JP2019097084A - Signal conversion device - Google Patents

Abstract

To provide a signal conversion device capable of reducing an error caused by interpolation and suppressing a data amount in a three-dimensional look-up table.SOLUTION: A signal conversion device 1 includes: 3D-LUT storage means 11 for storing a 3D-LUT; input conversion means 12 for dividing the signal value of an input video signal by a signal extraction interval and extracting the integer part and the decimal part of the divided value; 3D-LUT retrieving means 13 which retrieves the 3D-LUT and obtains a three-dimensional region corresponding to a cube of the integer part; and signal arithmetic processing means 14 which performs interpolation by a high-order equation representing a position corresponding to the decimal part in the three-dimensional region.SELECTED DRAWING: Figure 6

Description

  The present invention relates to a signal conversion device that converts a signal format.

With the development of imaging technology, many signal formats are proposed, and signal conversion processing for converting each signal format is also studied. Conventionally, for example, resolution (number of pixels) conversion from SDTV (Standard definition television) to HDTV (High definition television). In addition, with respect to the signal format of the video signal, converts the RGB and _YP B _{P R} (luminance and color difference signals), with respect to color space, BT. 709 (HDTV), BT. As for 2020 (4K, 8K) and DCI-P3 (digital cinema), SDR (standard dynamic range) and HDR (hydrodynamic range: HLG method or PQ method) can be cited as examples of the electro-optical conversion. There are ten or more types of combinations of these conversions. Because there are so many signal formats, separate signal conversion processes are required among them. Hereinafter, the signal format is abbreviated as “format”, the color space is abbreviated as “color gamut”, and the electric / light conversion is abbreviated as “gamma”.

These signal conversion processes differ in processing content according to the signal format before and after conversion. For example, when only gamma is different in luminance and color difference signals, signal conversion processing is as shown in FIG. That, _YP B _P a _R signal is converted into RGB signals at 3 × 3 matrix, the HLG lookup table: Convert the SDR in (LUT Look up table), then, _YP B P RGB signals in 3 × 3 matrix Convert to _R signal. On the other hand, when the color gamut is different, the signal conversion process is as shown in FIG. That is, the PQ method is linearly converted by the LUT, and the BT. BT.2020 in a 3 × 3 matrix. Convert to 709 and then convert the linear to PQ format with LUT. As described above, since signal conversion processing differs for each target signal format, it is difficult to use the same hardware to cope with a plurality of signal formats.

  As a method corresponding to a plurality of signal formats using the same hardware, as shown in FIG. 13, signal conversion processing may be performed using a three-dimensional look-up table (3D-LUT). It is considered (for example, patent document 1). The three-dimensional look-up table defines the correspondence between the input and output of the video signal in the three-dimensional space, and can correspond to various signal conversion processes.

In this 3D-LUT, the amount of data often becomes a problem. In FIG. 13, the input video signal requires 30 bits (3D × 10 bits) and a data area of 2 ³⁰ = 1.07 × 10 ⁹ is required, and since the output video signal is 30 bits, it is multiplied by 30. That is, the 3D-LUT requires a data area of 1.07 × 10 ⁹ × 30 = 3.2 × 10 ¹⁰ bits, which is not realistic. Therefore, the input to the 3D-LUT is roughened (the data of the 3D-LUT is thinned), and the thinned data is interpolated by linear interpolation. In FIG. 13, since the input to the 3D-LUT is thinned out at 100 steps per dimension, the data amount of the 3D-LUT can be reduced to 3.0 × 10 ⁷ bits.

JP, 2016-139096, A

  However, when data of the 3D-LUT is thinned out, there is a problem that an error due to interpolation increases.

  Then, this invention makes it a subject to reduce the difference | error by interpolation and to provide the signal conversion apparatus which can suppress the data amount of a three-dimensional lookup table.

  In view of the above problems, a signal conversion device according to the present invention is a signal conversion device that converts the signal format of an input signal, and includes a table storage unit, an extraction unit, a table search unit, and an interpolation unit. It was set as having composition.

  According to this signal conversion device, the table storage means represents a cube whose apex is a grid point arranged at a predetermined signal extraction interval in a three-dimensional space representing the signal value of the input signal, and represents the signal value of the output signal. A three-dimensional look-up table is stored in advance, which is associated with a three-dimensional area having vertices corresponding to the lattice points in a three-dimensional space.

