(m) obtained by quantizing the fitting coefficient t ^{(m )} as a parameter. As will be described later, the waveform fitting unit 31 The fitting coefficient t ^(m) may be output as a parameter.

The quantization unit 41 quantizes the inner product value p ^(m) by a predetermined quantization width q ^(m) (q ^(m) is a positive real number), and outputs the result c ^(m) . For example, when the basis vector b ^(m) is a unit vector, the following equation (3) is calculated.

は、ｚより大きくない最大の整数を表す（床関数）。 In addition,

Represents the largest integer not greater than z (floor function).

Alternatively, when the basis vector b ^(m) is not necessarily a unit vector, the quantization unit 41 more generally executes the following equation (5 ⁾ to obtain c ^(m) .

The inner product calculation unit 40 and the quantization unit 41 may not be separately provided, and the calculation of the following equation (6) may be performed by a processing unit in which these are integrated to directly obtain c ^(m) .

Any of the mathematical formulas (3), (5), and (6) corresponds to a mathematical formula that calculates a numerical value c ^(m) obtained by quantizing the fitting coefficient of the mathematical formula (2) by the quantization width q ^(m) .

The inverse quantization unit 42 multiplies the quantized numerical value c ^(m) by the quantization width q ^(m) and outputs the result as an approximate value u ^(m) of the fitting coefficient (here, scalar coefficient) t ^(m). To do. That is, the approximate value u ^(m) is obtained by the following equation (7).

The scalar multiplication unit 43 multiplies each component of the basis vector b ^{(m) by} the scalar u ^(m), and calculates the approximate waveform σ ^(m) (i) (i∈ {1, 2,... N}) (the arrangement of this as a column vector with respect to i is simply expressed as a vector σ ^(m)) .

Note that the inverse quantization unit 42 and the scalar multiplication unit 43 are not provided separately, and a processing unit in which these units are integrated performs the calculation of the following equation (9) to directly calculate the approximate waveform σ ^(m) (i). You can ask for it.

The subtraction unit 44 subtracts each sample value σ ^(m) (i) of the approximate waveform from each input residual component s ^(m−1) (i) based on the following equation (10), and obtains the result. Each residual component s ^(m ) is output as (i).

The adding unit 45 adds each sample value σ ^(m) (i) of the approximate waveform to each input transform coefficient sequence r ^(m−1) (i) based on the following equation (11), and the result is Each conversion coefficient sequence r ^(m ) is output as (i).

  Returning to FIG. 2, the operation of the waveform approximation unit 30 will be described.

The waveform fitting unit 31 _m (m is an integer greater than or equal to 1 and less than or equal to M, M = 3 in the example of FIG. 3) is an input sequence (s ^(m−1) (i)) _{i∈ {1, 2,. N}} , the sequence (r ^(m−1) (i)) _{iε {1, 2,..., N}} and the basis vector b ^(m) are processed as described above, and the parameter c ^(m) and the residual component are processed. (S ^(m) (i)) _{i∈ {1, 2,..., N}} , and approximate waveform conversion coefficient series (r ^(m) (i)) _{i∈ {1, 2,.} Is output. In the following description and drawings, a sequence (s ^(m) (i)) _{iε {1, 2,..., N}} or a sequence (r ^(m) (i)) _{iε {1, 2,. , N}} may be described as a column vector s ^(m) or a column vector r ^(m) arranged in a column with respect to i.

The first stage of the input of the waveform fitting portion 31 _1, all i∈ {1,2, ..., N} ^to, and ^{r (0) (i) =} 0. For iε {1, 2,..., N}, the value of the i-th term of the one-dimensional transform coefficient sequence output from the scanning unit 15 is set in s ⁽⁰⁾ (i).

Heretofore, it has been described that each waveform fitting unit 31 outputs a numerical value c ^(m) obtained by quantizing the fitting coefficient t ^(m) as a parameter. However, the fitting coefficient t ^(m) obtained by Expression (2 ⁾ is used. Alternatively, it may be output as a parameter for approximation. In the present invention, the parameter resulting from the waveform fitting means the fitting coefficient t ^(m) or a numerical value c ^(m) obtained by quantizing the fitting coefficient t ^(m) . As described above, each parameter may be a vector (including a plurality of scalars).

