JP2010166351A - Moving picture noise elimination device and moving picture noise elimination program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a moving picture noise elimination device that eliminates noise superimposed to a digital moving picture signal. <P>SOLUTION: The moving picture noise elimination device 1 includes: a block division means 3 to divide a digital moving picture signal into blocks; an orthogonal transforming means 5 to orthogonally transform each of blocks; a noise level estimating means 7 to estimate a noise level value indicating the size of noise elements superimposed to the block on the basis of the size of transformation factor of the predetermined high frequency element of the block; an in-block noise elimination means 9 to subtract the noise level value from transformation factors other than a.c. elements of the block; and an inverse orthogonal transforming means 11 to regenerate a digital moving picture signal by inversely transforming each of blocks. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a moving image noise removing apparatus and a moving image noise removing program for removing noise superimposed on a digital moving image signal.

In general, an analog moving image signal is superimposed with noise added through a transmission line in addition to the signal component of the original image itself. Conventionally, as a technique for removing such noise, there is a technique called a noise reducer, and various methods have been proposed so far (see, for example, Patent Document 1).
In order to operate this noise reducer, it is necessary to detect the level of the noise component in the screen (noise level) as a reference for removing noise. For example, a noise reducer (noise reduction circuit) described in Patent Document 1 detects a noise level from a signal in a blanking period of an analog moving image signal. Since the signal is flat (the voltage level is constant) during this blanking period, if even a slight signal component is superimposed during this period, it can be regarded as noise. Therefore, conventionally, noise removal processing corresponding to the amount of noise is performed by regarding the noise superimposed on the signal in the blanking period as the noise superimposed on the entire moving image signal.

  On the other hand, since the data of the moving image signal is a discrete value, the digital moving image signal is less likely to deteriorate than the analog moving image signal on the transmission path, and even when transmission noise is superimposed, Noise can be reduced.

  However, in recent years, noise (camera noise) superimposed on signals output from high-definition cameras has become a problem in digital moving image signals. Conventionally, the noise mixed on the transmission path has surpassed the camera noise, so that the noise has not been a problem. However, as in Super Hi-Vision, the reduction in the area of high-definition solid-state imaging devices (CCD, CMOS sensor, etc.) reduces the light energy obtained per unit photoreceptor, so noise is mixed into the output image. In particular, a large amount of noise is superimposed on the video shot especially at night.

Japanese Patent No. 3299026

As described above, a recent digital moving image signal has a problem in that camera noise is superimposed as the image becomes higher in definition.
This noise has fine sand particles and moves on the screen, so that the image quality is deteriorated. Further, the fact that the particles are fine means that they are high-frequency components, and this noise increases the amount of information during encoding. For this reason, when an image containing a lot of noise is encoded, transmission bits are occupied to transmit the noise, and there is a problem that the decoded image becomes a further deteriorated image than the original image.

  In addition, since a blanking period does not exist in a digital moving image signal like an analog moving image signal, the conventional noise reducer cannot detect noise and cannot properly remove noise. There is.

  The present invention has been made to solve the above-described problems, and is a moving image noise removal capable of removing superimposed noise even in a digital moving image signal having no blanking period. It is an object to provide an apparatus and a moving image noise removal program.

  The present invention was devised to achieve the above object. First, the moving image noise removing apparatus according to claim 1 is a moving image noise removing apparatus for removing noise superimposed on a digital moving image signal. The block division unit, the orthogonal transform unit, the noise level estimation unit, the intra-block noise removal unit, and the inverse orthogonal transform unit are provided.

  In such a configuration, the moving image noise removing apparatus divides the digital moving image signal into blocks having a predetermined size for each screen by the block dividing unit. And a moving image noise removal apparatus produces | generates a conversion coefficient by orthogonally transforming for every block by an orthogonal transformation means. The orthogonally transformed transform coefficient is a coefficient separated for each frequency component of the spatial frequency.

  Then, the moving image noise removal device estimates a noise level value indicating the magnitude of the noise component superimposed on the block based on the magnitude of the predetermined high-frequency component conversion coefficient by the noise level estimation means. . In the moving image signal, the signal component of the image itself is concentrated in a low spatial frequency region. Therefore, the noise level estimation means can estimate the noise level value from the spatial frequency region in which almost only the noise component is superimposed.