  Further, the extraction means receives the input signal to be converted, obtains the value obtained by dividing the signal value of the input signal to be converted by the signal extraction interval, and divides the value from the divided value to obtain the grid of the cube. Extract an integer part representing a point and a decimal part representing a position in the cube.

The table search means searches the three-dimensional lookup table to obtain a three-dimensional area corresponding to a cube of the integer part.
Then, a high-order equation for obtaining a position corresponding to the decimal part in the three-dimensional area acquired by the table search means is set in advance, and the interpolation means interpolates with the high-order equation to obtain the fraction The output signal of the signal value corresponding to the position of the part is generated.

According to the present invention, the following excellent effects can be obtained.
According to the present invention, since interpolation is performed in a high order equation, an error due to interpolation can be reduced, and the data amount of the three-dimensional lookup table can be suppressed.

It is the schematic of the signal conversion process in embodiment of this invention. It is explanatory drawing explaining 3D-LUT. It is explanatory drawing explaining the relationship between the input to 3D-LUT, and the output by high-order equation interpolation. It is explanatory drawing which demonstrates the cube of FIG. 3 on a plane. It is explanatory drawing explaining the output from 3D-LUT. It is a block diagram showing composition of a signal conversion device concerning an embodiment of the present invention. It is a flowchart which shows 3D-LUT production | generation processing by the signal conversion apparatus of FIG. It is a flowchart which shows the signal conversion process by the signal conversion apparatus of FIG. In the Example of this invention, it is a graph which shows the relationship between the bit number of 3D-LUT, and an error. In the Example of this invention, it is a graph which shows the relationship between the data amount of 3D-LUT, and an error. It is an explanatory view explaining an example of conventional signal conversion processing. It is an explanatory view explaining an example of conventional signal conversion processing. It is explanatory drawing explaining the conventional 3D-LUT.

Hereinafter, each embodiment of the present invention will be described in detail with reference to the drawings as appropriate. In each embodiment, the same reference numeral is given to the same means, and the description is omitted.
First, after the signal conversion processing in the first embodiment is described, the configuration and operation of the signal conversion device according to the first embodiment will be described.

First Embodiment
[Signal conversion processing]
In this signal conversion process, the input to the 3D-LUT is roughened to reduce the amount of data in the 3D-LUT, and high-order interpolation is performed to perform conversion with higher precision than simple linear interpolation. It is characterized by what is possible.

In the present embodiment, video signals are treated as input signals and output signals, but the present invention is not limited to this. The input video signal has three dimensions of V _ax , V _ay , and V _az , and takes values of 0 to N (x, y, z signals of a certain system a). The output video signal is three-dimensional of V _bx , V _by and V _bz (x, y, z signals of another system b). The signal value of these V _{ax to} V _bz takes a value of 0 to N, and in the case of 10 bits, the signal value becomes 0 to 1023 (N = 1023).

In order to roughen the input to the 3D-LUT, predetermined bits of the input video signal are extracted. The input to the _{3D-LUT m ax, m ay} , and _{m az,} the output from the 3D-LUT k-number of _{f k (m ax, m ay} , m az) and. This k represents the number of intermediate variables when obtaining the output of the 3D-LUT, which is a multiple of the number of dimensions of the x, y, z signals (in this embodiment, the number of dimensions = 3). Here, m _ax , m _ay and m _az take values of 0 to N / M. M represents a signal extraction interval, and is obtained by subtracting the number of bits to be extracted from the number of bits per one dimension of the video signal, and powering 2 by the value after subtraction. For example, when the video signal extracts 10 bits and high order 4 bits per one dimension, the signal extraction interval M = 2 ^{10 -4} = 2 ⁶ = 64.

Further, it is assumed that accurate output video signals are C _x (V _ax , V _ay , V _az ), C _y (V _ax , V _ay , V _az ), and C _z (V _ax , V _ay , V _az ). The accurate output video signals are the x, y and z signals of the system b corresponding to the input video signals V _ax , V _ay and V _az , which are signals calculated according to the processing required for the conversion of the video signal. That is, the correct output video signal represents a true value output video signal when converted without using the 3D-LUT so that a conversion error does not occur.

