JP2009111733A - Method, device and program for encoding image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To prevent generation of a block distortion, and to prevent the deterioration of a decoded image in the encoding system of a block base. <P>SOLUTION: A code quantity computing section 303 computes the quantity of codes on the basis of an encoded frame signal, a reference frame signal and a quantized parameter. A weighted-distortion quantity computing section 305 computes the quantity of a weighted distortion from the lower-limit value of the quantity of a change according to a mode stored in a mode storage section 302. An undefined multiplier computing section 307 computes an undefined multiplier on the basis of the input encoded frame signal, reference frame signal and quantized parameter. A cost computing section 309 computes a cost on the basis of the quantity of codes, the quantity of weighted distortion and the undefined multiplier. A minimum cost decision section 311 decides the cost as the minimum one. An optimum mode output section 316 outputs the mode in this case as the optimum one when the cost is determined as the minimum one. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an image encoding method, an image encoding device, and an image encoding program.

[Encoding method using cost function of square error criterion]
H. In H.264, with the introduction of intra prediction and variable shape motion compensation, the types of prediction modes are increased as compared with the conventional standardized method. For this reason, in order to reduce the amount of codes while maintaining a constant subjective image quality, it is necessary to select an appropriate prediction mode. H. In the H.264 reference software JM, a prediction mode that minimizes the RD cost shown in the following equation (1) is selected (see Non-Patent Document 1).

Here, S is the original signal, q is the number quantization parameter, m is representative of a prediction mode, S _{m, q} is predicted using the mode m against S, in the case of quantized using q It is a decoded signal. Further, λ is a Lagrange's undetermined multiplier used for mode selection. Further, D (S, (^) S _{m, q} ) is a square error sum represented by the following equation (2).

Here, S ^Y , S ^U and S ^V are Y, U and V components of the original signal, and S ^Y _{m, q} , S ^U _{m, q} and S ^V _{m, q} are Y, U and V of the decoded signal. It is an ingredient.

  H. The calculation procedure of the decoded signal in H.264 will be described below. The symbols used for the explanation are summarized in FIG. H. In the H.264 encoding process, orthogonal transformation using the transformation matrix Φ is applied to the prediction error signal R (= S−Pm) when the prediction of the mode number m is used as shown in the following equation (3). The

Φ ^t represents a transposed matrix with respect to the transformation matrix Φ. The transformation matrix Φ is an orthogonal matrix of integer elements expressed by the following equation (4).

  Next, since the matrix Φ is a non-normal matrix, processing corresponding to matrix normalization is performed according to the following equation (5).

  This corresponds to using φ in the following equation (6) instead of Φ in equation (3).

  Further, the quantization using the quantization parameter q is performed on C as shown in the following equation (7). In the reference software JM, normalization is incorporated in quantization.

  On the other hand, H. In the H.264 decoding process, V is inversely quantized as in the following equation (8) to obtain a decoded value of the transform coefficient.

Next, (^) C _q is subjected to inverse transformation as shown in the following equation (9) to obtain a decoded signal of prediction error.

  Finally, a decoded signal of the encoding target image is obtained by the following equation (10).

[Weighting distortion amount considering subjective image quality]
As described above, the subjective image quality measure used in the reference software JM is a square error. However, this square error is not necessarily a distortion amount reflecting subjective image quality degradation. For example, changes in high frequency components are less likely to be detected visually than changes in low frequency components. However, an encoder (for example, reference software JM) that does not use such visual characteristics has room for improvement in terms of efficient code amount reduction.

  Therefore, studies have been made to use the difference in visual sensitivity with respect to spatio-temporal frequency components. A distortion amount corresponding to the subjective image quality is defined by weighting the distortion amount for each spatial frequency component in accordance with the visual sensitivity with respect to the orthogonal transform coefficient. Further, in consideration of the visual sensitivity in the time direction, the weighted distortion amount described above is further weighted according to the shift amount. In this way, the amount of distortion weighted based on the spatiotemporal visual sensitivity is used in the cost function for selecting the encoding parameter.

  Next, weighting based on visual sensitivity for the quantization error signal will be described. As the weighting based on the visual sensitivity for the quantization error signal, for example, there is an RD cost represented by the following equation (11).

Here, C _m is a transform coefficient for the prediction residual signal R _m when the mode number m is used, and C _{m, q} is a coefficient obtained by quantizing and dequantizing C _m with the quantization parameter q. It is a decoded value. As a distortion amount used for calculating the RD cost, a weighted distortion amount represented by the following equation (12) is used.