  The moving image noise removing apparatus subtracts the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means by the intra-block noise removing means, By calculating the maximum value with zero, a noise removal conversion coefficient, which is a conversion coefficient from which noise components in the block are removed, is generated within a non-negative range. In the moving image signal, the noise component is distributed almost evenly in all the spatial frequency regions except the DC component. Therefore, the intra-block noise removing means can generate a transform coefficient from which noise components are removed for each block by subtracting the noise level value from the transform coefficient.

  Then, the moving image noise removing apparatus performs inverse orthogonal transform on the noise removal transform coefficient generated by the intra-block noise removing unit by the inverse orthogonal transform unit. As a result, the moving image signal can be regenerated from the transform coefficient from which the noise component has been removed for each block.

  According to another aspect of the present invention, there is provided a moving image noise removal program that performs block division means, orthogonal transform means, noise level estimation means, and intra-block noise removal in order to remove noise superimposed on a digital moving image signal. Means and inverse orthogonal transform means.

  In such a configuration, the moving image noise removal program divides the digital moving image signal into blocks having a predetermined size for each screen by the block dividing means. Further, the moving image noise removal program generates a transform coefficient by performing orthogonal transform for each block by the orthogonal transform unit.

  Then, the moving image noise removal program estimates the noise level value indicating the magnitude of the noise component superimposed on the block based on the predetermined high-frequency component conversion coefficient by the noise level estimation means. .

Then, the moving image noise removing program subtracts the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means by the intra-block noise removing means, By calculating the maximum value with zero, a noise removal conversion coefficient which is a conversion coefficient from which the noise component in the block is removed is generated.
Then, the moving image noise removal program performs inverse orthogonal transform on the noise removal transform coefficient generated by the intra-block noise removal unit by the inverse orthogonal transform unit.

  The moving image noise removing apparatus according to claim 3 is a moving image noise removing apparatus that removes noise superimposed on a digital moving image signal, and includes a block dividing unit, an orthogonal transform unit, and a noise level estimating unit. And an intra-block noise removing unit and an inverse orthogonal transform unit.

  In such a configuration, the moving image noise removing apparatus divides the digital moving image signal into blocks having a predetermined size for each screen by the block dividing unit. And a moving image noise removal apparatus produces | generates a conversion coefficient by orthogonally transforming for every block by an orthogonal transformation means. The orthogonally transformed transform coefficient is a coefficient separated for each frequency component of the spatial frequency.

  Then, the moving image noise removing device averages the level values of the high frequency components of a predetermined number of transform coefficients for each screen among transform coefficients orthogonally transformed by the orthogonal transform means by the noise level estimating means. Thus, the noise level value indicating the magnitude of the noise component for each screen superimposed on the digital moving image signal is estimated.

The moving image noise removing apparatus subtracts the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means by the intra-block noise removing means, By calculating the maximum value with zero, a noise removal conversion coefficient which is a conversion coefficient from which the noise component in the block is removed is generated.
Then, the moving image noise removing apparatus performs inverse orthogonal transform on the noise removal transform coefficient generated by the intra-block noise removing unit by the inverse orthogonal transform unit. As a result, the moving image signal can be regenerated from the transform coefficient from which the noise component has been removed for each block.

  According to a fourth aspect of the present invention, there is provided the moving image noise removing apparatus according to the third aspect, wherein the noise level estimating means includes a high frequency component detecting means and a noise level determining means. It was.

  In such a configuration, the moving image noise removing device averages the conversion coefficient of the predetermined high-frequency component converted by the orthogonal transform unit by the high-frequency component detection unit in the block, and detects it as the level value of the high-frequency component of the block. To do.

  Then, the moving image noise removing device uses the noise level determination unit to calculate a low level value of the level value of the high frequency component detected by the high frequency component detection unit or an average value for a predetermined number from the low level value as the noise level value. judge. This makes it possible to estimate the noise level (noise level value) superimposed on the entire screen from a plurality of blocks in the screen.

  Furthermore, the moving image noise removing device according to claim 5 is the moving image noise removing device according to claim 4, wherein the noise level estimating means further includes a luminance detecting means.