As shown in FIG. 1, the signal conversion process is composed of input conversion, search of a 3D-LUT, and signal operation process.
First, input conversion will be described. In this input conversion, the following equations (1) and (2) are calculated, and the input video signals V _ax , V _ay and V _az are converted into integer parts m _ax , m _ay and m _az and residuals Δx, Δy, Determine Δz. The relationship between the input video signal and the 3D-LUT is expressed by the following equation (1). That is, equation (1) means that integer parts m _ax , m _ay and m _az obtained by dividing the signal values of the input video signals V _ax , V _ay and V _az by the signal extraction interval M are obtained. In equation (1), '[]' represents a function that returns an integer number of arguments.

Further, residuals Δx, Δy, Δz are determined by the following (2). The residuals .DELTA.x, .DELTA.y, .DELTA.z represent fractional parts when the signal values of the input video signals V _ax , V _ay and V _az are divided by the signal extraction interval M, and take values of 0 to 1.0.

FIG. 2 illustrates a three-dimensional space of an orthogonal coordinate system consisting of an X-axis, a Y-axis and a Z-axis corresponding to the input video signals V _ax , V _ay and V _az respectively. In this case, the integer parts m _ax , m _ay and m _az of the equation (1) correspond to the grid points of the cube of FIG. 2 and are input values to the 3D-LUT. Further, the residuals Δx, Δy, Δz in the equation (2) indicate positions between lattice points, that is, positions in the cube.
Although FIG. 2 illustrates five lattice points in the X-axis, Y-axis, and Z-axis directions, the number of lattice points is not particularly limited, and may be five or more.

Subsequently, search of the 3D-LUT will be described.
The 3D-LUT is a cube whose apexes are grid points arranged at signal extraction intervals M in a three-dimensional space representing signal values of input video signals V _ax , V _ay and V _az , and output video signals V _bx , V _by , It corresponds to a three-dimensional area having a vertex corresponding to a grid point in a three-dimensional space representing a signal value of V _bz .

The search for the 3D-LUT, the integer part _{m ax, m ay,} searches the 3D-LUT by _{m az,} obtaining the output _f k from _{3D-LUT (m ax, m} ay, m az) a. Here, when the input to the 3D-LUT is m _ax , m _ay , and m _az , the output from the 3D-LUT is the third term of the equation (3).

Subsequently, signal calculation processing will be described. In this signal computation processing is performed the output _f k from _{3D-LUT (m ax, m} ay, m az) and the residual [Delta] x, [Delta] y, interpolation in two dimensions equation using Delta] z, output video signal V _{bx, V} by, determine the _{V bz.}

In the conventional signal arithmetic processing, linear interpolation is performed as shown in the following equation (4) to obtain the output video signal _Vbx . This equation (4) is, as shown in FIG. 3, residuals Δx, Δy, inside a cube corresponding to integer parts m _ax , m _ay and m _az in a three-dimensional space representing signal values of the input video signal. This means that the position of one point indicated by Δz is linearly interpolated from the values of eight vertices f ₁ which are the output of the 3D-LUT.
The output video signals V _by and V _bz can be expressed in the same manner as the output video signal V _bx of equation (4) using f ₂ and f ₃ of equation (3).

In the present embodiment, quadratic interpolation is used instead of the linear interpolation of equation (4). The conventional equation (4) is linear with respect to the residuals Δx, Δy, Δz. On the other hand, in the quadratic interpolation, the residuals Δx, Δy, Δz become quadratic expressions. Since the equation (4) is simply converted to a _quadratic equation, it becomes complicated, so the difference h _{1 x} between the output video signal V _bx of the equation (4) and the accurate output video signal C _x as shown in the following equation (5) Use

The equation (5) has the feature represented by the equation (6). Equation (6) represents that the difference h ₁ x becomes 0 at the vertex of the cube of FIG. Specifically, equation (6) represents that the difference h _1x (Δx, Δy, Δz) becomes 0 when all of the residuals Δx, Δy, Δz are 0 or 1.