Quantization unit 32, the sequence of the residual component output from the waveform fitting portion ^{_{31 3 (s (3) (}} i)) i∈ {1,2, ..., N} quantizes and outputs the quantization index column To do. For example, the quantization unit 32 quantizes the residual component s ⁽³⁾ (i) (i∈ {1, 2,..., N}) with the quantization width Q (i), and the result is a quantization index d. Output as (i). The quantization index d (i) is obtained by the following equation (12).

Note that the sequence (Q (i)) _{iε {1, 2,..., N}} of the quantization width Q (i) defined for each frequency can be defined as a quantization table, for example.

Here, it has been described that the waveform approximating unit 30 outputs a quantization index d (i) obtained by quantizing the sequence of residual components (s ⁽³⁾ (i)) _{iε {1, 2,..., N}} . However, in order to compensate for an error occurring in the approximation, the sequence of residual components (s ^(m) (i)) _{iε {1, 2,..., N}} obtained at the final stage of the waveform fitting unit 31 is used as it is. May be output as residual information. A part of the sequence may be used as residual information. In the present invention, the residual information is a sequence of residual components (s ^(m) (i)) _{iε {1, 2,..., N}} obtained at the final stage of the waveform fitting unit 31 or the residual components. Quantization index (d (i)) obtained by quantizing the number sequence (s ^(m) (i)) is information including at least a part of _{i∈ {1, 2,..., N}} .

  For example, the waveform approximating unit 30 may stop outputting the residual information halfway. When the absolute value of the quantization index after the nth (n is an integer value greater than or equal to 1 and less than or equal to the total number of index sequences) after the quantization index sequence becomes equal to or less than a predetermined threshold, the waveform approximation unit 30 Assuming that all of the coefficients are 0, the numerical value transfer to the entropy encoding unit 17 thereafter may be terminated. Similarly, with respect to the parameters of the waveform fitting result, when the absolute value of the entire sequence (vector) of residual components is equal to or less than a predetermined threshold, all the parameters of the waveform fitting result thereafter are 0. As a result, the numerical value transfer to the entropy encoding unit 17 may be terminated.

  The inverse quantization unit 33 multiplies each quantization index d (i) by each quantization width Q (i) in the quantization table, and outputs the result as e (i).

The adding unit 34 converts the approximate waveform conversion coefficient sequence (r ⁽³⁾ (i)) _{iε {1, 2,...} Output from the final stage of the waveform fitting unit 31 (the waveform fitting unit 31 _{3 in} the configuration of FIG. 2) _{. , N}} is added with (e (i)) _{iε {1, 2,..., N}} output from the inverse quantization unit 33, and the result is decoded transform coefficient sequence (f (i)) _iε. Output as _{{1, 2,..., N}} . That is, f (i) is expressed by the following equation (14).

The waveform approximating unit 30 does not include the quantizing unit 32, the inverse quantizing unit 33, and the adding unit 34, and the approximate waveform output from the final stage of the waveform fitting unit (the waveform fitting unit 31 _{3 in} the example of FIG. ₃ ). The transform coefficient sequence (r ⁽³⁾ ) _{i∈ {1, 2,..., N}} is output as a decoded transform coefficient string (f (i)) _{i∈ {1, 2,.} It doesn't matter. In this case, f (i) is expressed by the following equation (15).

In FIG. 1, the entropy encoding unit 17 includes parameter strings c ⁽¹⁾ , c ⁽²⁾ ,..., C ^(M) output from the waveform approximating unit 30 (M is a waveform fitting unit in the waveform approximating unit 30 ⁾ . 3 is an integer representing the number of stages, M = 3 in the example of FIG. 3, and residual information (quantized index sequence) d (1), d (2),..., D (N) (N is the orthogonal transform unit 14) An integer representing the total number of transform coefficient sequences to be output), entropy coding is applied to the motion vector output from the prediction unit 12, and the resulting bit sequence is output. In addition, the entropy encoding unit 17 also includes an identifier (mode information) indicating what prediction the prediction unit 12 has performed, the type of basis function applied by the orthogonal transformation unit 14, the block division shape in the block division unit 10, and the quantization unit 16 Information regarding the quantization width in the above may be included in the encoding target.