Here, C ^{γ (i)} _m [k, l] (γ = Y, U, V) is an element of C _m and is a macroblock (16 × 16 [pixel], U, V in the case of Y component). In the case of the component, it is a conversion coefficient included in the i-th sub-block scanned in the raster scan among the sub-blocks (N × N [pixel]) in 8 × 8 [pixel]). Further, (^) C ^{γ (i)} _m [k, l] (γ = Y, U, V) is an element of C _{m, q} , and a macroblock (16 × 16 [pixel] in the case of Y component) ), U, and V components, decoding transform coefficients included in the i-th sub-block scanned in the raster scanning among the sub-blocks (N × N [pixels]) in 8 × 8 [pixels)). .

Further, W ^γ _{k, l} (γ = Y, U, V) is a weighting coefficient set to 1 or less, and is hereinafter referred to as a sensitivity coefficient. The calculation of the sensitivity coefficient will be described in detail in [Calculation of sensitivity coefficient]. In Equation (12), setting W ^γ _{k, l} to a small value corresponds to estimating the quantization distortion D (C _m , C _{m, q} ) to be small.

From the normality of orthogonal transformation, if W ^γ _{k, l} = 1 ( ^∀ k, l; γ = Y, U, V), the above-described weighted distortion amount is equivalent to the square error sum. W ^γ _{k, l} (γ = Y, U, V) takes a smaller value as the spatial frequency and the temporal frequency are higher. From the normality of orthogonal transformation, if W ^γ _{k, l} = 1 ( ^∀ k, l; γ = Y, U, V), the above-described weighted distortion amount is equivalent to the square error sum.
KPLim and G. Sullivan and T. Wiegand, Text Description of Joint Model Reference Encoding Methods and Decoding Concealment Methods.Joint Video Team (JVT) of ISO / IEC MPEG and ITU-T VCEG, JVT-R95, Jan., 2006. http : //ftp3.itu.ch/av-arch/jvt-site/2006-01-Bangkok/JVT-R095.zip

  In the above-described method of weighting the distortion amount based on the cost sensitivity function, weighting by the sensitivity function is performed in units of macroblocks, and thus the discontinuity (block distortion) at the block boundary is not reflected in the weighted distortion amount. In a block-based coding scheme (for example, H.264) that combines inter-frame prediction based on motion compensation and orthogonal transform, block distortion is characteristic coding distortion. For this reason, when the block distortion is not taken into consideration, a case where the obtained weighted distortion amount cannot correctly reflect the subjective image quality occurs.

  The reason why the block distortion is not reflected in the coding distortion scale will be described. In the prior art, the DCT coefficient of each macroblock is weighted based on the contrast sensitivity function. For this reason, the contrast sensitivity with respect to the waveform in each macroblock was reflected, but the discontinuity between adjacent blocks was not considered. Taking the one-dimensional signal shown in FIG. 11 as an example, discontinuity between blocks (block k−1, block k, block k + 1) is closed in a block that performs weighting on the DCT coefficient of each block (block k−1, block k, block k + 1). The discontinuity between block k-1 and block k or the discontinuity between block k and block k + 1) cannot be known.

  That is, in the above-described conventional technology, the image quality is considered only in pixel units. For this reason, only subjective image quality in pixel units can be ensured, block distortion occurs, and the image quality of the decoded image is greatly degraded.

  The present invention has been made in consideration of such circumstances, and the purpose of the present invention is to improve the image degradation of a decoded image due to block distortion in a block-based encoding scheme that combines inter-frame prediction by motion compensation and orthogonal transformation. It is an object to provide an image encoding method, an image encoding device, and an image encoding program that can prevent the above-described problem and reduce the amount of codes.

  In order to solve the above-described problems, the present invention provides an image coding that performs information compression by transform coding and quantization on an image signal or a prediction error signal obtained by intraframe prediction and interframe prediction. In the method, based on a Lagrangian cost function including a distortion amount, a code amount, and an undetermined multiplier, a motion compensation block size in moving image coding, an inter prediction mode, a quantization parameter, a quantization parameter in still image coding, an intra When determining coding parameters such as the prediction mode, the step of measuring the spatial frequency component in the block and the discontinuity between adjacent blocks, and the frequency component in the block and the non-interval between adjacent blocks are measured. Estimating a temporal frequency component related to continuity, the spatial frequency component and the temporal frequency For each component, a step of calculating the importance based on the visual sensitivity function, and a weighted sum of the code amount using a distortion amount obtained as a square error weighted for each frequency based on the importance. And a step of setting a cost function and selecting a mode that minimizes the cost function.