Therefore, the difference h _{1 x} may be reduced by performing high-order interpolation such as second-order interpolation. Of course, in the three-dimensional space of FIG. 3, each cube corresponds to the input m _ax , m _ay , m _az to the 3D-LUT, so it is adjacent to the other cube (shown in dashed lines). Therefore, when adding a correction term to be described later, the correction term needs to be continuous between adjacent cubes (the same correction value is obtained). Otherwise, the signal value of the output video signal becomes discontinuous with respect to the input video signal in which the signal values are continuous. In order to avoid this discontinuity, it is necessary to make the correction value the same at each vertex, each side, and each face of the adjacent cube.

  Thereafter, a plane having the X axis as a normal (a plane parallel to the Y and Z axes) is a Y-Z plane, and a plane having the Y axis as a normal (a plane parallel to the X and Z axes) is an X-Z A plane (a plane parallel to the X-axis and the Y-axis) having a Z-axis as a normal is an XY plane.

Here, a supplementary explanation of the correction term will be made. In order to obtain an accurate output video signal, it is necessary to determine the correspondence between the input video signal and the output video signal for all points included in the cube of FIG. In the present embodiment, in order to reduce the data amount of the 3D-LUT, only the vertexes of the cube corresponding to the integer parts m _ax , m _ay , and m _az define the correspondence between the input video signal and the output video signal. . Then, in a three-dimensional area corresponding to a cube, one point corresponding to the residuals Δx, Δy, Δz is interpolated as a signal value of the output video signal. At this time, the calculation formula from the residuals Δx, Δy, Δz is taken as a correction term so as to approach the signal value of the output video signal more accurate than mere linear interpolation. In this embodiment, this correction term is a quadratic expression or a cubic expression of the residuals Δx, Δy, Δz.

FIG. 4 illustrates the cube of FIG. 3 on the XZ plane. In FIG. 4, a square (0 ≦ X ≦ 1, 0 ≦ Z ≦ 1) including the origin 0 corresponds to the plane of the cube illustrated by the solid line in FIG. Further, a square (0 ≦ X ≦ 1, 1 ≦ Z ≦ 2) illustrated by a broken line in FIG. 4 corresponds to a plane of a cube illustrated by a broken line in FIG. Here, it is assumed that the correct output video signal F (x, z) is known for all points included in the two squares. In order to minimize the difference between the interpolated output video signal F _a (x, z) and the accurate output video signal F (x, z) by performing interpolation at all points included in the solid square. Determine 3D-LUT. As with the solid square, the 3D-LUT is used to minimize the difference between the interpolated output video signal F _b (x, z) and the accurate output video signal F (x, z), also for the dashed square. Ask for As shown in FIG. 4, at Z = 1, F _a (x, z) and F _b (x, z) need to be continuous, but separately F _a (x, z) and F _b (x, There is no guarantee for calculating z). For this reason, in the present embodiment, the boundary of the square and the portion other than the boundary are distinguished so as not to be discontinuous at the boundary of the adjacent square.

Returning to FIG. 3, the description of the signal arithmetic processing is continued. In the cube of FIG. 3, a plane having eight vertices, four sides each parallel to each of the X axis, Y axis, and Z axis with a total of 12 and each having X axis, Y axis, and Z axis as a normal There are a total of eight in two on each side. For example, since each of i ₂ and i ₃ takes a value of 0 or 1, the correction terms at four sides passing two points (m _ax , m _ay + i ₂ , m _az + i ₃ ) parallel to the X axis are f _{It is} assumed that _line-x (m _ax + Δx, m _ay + i ₂ , m _az + i ₃ ). Also, since i ₁ takes a value of 0 or 1, the correction terms in the two YZ planes passing through the point (m _ax + i ₁ , m _ay , m _az ) are f _{plane −yz} (m _ax + i ₁ , It is set as _ma y + Δy, m _az + Δz). In this case, the quadratic interpolation is expressed by the following equation (7).

In the right side of the equation (7), three correction terms fline _{-x to} fline _-z interpolate correction values on sides parallel to the X axis, Y axis, and Z axis, respectively. The three correction terms f _{plane-yz to} f _plane-xy interpolate correction values in the Y-Z plane, the X-Z plane, and the XY plane. The final correction term f _vol is such that the correction value on each side and each surface is zero.
The differences h _1y and h _1z between the output video signals V _by and V _bz and the accurate output video signals C _y and C _z can also be expressed in the same manner as in equations (5) to (7).