The entropy encoding unit 17 may not encode part or all of the residual information (quantization index sequence) d (1), d (2),..., D (N). Further, the entropy encoding unit 17 may not encode a part of the parameter string c ⁽¹⁾ , c ⁽²⁾ ,..., C ^(M) , or may not encode all of them. I do not care.

For example, the quantization index string d (1), d (2),..., D (N) may not be encoded and output when it converges to a small numerical value. Further, when the residual component s ^{(3) of} the waveform fitting unit 31 ₃ is sufficiently small and the parameters below c ^{(4 )} are sufficiently small, c ⁽¹⁾ , c ⁽²⁾ , c ⁽ Only ³⁾ may be encoded and transmitted.

  As described above, it is possible to delete the less important data from the output bit string without encoding, as processing of the waveform approximation unit 30 or the entropy encoding unit. By omitting such encoding, further data compression can be performed.

  Even if all the output information from the waveform approximation unit 30 is entropy-encoded, the higher-order waveform fitting parameters and the residual information converge to almost 0, so the transform coefficient sequence is directly entropy-encoded. In many cases, the total amount of data decreases.

  Next, the video decoding device will be described.

  FIG. 5 is a block diagram showing an example of the configuration of the video decoding apparatus according to the present invention. 5, the video decoding apparatus 200 includes an entropy encoding unit 50, a coefficient restoration unit 51, a scanning unit 52, an inverse orthogonal transform unit 53, a memory 54, a prediction unit 55, and an addition unit 56. Is done.

The entropy decoding unit 50 decodes the bit sequence encoded by the entropy encoding unit 17 of the video encoding device 100, and parameters (for example, a numerical sequence obtained by quantizing the fitting coefficient) c ⁽¹⁾ , c ⁽²⁾ , .., C ^(M) and residual information (for example, quantization index sequence) d (1), d (2),..., D (N) are obtained.

The coefficient restoration unit 51 decodes the transformed transform coefficient sequence from the parameter sequence c ⁽¹⁾ , c ⁽²⁾ ,..., C ^(M) and the quantization index sequence d (1), d (2),. f (1), f (2),..., f (N) are obtained.

FIG. 6 is a block diagram illustrating an example of the configuration of the coefficient restoration unit 51. The coefficient decoding unit 51 includes an inverse quantization unit 59 (59 _{1 to} 59 _M ), a scalar multiplication unit 60 (60 _{1 to} 60 _M ), an addition unit 61 (61 _{1 to} 61 _M ), and an inverse quantization unit 62. It is comprised by. In FIG. 6, M = 3. Further, as described below, the addition unit 61 _M and the inverse quantization unit 62 is not essential.

Similarly to the inverse quantization unit 42 of the waveform fitting unit 31, the inverse quantization unit 59 _m (m is an integer of 1 or more and M or less) uses each parameter c ^(m) (m∈ {1, 2,..., M}). ) Is multiplied by each quantization width q ^(m ) in the quantization table, and the result is output as a coefficient u ^(m) .

The scalar multiplication unit 60 _m (m is an integer of 1 to M) multiplies the basis vector b ^(m) by a coefficient u ^(m) and outputs the result as a vector σ ^(m) .

Addition unit 61 ₁ includes a vector sigma ⁽¹⁾ to the output of the scalar multiplication unit ₆₀₁ performs vector addition of the vector sigma ⁽²⁾ to the output of the scalar multiplication portion 60 _2, and outputs the result.

The addition unit 61 _m performs vector addition of the vector output from the addition unit 61 _(m−1) and the vector σ ^{(m + 1)} output from the scalar multiplication unit 60 _{(m + 1)} when 2 ≦ m ≦ M−1. , Output the result.