In such a configuration, in the moving image noise removing apparatus, the noise level determination unit calculates a low level value of a high-frequency component level for each luminance detected by the luminance detection unit or an average value for a predetermined number of times from the low level value. The noise level value indicating the magnitude of each noise component is determined, and the noise removal means in the block determines from the conversion coefficient of the block for each luminance corresponding to the DC component of the block, for each luminance by the noise level determination means. A value obtained by subtracting the obtained noise level value and a maximum value of zero are calculated as a noise removal conversion coefficient.
Accordingly, the moving image noise removing device can detect the noise level value and remove noise for each luminance.

  According to another aspect of the present invention, there is provided a moving image noise removal program that performs block division means, orthogonal transform means, noise level estimation means, and intra-block noise removal in order to remove noise superimposed on a digital moving image signal. Means and inverse orthogonal transform means.

  In such a configuration, the moving image noise removal program divides the digital moving image signal into blocks having a predetermined size for each screen by the block dividing means. Then, the moving image noise removal program generates transform coefficients by performing orthogonal transform for each block by the orthogonal transform means. The orthogonally transformed transform coefficient is a coefficient separated for each frequency component of the spatial frequency.

  Then, the moving image noise removal program averages the level values of the high frequency components of a predetermined number of transform coefficients for each screen among the transform coefficients orthogonally transformed by the orthogonal transform means by the noise level estimation means. Thus, the noise level value indicating the magnitude of the noise component for each screen superimposed on the digital moving image signal is estimated.

Then, the moving image noise removing program subtracts the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means by the intra-block noise removing means, By calculating the maximum value with zero, a noise removal conversion coefficient which is a conversion coefficient from which the noise component in the block is removed is generated.
Then, the moving image noise removal program performs inverse orthogonal transform on the noise removal transform coefficient generated by the intra-block noise removal unit by the inverse orthogonal transform unit. As a result, the moving image signal can be regenerated from the transform coefficient from which the noise component has been removed for each block.

  The moving image noise removing apparatus according to claim 7 is a moving image noise removing apparatus that removes noise superimposed on a digital moving image signal, and includes a block dividing unit, an orthogonal transform unit, and a noise level estimating unit. And an intra-block noise removing unit and an inverse orthogonal transform unit.

  In such a configuration, the moving image noise removing apparatus divides the digital moving image signal into blocks having a predetermined size for each screen by the block dividing unit. And a moving image noise removal apparatus produces | generates a conversion coefficient by orthogonally transforming for every block by an orthogonal transformation means. The orthogonally transformed transform coefficient is a coefficient separated for each frequency component of the spatial frequency.

  Then, the moving image noise removing device displays a difference in pixel value between a horizontal or vertical line constituting the screen of the digital moving image signal and an adjacent line adjacent to the line in the spatial direction or the time direction by the noise level estimation unit. The noise level value indicating the magnitude of the noise component for each screen superimposed on the digital moving image signal is estimated. The horizontal line and the vertical line do not need to be a line having the maximum horizontal width or the maximum vertical height, and may be a partial line shorter than the maximum length.

Then, the moving image noise removing device subtracts the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means by the intra-block noise removing means, A noise removal conversion coefficient that is a conversion coefficient from which the noise component in the block is removed is generated.
Then, the moving image noise removing apparatus performs inverse orthogonal transform on the noise removal transform coefficient generated by the intra-block noise removing unit by the inverse orthogonal transform unit. As a result, the moving image signal can be regenerated from the transform coefficient from which the noise component has been removed for each block.

  The moving image noise removing device according to claim 8 is the moving image noise removing device according to claim 7, wherein the noise level estimating means includes a line extracting means, an inter-line noise level detecting means, and a noise level. And a determination unit.

  In such a configuration, the moving image noise removing apparatus sequentially extracts, from the digital moving image signal, a target line that is a line constituting the screen and an adjacent line adjacent to the target line as a correlation line group. Here, the adjacent line is one of the target lines on the screen (up or down for a horizontal line, right or left for a vertical line), or both. May be.

  Then, the moving image noise removing device averages the difference between the pixel values of the corresponding pixels between the target line and the adjacent line in the correlation line group extracted by the line extraction unit by the inter-line noise level detection unit, This is detected as the inter-line noise level in the correlation line group.

  Then, the moving image noise removing apparatus calculates an average value corresponding to the number of lines corresponding to a predetermined number of lines from the low level value of the inter-line noise level detected by the inter-line noise level detecting unit by the noise level determining unit. Is determined. As a result, the noise level (noise level value) superimposed on the entire screen can be estimated from a plurality of lines in the screen.