  In the present invention, in the above invention, when measuring spatial frequency components in a block, the number of samples in the horizontal and vertical directions is doubled with respect to the orthogonal transform base image used for encoding. Generating a corrected base image to which a zero value is added, and calculating a discrete Fourier transform coefficient for the corrected base image, thereby generating a spatial frequency component in the block and a discontinuity between adjacent blocks. The method further includes the step of calculating.

  In order to solve the above-described problem, the present invention provides an image in which information compression is performed by transform coding and quantization on an image signal or a prediction error signal obtained by intra-frame prediction and inter-frame prediction. In a coding device, based on a Lagrangian cost function consisting of distortion, coding amount, and undetermined multiplier, motion compensation block size in moving picture coding, inter prediction mode, quantization parameter, quantization parameter in still picture coding , When determining encoding parameters such as intra prediction mode, spatial frequency component measuring means for measuring the frequency component in the block and the spatial frequency component related to the discontinuity between adjacent blocks, and the frequency component in the block, And a time-frequency component estimation means for estimating a time-frequency component related to discontinuity between adjacent blocks; For each component of the spatial frequency component and the temporal frequency component, importance calculation means for calculating the importance based on the visual sensitivity function, and a square error weighted for each frequency based on the importance is obtained. An image encoding apparatus comprising: a cost function setting unit that sets a cost function as a weighted sum with a code amount using a distortion amount; and a mode selection unit that selects a mode that minimizes the cost function. It is.

  According to the present invention, in the above invention, when the spatial frequency measurement means measures the spatial frequency component in the block, the number of samples in the horizontal and vertical directions is respectively set with respect to the base image of the orthogonal transform used for encoding A corrected base image generating means for generating a corrected base image to which a zero value is added so as to be doubled, and by calculating a discrete Fourier transform coefficient for the corrected base image, a spatial frequency component in the block and Spatial frequency component calculation means for calculating a spatial frequency component related to discontinuity between adjacent blocks.

  In order to solve the above-described problem, the present invention provides an image in which information compression is performed by transform coding and quantization on an image signal or a prediction error signal obtained by intra-frame prediction and inter-frame prediction. Based on the Lagrangian cost function consisting of distortion, code amount, and undetermined multiplier, the computer that controls the encoding device, motion compensation block size, inter prediction mode, quantization parameter, still image coding in moving image coding When determining coding parameters such as a quantization parameter and an intra prediction mode, a step of measuring a frequency component in a block and a spatial frequency component related to discontinuity between adjacent blocks, and a frequency component in the block, And estimating a temporal frequency component related to discontinuity between adjacent blocks, and the spatial frequency For each component and component of the time frequency component, a step of calculating importance based on a visual sensitivity function, and a code using a distortion amount obtained as a square error weighted for each frequency based on the importance. An image encoding program, comprising: setting a cost function as a weighted sum with a quantity; and selecting a mode that minimizes the cost function.

  According to the present invention, the frequency component in the block and the spatial frequency component related to the discontinuity between adjacent blocks are measured, and the frequency component in the block and the time frequency component related to the discontinuity between adjacent blocks are estimated. Then, for each of the spatial frequency component and the temporal frequency component, the importance is calculated based on the visual sensitivity function, and the distortion amount obtained as the square error weighted for each frequency is used based on the importance. A cost function is set as a weighted sum with the code amount, and a mode for minimizing the cost function is selected. Therefore, in a block-based coding scheme that combines interframe prediction based on motion compensation and orthogonal transform, there is an advantage that image degradation of a decoded image due to block distortion can be prevented and the code amount can be reduced. can get.

  Further, according to the present invention, when measuring the spatial frequency component in the block, the number of samples in the horizontal and vertical directions is doubled with respect to the orthogonal transform base image used for encoding. A corrected base image to which values are added is generated, and a discrete Fourier transform coefficient is calculated for the corrected base image, thereby calculating a spatial frequency component related to discontinuity within a block and between adjacent blocks. Therefore, in a block-based coding scheme that combines interframe prediction based on motion compensation and orthogonal transform, there is an advantage that image degradation of a decoded image due to block distortion can be prevented and the code amount can be reduced. can get.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

A. Principle of the Present Invention The principle of the present invention will be described. In the present invention, a distortion scale considering block distortion is introduced, and an encoding parameter considering block distortion is set in a cost function used for setting the encoding parameter. Specifically, when calculating the same distortion measure, frequency analysis is also performed for discontinuity with adjacent blocks, and the discontinuity with adjacent blocks is reflected when calculating the distortion measure. Therefore, frequency analysis is performed in consideration of discontinuity between adjacent blocks together with the waveform in the block, and the distortion amount is weighted based on the contrast sensitivity function.