Here, as a premise of the equation (7), each correction term needs to be 0 at the start point and the end point as in the following equation (8). For example, in the equation (8), when the residual Δx is 0 or 1, it represents that the correction term f _{line -x} becomes 0. The start point and the end point indicate when the residuals Δx, Δy, Δz are 0 or 1, and correspond to the vertexes (integer parts m _ax , m _ay , m _az ) of the cube in FIG.

Under the condition of this equation (8), for example, on the Y-Z plane where residual Δx = 0, equation (7) can be used as a correction term f _line-y (m _ax , m _ay + Δy, m _az + i ₃ ), correction term of the Z-axis component f _line-z (m _ax , _ma y + i ₂ , m _az + Δz), correction term of the Y-Z plane f _plane-yz (m _ax , m _ay It is represented only by + Δy, m _az + Δz). Since these correction terms are completed in this YZ plane, they have the same correction value in adjacent cubes. Therefore, in the equation (7), the correction of each side by the correction terms f _{line -x to} f _line -z does not affect the other side (correction value = 0), and the correction terms f _{plane -yz to} f _{plane -xy} The correction of each surface does not affect the other surface (correction value = 0), and the correction by the correction term f _vol does not affect the sides and surfaces (correction value = 0).

In quadratic interpolation, the basic quadratic equation g ₂ is defined as the following equation (9) (the same applies to the residuals Δy and Δz). This equation (9) represents a function whose return value is 0 when the residual Δx is 0 or 1. Moreover, each correction term of Formula (7) is represented by the following Formula (10).

  Using this equation (10), equation (7) can be defined as the following equation (11).

Then, Equation (5) and Equation (11) are put together, and the accurate output video signal _Cx is approximated by the following Equation (12) (the same applies to the accurate output video signals C _y and C _z ). This equation (12) represents a position corresponding to the residuals Δx, Δy, Δz within a three-dimensional area corresponding to the cube of the integer parts m _ax , m _ay , m _az .

In correspondence with the signal conversion process of FIG. 1, the signal arithmetic process is expressed by equation (12). Therefore, the output of the 3D-LUT is each coefficient included in equation (12). That is, the output of the 3D-LUT has eight coefficients f ₁ related to the correction terms at each vertex of the cube, and 12 coefficients C _{line-x to} C _line-z related to the correction terms at each side of the cube The number of coefficients C _{plane-yz to} C _{plane-x x} related to the correction terms in each surface of the cube and the cube is six, and the coefficient C _vol related to the correction terms in the cube is one in total, 27 in total.

In other words, the coefficients f _{1 to} C _plane-xy in equation (12) represent the positional relationship between the cube, in the input video signal, and the three-dimensional region in the output video signal, with respect to vertices, sides, and faces. For example, eight vertexes contained in a cube of FIG. _{3 (m ax, m ay,} m az) ~ (m ax + 1, m ay + 1, m az +1) is eight vertexes of the three-dimensional region of FIG. 5 It corresponds to f ₁ (m _ax , m _ay , m _az ) to f ₁ (m _ax +1, m _ay +1, m _az +1).
Further, the coefficient C _vol of the equation (12) represents a position corresponding to the residuals Δx, Δy, Δz in the three-dimensional area of FIG.

Here, each vertex of a cube is shared by eight adjacent cubes, each side is shared by four cubes, and each face is shared by two cubes. As a result, the output from the 3D-LUT is eight as shown in the following equation (13).
Since these coefficients relate the output video signal _{C x,} the output video signal _C y, _{C y} likewise required, the output from the 3D-LUT is three times of the formula (13).

Hereinafter, the calculation method of each coefficient of Formula (12), ie, the production | generation method of 3D-LUT, is demonstrated. The equation (12) is an approximation of the accurate output video signal C _x , so the error g between the accurate output video signal C _x and the approximation is defined as the following equation (14) (accurate output video signal The same applies to C _y and C _z ). The equation (14) represents an error (error due to interpolation) g between the output signal interpolated by the equation (12) and the accurate output video signal C _x which is a true value.

Each coefficient is expressed by the following general formula (15) so that this error g is minimized. In addition, Min of Formula (15) is a minimum value. That is, equation (15) means that coefficients such as C _line included in equation (14) are determined such that the solution of the equation is minimized.