  Furthermore, the moving image noise removing device according to claim 9 is the moving image noise removing device according to claim 8, wherein the noise level estimating means further includes a luminance detecting means.

  In such a configuration, in the moving image noise removing device, the noise level determination unit calculates a low level value of the inter-line noise level for each luminance detected by the luminance detection unit or an average value for a predetermined number from the low level value. The noise level value indicating the magnitude of each noise component is determined, and the noise removal means in the block determines from the conversion coefficient of the block for each luminance corresponding to the DC component of the block, for each luminance by the noise level determination means. Subtracted noise level value. Accordingly, the moving image noise removing device can detect the noise level value and remove noise for each luminance.

  In addition, the moving image noise removing program according to claim 10, in order to remove the noise superimposed on the digital moving image signal, the computer is divided into block dividing means, orthogonal transform means, noise level estimating means, and noise removal within the block. Means and inverse orthogonal transform means.

  In such a configuration, the moving image noise removal program divides the digital moving image signal into blocks having a predetermined size for each screen by the block dividing means. Then, the moving image noise removal program generates transform coefficients by performing orthogonal transform for each block by the orthogonal transform means. The orthogonally transformed transform coefficient is a coefficient separated for each frequency component of the spatial frequency.

  Then, the moving image noise elimination program uses the noise level estimation means to average the pixel value difference between the line constituting the screen of the digital moving image signal and the adjacent line adjacent to the line, thereby digitally A noise level value indicating the magnitude of the noise component for each screen superimposed on the moving image signal is estimated.

Then, the moving image noise removing program subtracts the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means by the intra-block noise removing means, By calculating the maximum value with zero, a noise removal conversion coefficient which is a conversion coefficient from which the noise component in the block is removed is generated.
Then, the moving image noise removal program performs inverse orthogonal transform on the noise removal transform coefficient generated by the intra-block noise removal unit by the inverse orthogonal transform unit. As a result, the moving image signal can be regenerated from the transform coefficient from which the noise component has been removed for each block.

The present invention has the following excellent effects.
According to the first and second aspects of the invention, noise superimposed on the digital moving image signal can be detected based on the magnitude of the high-frequency component conversion coefficient obtained by orthogonally converting the digital moving image signal for each block. Even in the case of a digital moving image signal having no blanking period, noise can be detected and removed. In addition, according to the present invention, since the noise level is estimated from the high frequency component of the transform coefficient that has been orthogonally transformed, the camera noise of the high frequency component superimposed on the digital moving image signal can be reduced. Even in a high-definition video such as Super Hi-Vision having twice or 32 times the number of pixels, noise can be accurately removed.

  According to the third, fourth, and sixth aspects of the invention, in addition to the effect that noise can be removed even in the case of a digital moving image signal that does not have a blanking period, it is possible to remove the noise from the high frequency components in the screen. Since the noise level is estimated, the noise level superimposed on the screen can be detected with high accuracy. For this reason, the present invention can accurately remove noise from a digital moving image signal.

  According to the seventh, eighth, and tenth aspects of the present invention, in addition to the effect that noise can be removed even in the case of a digital moving image signal that does not have a blanking period, the screen is determined from the correlation between lines in the screen. Since the overall noise level is estimated, the noise level superimposed on the screen can be accurately detected. For this reason, the present invention can accurately remove noise from a digital moving image signal.

  According to the fifth and ninth aspects of the present invention, noise is superimposed on a digital moving image signal due to a decrease in light energy obtained per unit photoreceptor due to a reduction in area of a solid-state imaging device with high definition. For example, even when a lot of noise is superimposed on a dark part (a part with low luminance), the noise can be removed according to the luminance. For this reason, the present invention can accurately remove noise even in a high-definition image such as Super Hi-Vision having about 16 or 32 times the number of pixels of Hi-Vision.