[Calculation of sensitivity coefficient]
In the calculation of the weighting coefficient described above, the input is the transformation matrix and the shift amount. In the following, a case where an image having a vertical width H is observed at a viewing distance rH is considered. r is called a viewing distance parameter.

When the k-th column vector of the transformation matrix [Phi (N × N matrix) to (N-dimensional vector) and phi _k, base image for the matrix is obtained from the following equation. H. In the case of H.264, the possible value for N is either 4 or 8.

Here, φ ^t _l is a transposed vector of φ _l . Each base image f _{k, l} (x, y) (0 ≦ x, y ≦ N−1) is doubled in size by zero padding. As a zero padding method, for example, there are methods like the following formulas (14) and (15).
Zero padding method 1:

Zero padding method 2:

The (˜) f _{k, l} (x, y) obtained as a result of zero padding is called a corrected base image. For example, when N = 4, as shown in FIGS. 9A and 9B, the base image is arranged at 4 × 4 pixels in the shaded portion, and zero values are padded at other positions. (Zero padding). FIG. 9A shows the above-described cell padding method 1, and FIG. 9B shows the above-described cell padding method 2. The modified base image is subjected to a discrete Fourier transform represented by the following equation (16) to obtain a Fourier coefficient. In the following, it is assumed that N = 2 ^m .

Here, j is an imaginary unit.
The Fourier coefficient according to the above equation (16) is shown in the following equation (17).

  Furthermore, the weighting shown in the following equation (18) is performed on the equation (17).

Hereinafter, (^) F _{k, l} (u, v) of the first term on the right side of Expression (18) will be described. Here, g (η, w) is a function known as a visual sensitivity function, and is expressed in a function form as in the following equation (19).

  Here, a, b, c, and d are parameters (hereinafter referred to as model parameters) that define the function form of the visual sensitivity function. For example, values as shown in the following equation (20) are used.

  In the equation (19), η is a value represented by the following equation (21).

  θ (r, H) is an angle per pixel when an image having a vertical width H is observed at a viewing distance rH, and is given by the following equation (22).

ω is an angle change amount per unit time [degrees / sec], and is set as follows. When the macroblock displacement is estimated as (d _x , _dy ) and an image having a vertical width H is observed at the viewing distance rH, the angle variation per unit time is given by the following equation (23). .

Here, _fr is a frame rate. The sensitivity coefficient for the k, l base is defined as the power ratio represented by the following equation (24).

Note that W ^U _{k, l} (ω) and W ^v _{k, l} (ω) can be similarly obtained. At this time, it is also possible to change the model parameter between the luminance component and the color difference component.

B. First Embodiment FIG. 1 is a block diagram showing a configuration of a mode selection device according to a first embodiment of the present invention. In the figure, a transition amount storage unit 300 stores an estimated transition amount. The initial mode setting unit 301 sets an initial mode. The mode storage unit 302 stores the initially set mode and appropriately stores (updates) the mode set by the mode setting unit 316 described later.

  The code amount calculation unit 303 calculates the code amount based on the input encoding target frame signal, the reference frame signal, and the quantization parameter. The code amount storage unit 304 stores the calculated code amount. The weighted distortion amount calculation unit 305 calculates the weighted distortion amount from the transition amount stored in the transition amount storage unit 300 according to the mode stored in the mode storage unit 302. The weighted distortion amount storage unit 306 stores the calculated weighted distortion amount. The undetermined multiplier calculation unit 307 calculates an undetermined multiplier based on the input encoding target frame signal, the reference frame signal, and the quantization parameter. The undetermined multiplier storage unit 308 stores the calculated undetermined multiplier.

  The cost calculation unit (cost function setting unit) 309 stores the code amount stored in the code amount storage unit 304, the weighted distortion amount stored in the weighted distortion amount storage unit 306, and the undetermined multiplier storage unit 308. The cost is calculated based on the stored undetermined multiplier. The cost storage unit 310 stores the calculated cost. The minimum cost determination unit (mode selection unit) 311 determines the minimum cost from the costs stored in the cost storage unit 310. The minimum cost storage unit 312 stores the minimum cost determined by the minimum cost determination unit 311.