In this embodiment, the coefficient f ₁ related to the correction term at the vertex of the cube, the coefficient C _{line-x to} C _line-z related to the correction term on each side of the cube, and the correction term on each surface of the cube The related coefficients C _{plane-yz to} C _plane-xy and the coefficient C _vol regarding the correction term inside the cube are separately determined. For example, the coefficient C _line-x of the side parallel to the X axis and passing the point (m _ax , m _ay + i ₂ , m _az + i ₃ ) can be calculated from the following equation (16).

The coefficient C _plane-yz (m _ax + i ₁ , m _ay + Δy, m _az + Δz) in the YZ plane passing through the point (m _ax + i ₁ , m _ay , m _az ) is calculated from the following equation (17) it can.

The coefficient C _vol inside the cube can be calculated from the following equation (18).

[Configuration of signal converter]
The configuration of the signal conversion device 1 will be described with reference to FIG.
The signal conversion device 1 converts the signal format of an input video signal. As shown in FIG. 6, the signal conversion device 1 includes a 3D-LUT generation unit (table generation unit) 10, a 3D-LUT storage unit (table storage unit) 11, an input conversion unit (extraction unit) 12, and 3D. -LUT search means (table search means) 13 and signal operation processing means (interpolation means) 14 are provided.

The 3D-LUT generation unit 10 obtains a coefficient that minimizes the error according to a preset error calculation formula, and writes the obtained coefficient in the 3D-LUT storage unit 11 as a 3D-LUT.
In the present embodiment, in the 3D-LUT generation unit 10, equation (14) is preset as an error calculation equation, and an input video signal for 3D-LUT generation and an accurate output video signal are input. Then, the 3D-LUT generation unit 10 obtains coefficients f _{1 to} C _vol as a 3D-LUT from the input video signal for 3D-LUT generation and the accurate output video signal using this equation (14).
Thereafter, the 3D-LUT generation unit 10 writes the obtained 3D-LUT into the 3D-LUT storage unit 11.

  The 3D-LUT storage unit 11 is a storage device such as a memory that stores the 3D-LUT. The 3D-LUT is written by the 3D-LUT generation unit 10 and referred to by the 3D-LUT search unit 13 described later.

The input conversion means 12 receives an input video signal to be converted, obtains a value obtained by dividing the signal value of the input video signal input by the signal extraction interval, and extracts an integer part and a decimal part from the divided value It is.
Specifically, the input conversion means 12 divides the signal values of the input video signals V _ax , V _ay and V _az by the signal extraction interval M using the above equation (1), and the integer part m _ax , _Ask for m _ay and m _az . Further, the input conversion means 12 divides the signal values of the input video signals V _ax , V _ay and V _az by the signal extraction interval M using the above equation (2) to obtain residuals Δx, Δy and Δz Ask.
Thereafter, the input conversion means 12 outputs the obtained integer parts m _ax , m _ay , m _az to the 3D-LUT search means 13 and the signal operation processing means 14 and performs the signal operation processing on the obtained residuals Δx, Δy, Δz. Output to means 14.

The 3D-LUT search unit 13 searches the 3D-LUT of the 3D-LUT storage unit 11 and acquires a three-dimensional area corresponding to the cube of the integer part obtained by the input conversion unit 12.
Specifically, the 3D-LUT search unit 13 acquires, from the 3D-LUT, coefficients f _{1 to} C _vol representing three-dimensional regions corresponding to cubes of integer parts m _ax , m _ay , and m _az .
Thereafter, the 3D-LUT search unit 13 outputs the acquired coefficient to the signal operation processing unit 14.

The signal arithmetic processing means 14 generates an output signal of a signal value corresponding to the position indicated by the residual by performing interpolation with a preset high-order equation.
In the present embodiment, the signal arithmetic processing means 14 has equation (12) set in advance as a quadratic equation, and in this equation (12), integer parts m _ax , m _ay , m _az , residuals Δx, Δy, Δz The coefficients f _{1 to} C _vol are applied to obtain the signal value of the output video signal.
Thereafter, the signal arithmetic processing means 14 outputs the obtained output video signal to the outside.

[Operation of signal conversion device: 3D-LUT generation processing]
The procedure of the 3D-LUT generation process will be described with reference to FIG.
As shown in FIG. 7, the 3D-LUT generation unit 10 generates a 3D-LUT from the input video signal and the output video signal for 3D-LUT generation using an error calculation formula set in advance (step S1).
The 3D-LUT generation unit 10 writes the 3D-LUT obtained in step S1 in the 3D-LUT storage unit 11 (step S2).