It is a block block diagram which shows the structure of the moving image noise removal apparatus which concerns on 1st Embodiment of this invention. It is explanatory drawing for demonstrating the relationship between a moving image signal and a block. It is explanatory drawing for demonstrating the arrangement | sequence of the transform coefficient after orthogonal transformation. It is explanatory drawing for demonstrating the area | region of the frequency component used as the object in order to estimate a noise level. It is a figure which shows the other example of a pattern for demonstrating the area | region of the frequency component used as the object in order to estimate a noise level. It is a flowchart which shows operation | movement of the moving image noise removal apparatus which concerns on 1st Embodiment of this invention. It is a block block diagram which shows the structure of the moving image noise removal apparatus which concerns on 2nd Embodiment of this invention. It is a block block diagram which shows the structural example of the frame noise level estimation means of the moving image noise removal apparatus which concerns on 2nd Embodiment of this invention. It is typical explanatory drawing for demonstrating the method to detect the level of a high frequency component from the frequency component of a block. It is the graph which represented the cumulative value of the block for every level of a high frequency component as a histogram. It is a block block diagram which shows the structure of the moving image noise removal apparatus which concerns on 3rd Embodiment of this invention. It is a block block diagram which shows the structural example of the frame noise level estimation means of the moving image noise removal apparatus which concerns on 3rd Embodiment of this invention. It is explanatory drawing for demonstrating the extraction method of the correlation line group of 2 lines in a line extraction means. It is typical explanatory drawing for demonstrating the method of detecting the noise level between lines from the pixel value of an adjacent line. It is the graph which represented the cumulative value (frequency) of the number of lines for every noise level between lines as a histogram. It is explanatory drawing for demonstrating the extraction method of the correlation line group of 3 lines in a line extraction means. It is a block block diagram which shows the structure of the moving image noise removal apparatus which concerns on 4th Embodiment of this invention. It is a block block diagram which shows the structural example of the noise level estimation means according to the brightness | luminance of the moving image noise removal apparatus which concerns on 4th Embodiment of this invention. It is a block block diagram which shows the structure of the moving image noise removal apparatus which concerns on 5th Embodiment of this invention. It is a block block diagram which shows the structural example of the noise level estimation means according to brightness | luminance of the moving image noise removal apparatus which concerns on 5th Embodiment of this invention. It is a flowchart which shows operation | movement of the moving image noise removal apparatus which concerns on 2nd-5th embodiment of this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[First Embodiment: Configuration of Moving Image Noise Removal Device]
First, the configuration of the moving image noise removal device according to the first embodiment of the present invention will be described with reference to the block configuration diagram shown in FIG.

  The moving image noise removing apparatus 1 according to the first embodiment estimates a noise level in units of a block having a predetermined size from a digital moving image signal on which noise is superimposed (hereinafter simply referred to as a moving image signal). Is to be removed. Here, the moving image noise removing apparatus 1 includes a block dividing unit 3, an orthogonal transform unit 5, a block noise level estimating unit 7, an intra-block noise removing unit 9, and an inverse orthogonal transform unit 11. .

  The block dividing means 3 divides the moving image signal into blocks having a predetermined size for each frame constituting the screen. Although the size of this block is not particularly limited, for example, the block is divided into sizes of about 4 × 4 pixels to 128 × 128 pixels. The blocks divided by the block dividing unit 3 are output to the orthogonal transform unit 5.

That is, as shown in FIG. 2, the block dividing means 3 divides a frame F (t) at an arbitrary time position of a moving image signal into blocks of horizontal m pixels × vertical n lines (hereinafter referred to as m × n size). . Thus, the frame F (t) is divided into m × n size blocks B _{I and J.} Incidentally, subscript I, J of the block B _{I, J,} it is assumed that a horizontal I-th vertical J-th block.

  The orthogonal transform means 5 performs orthogonal transform (spatial frequency transform) on the block output from the block dividing means 3 for each block. This orthogonal transform means 5 transforms each block into a frequency component (spatial frequency) coefficient by performing orthogonal transform on each block. The frequency component coefficient (transform coefficient) is output to the block noise level estimating means 7 and the intra-block noise removing means 9.

  For example, the orthogonal transform unit 5 transforms a block into a frequency component by a discrete cosine transform (DCT) which is a kind of orthogonal transform. In addition, as long as this orthogonal transformation transforms the discrete signal of a block into a frequency domain, you may use transformations other than DCT transformation. For example, discrete sine transform (DST), fast Fourier transform (FFT), Hadamard transform, or the like can be used.