  When the minimum cost determination unit 311 determines that the calculated cost is smaller than the minimum cost stored in the register, the optimum mode update unit 313 stores the mode stored in the mode storage unit 302 at that time. Are stored in the optimum mode storage unit 314. The optimum mode storage unit 314 stores the mode at the time when it is determined that the calculated cost is smaller than the minimum cost stored in the register as the optimum mode.

  The final mode determination unit 315 determines whether or not the processing has been completed for all modes, that is, whether or not the mode is the final mode. If the final mode is not the final mode, the determination result of the minimum cost determination 311 is displayed as a mode setting unit (mode Selection means) 316, and in the final mode, the determination result of the minimum cost determination 311 is supplied to the optimum mode output unit 316. When the determination result of the minimum cost determination 311 is supplied from the final mode determination unit 315, that is, when the processing is completed for all modes, the optimal mode output unit 316 displays the optimal mode stored in the optimal mode storage unit 314. Output.

  FIG. 2 is a block diagram showing the configuration of the weighted distortion amount calculation unit 305 described above. In the figure, a conversion coefficient normalization unit 401 normalizes input conversion coefficients. The normalized conversion coefficient storage unit 402 stores the normalized conversion coefficient. The transform coefficient decoding unit 403 decodes input transform coefficients. The decoded transform coefficient storage unit 404 stores the decoded transform coefficient. The transition amount storage unit 408 stores the input conversion amount.

  A sensitivity coefficient calculation unit (spatial frequency component measurement unit, temporal frequency component estimation unit, importance calculation unit, spatial frequency component calculation unit) 409 calculates a sensitivity coefficient from the shift amount stored in the shift amount storage unit 408. The sensitivity coefficient storage unit 410 stores the calculated sensitivity coefficient. The sensitivity coefficient multiplication unit 407 reads the distortion amount from the distortion amount storage unit 406 and multiplies the sensitivity coefficient read from the sensitivity coefficient storage unit 410. The distortion amount storage unit 411 stores the distortion amount multiplied by the sensitivity coefficient. The distortion amount sum calculation unit 412 adds the distortion amounts read from the distortion amount storage unit 411 and outputs the result as a weighted distortion amount.

  FIG. 3 is a block diagram illustrating a configuration of the sensitivity coefficient calculation unit 409 described above. In the figure, a conversion coefficient storage unit 601 stores input conversion coefficients. The transformation matrix size storage unit 602 stores the inputted transformation matrix size. The base image calculation unit 603 calculates a base image according to the transform coefficient read from the transform coefficient storage unit 601 and the transform matrix size read from the transform matrix size storage unit 602. The base image storage unit 604 stores the calculated base image.

  A corrected base image calculation unit (corrected base image generation unit) 605 calculates a corrected base image from the base image read from the base image storage unit 604 according to the transformation matrix size read from the transformation matrix size storage unit 602. The corrected base image storage unit 606 stores the corrected base image. The contrast sensitivity function storage unit 607 stores the input contrast sensitivity function. The multiplication unit 608 multiplies the corrected base image read from the corrected base image storage unit 606 by the contrast sensitivity function read from the contrast sensitivity function storage unit 607. Adder 609 adds the multiplied values and outputs the result as a sensitivity coefficient.

  FIG. 4 is a flowchart for explaining the operation of the encoding mode selection process according to this embodiment. First, the initial mode setting unit 301 sets the initial value of the mode (step S101). Specifically, a register storing the minimum cost and optimum mode is initialized. Next, the transition amount is stored in the transition amount storage unit 300 (step S102). Next, the code amount calculation unit 303 calculates the code amount (step S103), and the weighted distortion amount calculation unit 305 calculates the weighted distortion amount (step S104). Details of the calculation of the weighted distortion amount will be described later.

  Next, an undetermined multiplier calculation unit calculates an undetermined multiplier (step S105), and a cost calculation unit 309 calculates a cost based on the code amount, the weighted distortion amount, and the undetermined multiplier (step S106). Next, the minimum cost determination unit 311 determines whether or not the calculated cost is smaller than the minimum value of the cost stored in the minimum cost storage unit 312 (register) (step S107). The mode is changed (step S111), the process returns to step S103, and the code amount, weighted code amount, undetermined multiplier, and cost are calculated in the changed mode.

  On the other hand, when the calculated cost is smaller than the minimum cost stored in the register, the calculated cost is stored (step S108), and the mode at that time is stored (step S109). Next, it is determined whether or not the processing has been completed for all modes (step S110). If the processing has not been completed, the mode is changed (step S111), and the process returns to step S103. Repeat the process. On the other hand, when the process has been completed for all modes, the process ends.