[Operation of signal conversion device: signal conversion processing]
The procedure of signal conversion processing will be described with reference to FIG.
The input conversion means 12 obtains a value obtained by dividing the signal value of the input video signal by the signal extraction interval (step S10), and extracts an integer part and a residual from the divided value (step S11).
The 3D-LUT search unit 13 searches the 3D-LUT, and acquires a three-dimensional area corresponding to the cube of the integer part obtained in step S11 (step S12).
The signal arithmetic processing unit 14 interpolates with a preset quadratic equation to generate an output signal of a signal value corresponding to the position of the residual in the three-dimensional area acquired in step S12 (step S13).

[Operation / effect]
As described above, since the signal conversion device 1 performs the quadratic interpolation, the error due to the interpolation can be reduced compared to the conventional linear interpolation. Therefore, the signal conversion device 1 can generate a highly accurate output video signal while suppressing the data amount of the 3D-LUT. Furthermore, the signal conversion device 1 can minimize the increase in the amount of operation by quadratic interpolation, and can accelerate the signal conversion processing.

Second Embodiment
[Signal conversion processing]
The points of the signal conversion processing in the second embodiment which are different from the first embodiment will be described.
While the first embodiment performs quadratic interpolation, the second embodiment differs in performing cubic interpolation.

First, it defined as the following equation 3 equation g ₃ underlying (19) (residual [Delta] y, Delta] z as well). This equation (19) represents a function whose return value is 0 when the residual Δx is 0 or 1 as in the following equation (20).

In the second embodiment, the following equation (21) is used instead of the equation (10).
In Expression (21), in order to distinguish the coefficients C _line , C _{plane − xy} , and C _vol , numerical values of 1 to 8 are added to the end of these coefficients.

  Therefore, in the second embodiment, the following equation (22) is used instead of the equation (12), and the following equation (23) is used instead of the equation (14).

In the case of cubic interpolation, eight coefficients f ₁ related to the correction term at the vertex of the cube and coefficients C _{line-x-1 to} C _line-z-2 related to the correction term at each side of the cube Is 2 × 12 = 24, and the coefficients C _{plane-yz-1 to} C _plane-xy-4 related to the correction term in each surface of the cube are 4 × 6 = 24, related to the correction term in the cube The coefficients C _{vol-1 to} C _vol-8 are 8 × 1 = 8. Therefore, the output from the 3D-LUT is 27 coefficients as shown in the following equation (24).
Since these coefficients relate the output video signal _{C x,} the output video signal _C y, _{C y} likewise required, the output from the 3D-LUT is three times of the formula (24).

[Configuration of signal converter]
Returning to FIG. 6, the configuration of the signal conversion device 1B will be described about differences from the first embodiment.
As shown in FIG. 6, the signal conversion device 1B includes 3D-LUT generation means (table generation means) 10B, 3D-LUT storage means 11, input conversion means 12, 3D-LUT search means 13, and signal operation. And processing means (interpolation means) 14B.
In addition, except for the 3D-LUT generation unit 10B and the signal operation processing unit 14B, the description is omitted because it is the same as that of the first embodiment.
Further, the operation of the signal conversion device 1B is the same as that of the first embodiment, and thus the description thereof is omitted.

The 3D-LUT generation unit 10B calculates a coefficient that minimizes the error according to a preset error calculation formula, and writes the calculated coefficient in the 3D-LUT storage unit 11 as a 3D-LUT.
The 3D-LUT generation unit 10B is the same as that of the first embodiment except that the equation (23) is used as an error calculation equation.

The signal operation processing means 14B generates an output signal of a signal value corresponding to the position indicated by the residual by performing interpolation with a preset high-order equation.
The signal arithmetic processing unit 14B is the same as the first embodiment except that the cubic equation of the equation (22) is used as the high order equation.

  As described above, since the signal conversion device 1B performs cubic interpolation, errors due to interpolation can be reduced compared to conventional linear interpolation. Therefore, the signal conversion device 1B can generate a more accurate output video signal while suppressing the data amount of the 3D-LUT.