  Similar to the inverse quantization unit 33 of the waveform approximation unit 30, the inverse quantization unit 62 assigns each quantization index d (i) (i∈ {1, 2,..., N}) to each quantum in the quantization table. Multiply by the quantization width Q (i) and output the result as e (i).

The adder 61 _M performs vector addition of the output value of the adder 61 _(M−1) and the output value e (i) of the inverse quantization unit 62, and the result is decoded into a transform transform coefficient sequence f (1), Output as f (2),..., f (N).

Incidentally, in the coefficient restoring unit 51, inverse quantization unit 62 and without providing the adding unit 61 _M, the addition unit _{61 (M-1)} output with a decoded transform coefficients sequence f of (1), f (2) , ..., f (N) may be used.

  Returning to FIG. 5, the description of the configuration and operation of the video decoding apparatus 200 will be continued.

  The scanning unit 52 performs one-dimensional decoded transform coefficient sequences f (1), f (2),..., F (N) output from the coefficient restoration unit 51 by the reverse operation of the scanning unit 16 of the video encoding device 100. Is converted into a two-dimensional decoded transform coefficient sequence.

  The inverse orthogonal transform unit 53 performs the inverse transform of the orthogonal transform unit 14 of the video encoding device 100, and converts the two-dimensional decoded transform coefficient sequence from the scanning unit 52 into a two-dimensional residual pixel value sequence (decoded residual block). ). For example, when the orthogonal transform unit 14 performs discrete cosine transform, the inverse orthogonal transform unit 53 performs inverse discrete cosine transform. For example, when the orthogonal transform unit 14 performs discrete sine transform, the inverse orthogonal transform unit 53 performs inverse discrete sine transform. The obtained decoded residual block corresponds to the residual block of the output of the subtracting unit 13 of the video encoding device 100 and is the same as the output of the inverse orthogonal transform unit 20.

  The memory 54 constructs a decoded image by sequentially writing partial areas of video sequentially decoded in units of blocks into a predetermined storage area corresponding to the block position. The memory 54 stores frames of video that have been decoded so far in each stage of the decoding process.

  The prediction unit 55 tries to approximate (predict) the pixel value sequence in the block (target block) output from the block dividing unit 10 from the decoded image stored in the memory 54, and outputs the approximation result as a prediction block. The prediction operation of the prediction unit 55 is the same as the operation of the prediction unit 12 of the video encoding device 100.

  The adding unit 56 generates a decoded block by obtaining, for each pixel, the sum of each pixel value of the prediction block by the prediction unit 55 and each pixel value of the decoded residual block from the inverse orthogonal transform unit 53. The pixel value sequence of the decoded block generated by the adding unit 56 is recorded in a storage area corresponding to the block position in the memory 54.

  The above operation is repeated until a predetermined number of decoded image frames are configured in the memory 54. Here, the predetermined number is the number necessary to display the image frames in the correct display order, and is, for example, one period of a video coding GOP (Group Of Pictures) structure. An output video is obtained from the memory 54 by reading the decoded image frames in the display order.

Various combinations of the bit string transmitted from the video encoding device 100 side and the configuration of the video decoding device 200 can be considered. However, processing may be performed by appropriately deleting or supplementing excess / deficiency information. it can. For example, an encoded bit sequence transmitted from the video encoding device 100 side is a parameter sequence c ⁽¹⁾ , c ⁽²⁾ ,..., C ^(M) and a quantization index sequence d (1), d (2), ... includes a d (N), if no and the coefficient decoding unit 51 inverse quantization unit 62 of video decoder 200 to the adding unit 61 _M, the video decoder 200, the entropy decoded data, parameter sequence ^{c (1), c (2} ), ..., c by extracting only ^(M), this parameter sequence ^{c (1), c (2} ), ..., c decoded transform coefficients sequence f from ^{(M) (1)} , F (2),..., F (N) can be obtained and processed.