  FIG. 5 is a flowchart showing the calculation procedure of the weighted distortion amount in step S104. First, the transform coefficient normalization unit 401 normalizes the input transform coefficient (step S201), and the transform coefficient decoding unit 403 decodes the input transform coefficient (step S202). Next, the conversion amount is read (step S203), the sensitivity coefficient calculation unit 409 calculates the sensitivity coefficient, and sets the sensitivity coefficient (step S204). Next, the counter i and the register S (distortion amount storage unit 411) are initialized to 0 (step S205), and the distortion amount calculation unit 405 calculates the encoding distortion of the i-th component of the transform coefficient (step S206).

  Next, the sensitivity coefficient multiplication unit 407 multiplies the encoding distortion of the i-th component of the transform coefficient by the sensitivity coefficient (step S207), and the distortion amount sum calculation unit 412 multiplies the sensitivity coefficient by the encoding distortion (distortion amount). Is added to the register S (distortion amount storage unit 411) (step S208). Next, it is determined whether or not the processing has been completed for all components of the transform coefficient (step S209). If not all have been completed, 1 is added to the counter i (step S210), and the process returns to step S206. The above-described process is repeated for the other component. Thereafter, 1 is added to the counter i until the processing is completed for all components, and the above-described processing is repeated. Then, when the process is completed for all components of the conversion coefficient, the process ends.

  FIG. 6 is a flowchart showing the procedure for setting the sensitivity coefficient in step S204. First, the size (N × N) of the transformation matrix (DCT) is read (step S501), and the counter i is initialized to 0 (step S502). Next, the base image calculation unit 603 calculates the i-th base image in the transformation matrix (DCT) (step S503), and the corrected base image calculation unit 605 performs zero padding according to the equation (21) described above. Then, an image twice as large as the base image (2N × 2N image: corrected base image) is generated (step S504).

  Next, frequency analysis (DFT) is performed on the corrected base image, and the distribution of frequency components in the corrected base image is calculated (step S505). Next, the contrast sensitivity function (CSF) is read (step S506), and the counter k and the register W [i] are initialized to 0 (step S507). Next, the multiplication unit 608 multiplies the k-th component of the frequency component of the corrected base image by the corresponding value of the contrast sensitivity function according to the equation (22) described above (step S508), and the addition unit 609 multiplies the multiplication. Is added to the register W [i] (step S509).

  Next, it is determined whether or not the processing has been completed for all the frequency components of the modified base image (step S510). If the processing has not been completed for all the frequency components, 1 is added to the counter k (step S511). Returning to S508, the above-described processing is repeated. On the other hand, if the processing has been completed for all the frequency components, it is determined whether or not the processing has been completed for all the base images (step S512). 1 is added (step S513), the process returns to step S503, and the above-described processing is repeated. Then, when the processing is completed for all base images, the processing ends.

C. Second Embodiment Next, a second embodiment of the present invention will be described. In the first embodiment described above, the example is based on the mode selection, but in the second embodiment, the quantization parameter is selected.

  FIG. 7 is a block diagram showing the configuration of the quantization parameter selection apparatus according to the second embodiment. In the figure, a transition amount storage unit 800 stores an estimated transition amount. The initial quantization parameter setting unit 801 sets initial quantization parameters. The quantization parameter storage unit 802 stores the initially set quantization parameter, and appropriately stores (updates) the quantization parameter set by the quantization parameter setting unit 816 described later.

  The code amount calculation unit 803 calculates the code amount based on the input encoding target frame signal, the reference frame signal, and the quantization parameter. The code amount storage unit 804 stores the calculated code amount. The weighted distortion amount calculation unit 805 calculates the weighted distortion amount from the lower limit value of the transition amount stored in the transition amount storage unit 800 according to the quantization parameter stored in the quantization parameter storage unit 802. The weighted distortion amount storage unit 806 stores the calculated weighted distortion amount. The undetermined multiplier calculation unit 807 calculates an undetermined multiplier based on the input encoding target frame signal, the reference frame signal, and the quantization parameter. The undetermined multiplier storage unit 808 stores the calculated undetermined multiplier.

  The cost calculation unit 809 includes a code amount stored in the code amount storage unit 804, a weighted distortion amount stored in the weighted distortion amount storage unit 806, and an undetermined multiplier stored in the undetermined multiplier storage unit 808. Based on the above, the cost is calculated. The cost storage unit 810 stores the calculated cost. The minimum cost determination unit 811 determines the minimum cost from the costs stored in the cost storage unit 810. The minimum cost storage unit 812 stores the minimum cost determined by the minimum cost determination unit 811.