(Example)
Hereinafter, examples of the present invention will be described.
The errors of linear interpolation, quadratic interpolation and cubic interpolation were compared for the same input video signal.
Here, in order to increase the error, the input video signal is defined as JND (Just noticeable difference), color space as XYZ color space, and gamma as JND. Also, the output video signal has a format of RGB, a color space of BT. 709, gamma is the HDR PQ method.

  Also, the input video signal and the output video signal are 10 bits (corresponding to N = 1023), and the roughness of the input to the 3D-LUT is 4 bits, 5 bits, and 6 bits, respectively. Therefore, the signal extraction intervals M are 64, 32, 16 respectively. Then, for each of linear interpolation, quadratic interpolation, and cubic interpolation, the standard deviation with respect to the error of equation (14) was determined and compared.

  In FIG. 9, the abscissa represents the number of bits of the 3D-LUT, and the ordinate represents the standard deviation of the error. Here, the solid line represents linear interpolation, the broken line represents quadratic interpolation, and the dashed-dotted line represents cubic interpolation. From this FIG. 9, it can be seen that if the number of bits is the same, the errors decrease in the order of cubic interpolation, quadratic interpolation and linear interpolation, and high accuracy can be obtained. However, since the data amount of the 3D-LUT increases in the quadratic interpolation and the cubic interpolation, it is necessary to take that into account. Therefore, in FIG. 10, the amount of data is shown on the horizontal axis, and the standard deviation of the error is shown on the vertical axis. From FIG. 10, it was found that high accuracy can be obtained in the order of cubic interpolation, quadratic interpolation, and linear interpolation even with the same data amount. As described above, according to the present invention, conversion can be performed with higher accuracy if the amount of data is the same.

As mentioned above, although the embodiment of the present invention has been described in detail, the present invention is not limited to each of the above-described embodiments, and includes design changes and the like within the scope of the present invention.
In each of the above-described embodiments, the quadratic interpolation and the cubic interpolation have been described. However, in the present invention, the interpolation may be performed using a higher order equation than a quaternary equation.

1, 1B Signal conversion device 10, 10B 3D-LUT generation means (table generation means)
11 3D-LUT storage means (table storage means)
12 Input conversion means (extraction means)
13 3D-LUT Search Means (Table Search Means)
14, 14 B Signal operation processing means (interpolation means)

Claims (6)

A signal conversion apparatus for converting the signal format of an input signal, comprising:
A cube having at its vertex a grid point arranged at a predetermined signal extraction interval in a three-dimensional space representing signal values of the input signal, and a vertex having a vertex corresponding to the grid point in a three-dimensional space representing a signal value of the output signal Table storage means for storing in advance a three-dimensional look-up table in which the three-dimensional look-up table is associated with a three-dimensional area;
The input signal to be converted is input, and a value obtained by dividing the signal value of the input signal to be converted input is determined by the signal extraction interval, and an integer portion representing a grid point of the cube is calculated from the divided value. Extracting means for extracting a decimal part representing a position in the cube;
Table search means for searching the three-dimensional look-up table and obtaining a three-dimensional area corresponding to the cube of the integer part;
A high-order equation representing a position corresponding to the fractional part is set in advance in the three-dimensional area acquired by the table search means, and interpolation is performed using the high-order equation to obtain a signal corresponding to the position of the fractional part Interpolation means for generating said output signal of values;
A signal conversion device comprising:   The signal conversion apparatus according to claim 1, wherein the high order equation is a quadratic equation or a cubic equation.   The high-order equation corrects an error of a position represented by the decimal part in the three-dimensional area, a correction term in which positions of vertices and sides of the cube and the three-dimensional area are associated with each other The signal conversion device according to claim 1 or 2, wherein a correction term is included.   4. The signal conversion apparatus according to claim 3, wherein the table storage means stores, as the three-dimensional lookup table, coefficients of a correction term included in the high-order equation in advance.   An error calculation expression representing an error between the output signal interpolated by the high-order expression and the output signal of the true value is set in advance, and the coefficient which minimizes the error is calculated by the error calculation formula. 5. The signal conversion apparatus according to claim 4, further comprising table generation means for writing the table storage means as the three-dimensional lookup table.   The signal conversion apparatus according to any one of claims 1 to 5, wherein the high-order equation includes a function in which the return value is 0 when the fractional part is 0 or 1.

Images