  As a result, the m × n size block is converted into a conversion coefficient (here, a DCT conversion coefficient) indicating the size of the frequency component equal to the number of pixels of the block, as shown in FIG. In the DCT conversion coefficient, the DCT conversion coefficient at the upper left in the figure represents a direct current component (DC), and the frequency increases toward the lower right and the lower left.

  The block noise level estimation means (noise level estimation means) 7 is superimposed on the block based on the size of the transform coefficient (DCT transform coefficient) of the predetermined high frequency component of the block transformed by the orthogonal transform means 5. The noise level value indicating the size of the noise component is estimated.

In general, when a frame (image) is converted into a spatial frequency, its signal component is concentrated in a low spatial frequency region (“A region” in FIG. 3). In general, since the noise component has whiteness and Gaussianity, the noise component is distributed almost evenly in all the spatial frequency regions (“B region” in FIG. 3) except the direct current component (DC).
That is, the DCT transform coefficient in any block _{BI, J} has a signal component (S: Signal) to noise component (N: Noise) ratio (S / N: Signal to Noise Ratio) as the spatial frequency is lower. The higher the spatial frequency, the lower the S / N ratio.

Therefore, as shown in FIG. 4, the block noise level estimation means 7 sets a predetermined region hm × hn region on the high frequency side in the spatial frequency among the DCT transform coefficients as the noise level estimation region R, and the noise By averaging the DCT transform coefficients in the level estimation region R, the noise level value is estimated. For example, the block noise level estimation means 7 calculates the root mean square (RMS) of the DCT transform coefficient in the noise level estimation region R, thereby calculating the noise level values B _{I, J_RSM} in the blocks B _{I, J.} presume.

Specifically, when the noise level estimation region R of hm × hn = 3 × 3 is predetermined among the 8 × 8 pixel blocks BI _{, J} shown in FIG. The noise level values B _{I, J_RSM} are estimated (calculated) from the DCT transform coefficients x _{1 to} x ₉ in the noise level estimation region R by the following equation (1).

In the equation (1), the coefficient 64/9 is a normalization coefficient, and is a coefficient for normalizing nine DCT transform coefficients to 64 coefficients of the entire block. By normalizing in this way, the noise level value of the high frequency component for the entire block can be obtained.
The block noise level estimation means 7 may obtain the noise level value by averaging the absolute values of the respective transform coefficients (absolute value average).
Further, the noise level estimation region R is not necessarily a predetermined rectangular region on the high frequency side at the spatial frequency. For example, as shown in FIG. 5, various patterns (patterns A to C) can be used. Returning to FIG. 1, the description of the configuration of the moving image noise removing apparatus 1 will be continued.

The intra-block noise removing unit 9 subtracts the noise level value estimated by the block noise level estimating unit 7 from the transform coefficient (DCT transform coefficient) other than the DC component of the block transformed by the orthogonal transform unit 5; The maximum value with zero is calculated. That is, the intra-block noise removing unit 9 calculates max (0, I−B _{I, J_RSM} ) where the value of the DCT transform coefficient is I and the noise level value is B _{I, J_RSM} . Thus, by subtracting the noise level value within a range that does not become negative, the noise component is removed in units of blocks. The reason why the noise level value is not subtracted from the DC component is that the DC component is a constant value in the block and represents the brightness (luminance). It is because it affects.
The transform coefficient from which noise has been removed (noise removal transform coefficient) and the transform coefficient of the DC component are output to the inverse orthogonal transform means 11.

  The inverse orthogonal transform unit 11 performs inverse orthogonal transform for each block of the noise removal transform coefficient generated by the intra-block noise removal unit 9 and the DC component transform coefficient. The inverse orthogonal transform unit 11 generates a moving image signal by performing inverse transform of the transform performed by the orthogonal transform unit 5. For example, when the orthogonal transform unit 5 performs discrete cosine transform (DCT transform), the inverse orthogonal transform unit 11 performs inverse discrete cosine transform (IDCT) for each block.

  As described above, the inverse orthogonal transform unit 11 performs the inverse orthogonal transform using the noise removal transform coefficient obtained by removing the noise component from the transform coefficient (DCT transform coefficient) generated by the orthogonal transform unit 5. A signal from which noise is removed from the signal (noise-removed video signal) can be generated.