  When the minimum cost determination unit 811 determines that the calculated cost is smaller than the minimum value of the cost stored in the register, the optimal quantization parameter update unit 813 stores it in the quantization parameter storage unit 802 at that time. The quantization parameter being stored is stored in the optimum quantization parameter storage unit 814. The optimum quantization parameter storage unit 814 stores, as the optimum quantization parameter, the quantization parameter at the time when it is determined that the calculated cost is smaller than the minimum cost stored in the register.

  The final quantization parameter determination unit 815 determines whether or not processing has been completed for all the quantization parameters, that is, whether or not the final quantization parameter, and if not, the minimum cost determination 811 The determination result is supplied to the quantization parameter setting unit 816. If the determination result is the final quantization parameter, the determination result of the minimum cost determination 811 is supplied to the optimal quantization parameter output unit 816. The optimal quantization parameter output unit 816 receives the determination result of the minimum cost determination 811 from the final quantization parameter determination unit 815, that is, when the processing is completed for all the quantization parameters, the optimal quantization parameter storage unit 814 The optimal quantization parameter stored in is output.

  FIG. 8 is a flowchart for explaining the operation of the quantization parameter selection process according to the second embodiment. First, the initial quantization parameter setting unit 801 sets the initial value of the quantization parameter (step S700). Specifically, a register storing the minimum cost and optimum quantization parameter is initialized. Next, the transition amount is stored in the transition amount storage unit 800 (step S701). Next, the code amount calculation unit 803 calculates the code amount (step S703), and the weighted distortion amount calculation unit 805 calculates the weighted distortion amount (step S704). The details of the calculation of the weighted distortion amount are the same as in the first embodiment described above.

  Next, the undetermined multiplier calculation unit 807 calculates the undetermined multiplier (step S705), and the cost calculation unit 809 calculates the cost based on the code amount, the weighted distortion amount, and the undetermined multiplier (step S706). Next, the minimum cost determination unit 811 determines whether or not the calculated cost is smaller than the minimum value of the cost stored in the minimum cost storage unit 812 (register) (step S707). In step S711, the quantization parameter is changed, and the process returns to step S703, and the code amount, the weighted code amount, the undetermined multiplier, and the cost are calculated using the changed quantization parameter.

  On the other hand, when the calculated cost is smaller than the minimum value of the cost stored in the register, the calculated cost is stored in the minimum cost storage unit 812 (step S108), and the quantization parameter at that time is stored (step S108). Step S709). Next, it is determined whether or not the processing has been completed for all the quantization parameters (step S710). If not, the quantization parameter is changed (step S711), and the process returns to step S703 to change the changed quantum. The above-described processing is repeated with the optimization parameter. On the other hand, when the processing has been completed for all the quantization parameters, the processing ends.

  According to the first and second embodiments described above, a distortion measure considering block distortion is introduced, and in the cost function used for setting the encoding parameter, the discontinuity with adjacent blocks is calculated when the distortion measure is calculated. In contrast, frequency analysis is performed and the amount of distortion is weighted based on the contrast sensitivity function in order to reflect discontinuity with adjacent blocks when calculating the distortion measure. There is an advantage that image degradation of the decoded image can be prevented.

It is a block diagram which shows the structure of the mode selection apparatus by 1st Embodiment of this invention. It is a block diagram which shows the structure of the weighted distortion amount calculation part 305 by this 1st Embodiment. It is a block diagram which shows the structure of the sensitivity coefficient calculation part 409 by this 1st Embodiment. It is a flowchart for demonstrating operation | movement of the encoding mode selection process by this embodiment. It is a flowchart which shows the calculation procedure of the weighted distortion amount by this 1st Embodiment. It is a flowchart which shows the setting procedure of the sensitivity coefficient by this 1st Embodiment. It is a block which shows the structure of the quantization parameter selection apparatus by 2nd Embodiment of this invention. It is a flowchart for demonstrating operation | movement of the quantization parameter selection process by this 2nd Embodiment. It is a conceptual diagram for demonstrating the zero padding method. It is a table | surface figure which shows the example of the symbol used for description in this specification. It is a conceptual diagram for demonstrating the subject of a prior art.