On the other hand, the coefficient decoding unit 51 of the video decoding device 200 includes an inverse quantization unit 59 (59 _{1 to} 59 _M ), a scalar multiplication unit 60 (60 _{1 to} 60 _M ), and an addition unit 61 (61 _{1 to} 61 and _M), when provided with all the inverse quantization unit 62, the bit string transmitted from the video encoding apparatus 100 side, parameter sequence ^{c (1), c (2} ), ..., c ^{(M) is} When the residual information is not included, the decoding apparatus side appropriately supplements the residual information (quantization index sequence) d (1), d (2),..., D (N) (for example, these quantizations) Decoding conversion coefficient sequences f (1), f (2),..., F (N) can be obtained by performing processing by creating all index sequences as 0 data.

(Embodiment 2)
The video (moving image) encoding device and decoding device including the prediction process have been described above. However, the waveform approximating unit 30 according to the present invention is applied to the still image encoding device, and the coefficient restoration unit 51 is provided. You may apply to the encoding apparatus for still images. Hereinafter, as a second embodiment of the present invention, a coding apparatus and a decoding apparatus for still images that do not include a prediction process will be described.

  In the case of a still image video encoding device (including a case where a moving image is encoded as a set of still images) that does not have a prediction process such as motion compensation prediction or in-screen prediction, a decoding process in the encoding device, A process for obtaining a prediction block is not necessary, and the encoding apparatus can be simplified.

  FIG. 7 is a block diagram showing an example of the configuration of a video encoding device (without a prediction processing unit) according to the present invention.

  In this configuration, the video encoding device 101 includes a block dividing unit 10, an orthogonal transform unit 14, a scanning unit 15, a waveform approximation unit 30, and an entropy encoding unit 17.

  The block dividing unit 10 is the same as the operation of the conventional video encoding apparatus in FIG.

  The orthogonal transform unit 14 applies orthogonal transform to the pixel values of the image output from the block dividing unit 10 instead of the prediction residual, and outputs a transform coefficient sequence as a result. As the orthogonal transform, for example, discrete cosine transform, discrete sine transform, Hadamard transform, wavelet transform, and transforms obtained by approximating them can be used as in the conventional case.

  The scanning unit 15 scans the two-dimensional transform coefficient sequence output from the orthogonal transform unit 14 in a predetermined order, rearranges it into a one-dimensional number sequence, and outputs the result. This is the same as the operation of the conventional video encoding apparatus.

The waveform approximating unit 30 is basically the same as the configuration of the waveform approximating unit 30 described with reference to FIG. 2, and includes a waveform fitting unit 31 (31 ₁ , 31 ₂ , 31 ₃ ) and a quantizing unit 32, and parameters. (Numerical sequence obtained by quantizing the fitting coefficient) c ⁽¹⁾ , c ⁽²⁾ ,..., C ^(M) and residual information (quantized index sequence) d (1), d (2),. N) is output. Since the video encoding apparatus 101 does not perform the prediction process, the inverse quantization unit 33 and the addition unit 34 are unnecessary, and the decoded transform coefficient sequences f (1), f (2),..., F (N) Is not output.

The entropy encoding unit 17 is similar to the entropy encoding unit 17 of the video encoding device 100 of FIG. 1, and the parameter sequence c ⁽¹⁾ , c ^{(2) ,. ),} And entropy coding is applied to the quantized index string d (1), d (2),..., D (N), and the resulting bit string is output.

  The above is the outline of the configuration and operation of the encoding apparatus 101. Next, a configuration example of a decoding device corresponding to the encoding device in FIG. 7 will be described.

  FIG. 8 is a block diagram showing an example of the configuration of a video decoding apparatus (without a prediction processing unit) according to the present invention.

  In this configuration, the video decoding apparatus 201 includes an entropy encoding unit 50, a coefficient restoration unit 51, a scanning unit 52, an inverse orthogonal transform unit 53, and a memory 54.