Explanation of symbols

300 Transition amount storage unit 301 Initial mode setting unit 302 Mode storage unit 303 Code amount calculation unit 304 Code amount storage unit 305 Weighted distortion amount calculation unit 306 Weighted distortion amount storage unit 307 Undetermined multiplier calculation unit 308 Undetermined multiplier storage unit 309 Cost Calculation unit 310 Cost storage unit 311 Minimum cost determination unit 312 Minimum cost storage unit 313 Optimal mode update unit 314 Optimal mode storage unit 315 Final mode determination unit 316 Optimal mode output unit 401 Conversion coefficient normalization unit 402 Normalized conversion coefficient storage unit 403 Transform coefficient decoding unit 404 Decoding transform coefficient storage unit 407 Sensitivity coefficient multiplication unit 408 Transition amount storage unit 409 Sensitivity coefficient calculation unit 410 Sensitivity coefficient storage unit 411 Distortion amount storage unit 412 Distortion amount sum calculation unit 601 Conversion coefficient storage unit 602 Conversion matrix size Storage unit 603 Base image calculation 604 base image storage unit 605 fixes the base image calculating unit 606 fixes the base image storage unit 607 contrast sensitivity function storage unit 608 multiplying unit 609 adding unit

Claims (5)

In an image coding method for performing information compression by transform coding and quantization on an image signal or a prediction error signal obtained by intra-frame prediction and inter-frame prediction,
Based on Lagrangian cost function consisting of distortion, code amount, and undetermined multiplier, motion compensation block size in video coding, inter prediction mode, quantization parameter, quantization parameter in still image coding, intra prediction mode, etc. Measuring the frequency component in the block and the spatial frequency component related to the discontinuity between adjacent blocks when determining the encoding parameters of
And estimating a frequency component within the block and a time frequency component related to the discontinuity between adjacent blocks;
Calculating importance based on a visual sensitivity function for each of the spatial frequency component and the temporal frequency component; and
Setting a cost function as a weighted sum with a code amount using a distortion amount obtained as a square error weighted for each frequency based on the importance;
Selecting a mode that minimizes the cost function;
An image encoding method comprising: When measuring the spatial frequency components in a block, a modified base image with zero values added so that the number of samples in the horizontal and vertical directions is doubled with respect to the base image of orthogonal transform used for encoding. Generating step;
Calculating a spatial frequency component in a block and a discontinuity between adjacent blocks by calculating a discrete Fourier transform coefficient for the modified base image, further comprising: Item 2. The image encoding method according to Item 1. In an image coding apparatus that performs information compression by transform coding and quantization on an image signal or a prediction error signal obtained by intra-frame prediction and inter-frame prediction,
Based on Lagrangian cost function consisting of distortion, code amount, and undetermined multiplier, motion compensation block size in video coding, inter prediction mode, quantization parameter, quantization parameter in still image coding, intra prediction mode, etc. Spatial frequency component measurement means for measuring the frequency component in the block and the spatial frequency component related to the discontinuity between adjacent blocks when determining the encoding parameter of
Furthermore, a time frequency component estimation means for estimating a frequency component in a block and a time frequency component related to discontinuity between adjacent blocks;
Importance calculating means for calculating importance based on a visual sensitivity function for each of the spatial frequency component and the temporal frequency component;
Cost function setting means for setting a cost function as a weighted sum with a code amount using a distortion amount obtained as a square error weighted for each frequency based on the importance,
Mode selection means for selecting a mode that minimizes the cost function;
An image encoding device comprising: The spatial frequency measuring means includes
When measuring the spatial frequency components in a block, a modified base image with a zero value added so that the number of samples in the horizontal and vertical directions is twice that of the orthogonal transform base image used for encoding. Modified base image generation means for generating;
Spatial frequency component calculation means for calculating a spatial frequency component in the block and a spatial frequency component related to discontinuity between adjacent blocks by calculating a discrete Fourier transform coefficient for the modified base image. The image encoding device according to claim 3. To a computer that controls an image encoding device that performs information compression by transform coding and quantization on an image signal or a prediction error signal obtained by intra-frame prediction and inter-frame prediction,
Based on Lagrangian cost function consisting of distortion, code amount, and undetermined multiplier, motion compensation block size in video coding, inter prediction mode, quantization parameter, quantization parameter in still image coding, intra prediction mode, etc. Measuring the frequency component in the block and the spatial frequency component related to the discontinuity between adjacent blocks when determining the encoding parameters of
And estimating a frequency component within the block and a time frequency component related to the discontinuity between adjacent blocks;
Calculating importance based on a visual sensitivity function for each of the spatial frequency component and the temporal frequency component; and
Setting a cost function as a weighted sum with a code amount using a distortion amount obtained as a square error weighted for each frequency based on the importance;
Selecting a mode that minimizes the cost function;
An image encoding program characterized in that

Images