An entropy coding unit 50, a coefficient restoring unit 51, the configuration and operation of the scanning unit 52 has the same configuration and operation of each part of the video decoding apparatus 200 of FIG. 5, the coefficient restoring unit 51, parameter sequence c ^{( 1)} , c ⁽²⁾ ,..., C ^(M) and residual information (quantized index sequence) d (1), d (2),..., D (N) to decoded transform coefficient sequence f (1), f (2),..., f (N) are obtained. The scanning unit 52 converts the one-dimensional decoded transform coefficient sequence f (1), f (2),..., F (N) output from the coefficient restoring unit 51 into a two-dimensional decoded transform coefficient sequence.

  Since the video decoding apparatus 201 in FIG. 8 does not perform the prediction process, the inverse orthogonal transform unit 53 outputs the decoded value of the image instead of the decoded value of the prediction residual, and writes the decoded value of the image into the memory 54. A decoded image is constructed.

  As described above, according to the video encoding device 101 and the video decoding device 201 according to the second embodiment, it is possible to increase the encoding efficiency of a still image as compared with the conventional art.

  Although the above embodiment has been described as a representative example, it will be apparent to those skilled in the art that many changes and substitutions can be made within the spirit and scope of the invention. Therefore, the present invention should not be construed as being limited by the above-described embodiments, and various modifications and changes can be made without departing from the scope of the claims. For example, a plurality of constituent blocks described in the embodiments can be combined into one, or one constituent block can be divided.

10 block division unit 11 memory 12 prediction unit 13 subtraction unit 14 orthogonal transformation unit 15 scanning unit 16 quantization unit 17 entropy encoding unit 18 inverse quantization unit 19 scanning unit 20 inverse orthogonal transformation unit 21 addition unit 30 waveform approximation unit 31 ( 31 ₁ , 31 ₂ , 31 ₃ , 31 _m ) Waveform fitting section 32 Quantization section 33 Inverse quantization section 34 Addition section 40 Inner product calculation section 41 Quantization section 42 Inverse quantization section 43 Scalar multiplication section 44 Subtraction section 45 Addition section 50 Entropy Decoding Unit 51 Coefficient Restoration Unit 52 Scanning Unit 53 Inverse Orthogonal Transform Unit 54 Memory 55 Prediction Unit 56 Addition Unit 59 (59 ₁ , 59 ₂ , 59 ₃ , 59 _m ) Inverse Quantization Unit 60 (60 ₁ , 60 ₂ , 60 ₃ , 60 _m ) Scalar multiplication unit 61 (61 ₁ , 61 ₂ , 61 ₃ , 61 _m ) Adder 62 Inverse quantization unit 100, 101 Video encoding device 200, 201 Video decoding device

Claims (4)

A video encoding device having an orthogonal transform unit and an entropy encoding unit,
A waveform approximation unit that performs parametric waveform fitting on the orthogonal transform coefficient sequence output by the orthogonal transform unit,
The waveform approximation unit outputs a parameter as a result of the waveform fitting,
The entropy encoding unit includes the parameter as an encoding target when generating a bit string. The video encoding device according to claim 1,
The waveform approximation unit further outputs residual information based on a residual sequence obtained by subtracting an approximate waveform approximated using the parameter from the orthogonal transform coefficient sequence,
The entropy encoding unit further includes the residual information in an encoding target when generating a bit string. A video decoding device that decodes a bit string output by the video encoding device according to claim 1 or 2,
An entropy decoding unit corresponding to the entropy encoding unit of the video encoding device,
The entropy decoding unit decodes the parameter from the bit string,
A video decoding device, wherein a decoded value of an orthogonal transform coefficient sequence is obtained by reproducing an approximate waveform of the orthogonal transform coefficient sequence based on the parameter. The video decoding device according to claim 3,
The entropy decoding unit further decodes or creates the residual information from the bit string,
A video decoding device characterized in that an approximate waveform of an orthogonal transform coefficient sequence is reproduced based on the parameter, and a decoded value of the orthogonal transform coefficient sequence is obtained by adding a residual sequence obtained from the residual information to the approximate waveform .

Images