  As described above, the moving image noise removal apparatus 1 orthogonally transforms a moving image signal for each block, and removes a noise component determined based on the magnitude of a high-frequency component conversion coefficient for each block. For this reason, the moving image noise removing apparatus 1 can detect a noise level and can remove noise even for a digital moving image signal having no blanking period.

  The moving image noise removing apparatus 1 can be operated by a moving image noise removing program that causes a general computer to function as each of the above-described means. The moving image noise elimination program can be distributed via a communication line, or can be recorded on a recording medium such as a CD-ROM for distribution.

[First Embodiment: Operation of Moving Image Noise Removal Device]
Next, the operation of the moving image noise removing apparatus according to the first embodiment of the present invention will be described with reference to the flowchart shown in FIG.

  First, the moving image noise removing apparatus 1 divides the input moving image signal into blocks having a predetermined size for each screen (frame) by the block dividing unit 3. Here, the block dividing means 3 divides the frame into horizontal I blocks and vertical J blocks (step S1). Each block is assumed to be composed of m horizontal pixels and n vertical pixels.

Thereafter, the moving image noise removing apparatus 1 sequentially orthogonally transforms the blocks divided in step S1 by the orthogonal transform unit 5. Here, i and j are variables for specifying a block, and the orthogonal transform means 5 first initializes the variables i and j (“0”) (step S2). Thereafter, the orthogonal transform means 5 performs orthogonal transform on the block B _{i, j} (step S3). As a result, the same number of transform coefficients as the pixels of the block B _{i, j} are generated.

Then, the moving image noise removing apparatus 1 uses the block noise level estimation means 7 to determine the noise component superimposed on the block based on the size of the predetermined high frequency component conversion coefficient of the block converted in step S3. A noise level value indicating the magnitude of the is estimated. Here, the block noise level estimation means 7 calculates the root mean square (RMS) for a predetermined high-frequency component region (noise level estimation region R: see FIG. 4) of the block B _{i, j} , thereby calculating the block B _{i. , J} are estimated as noise level values B _{i, j_RSM} (step S4).

  Then, the moving image noise removing apparatus 1 subtracts the noise level value estimated in step S4 from the conversion coefficient other than the DC component of the block converted in step S3 by the intra-block noise removing unit 9 in a range that does not become negative. By doing so, the noise component for each block is removed.

Here, x and y are variables for specifying a conversion coefficient for each block. Further, the conversion coefficient of x = 0 and y = 0 is a direct current component. Then, the intra-block noise removing unit 9 first initializes variables “x” and “y” (“0”) (step S5).
Then, the intra-block noise removing unit 9 determines whether or not both the variables x and y are “0”, that is, whether or not the conversion coefficient specified by the variable is a DC component (step S6). .

Here, when the conversion coefficient is a DC component (Yes in step S6), the intra-block noise removing unit 9 advances the operation to step S8. On the other hand, if the conversion coefficient is not a direct current component (No in step S6), the intra-block noise removal means 9 starts from the conversion coefficient B _{i, j} (x, y) of the block B _{i, j} specified by the variables x and y. Then, the maximum value of the value _obtained by subtracting the noise level value B _{i, j_RSM} of the noise of the block estimated in step S4 and zero is calculated (step S7).

Then, the intra-block noise removing means 9 sequentially increments (+1) the values of the variables x and y until the value of the variable x becomes “m” and the value of the variable y becomes “n”, and proceeds to step S6. Returning, the operation of removing the noise component from all the transform coefficients (excluding the DC component) in the block B _{i, j} is repeated (steps S8 to S11).

Then, the moving image noise removal apparatus 1 performs inverse orthogonal transformation on the noise removal transformation coefficient from which the noise component in the block B _{i, j} is removed by the inverse orthogonal transformation unit 11 and the transformation coefficient of the DC component, A conversion coefficient is generated as a moving image signal (step S12). The moving image signals for each block B _{i, j} are sequentially output to the outside.

  After that, the intra-block noise removing unit 9 sequentially increments (+1) the values of the variables i and j until the value of the variable i becomes “I” and the value of the variable j becomes “J”. Returning, the noise level estimation and the noise removing operation are repeated for all the blocks in the frame (steps S13 to S16).

  With the above operation, the moving image noise removing apparatus 1 removes noise in units of blocks in a frame. In addition, although the operation | movement which performs the noise removal for 1 frame was demonstrated in FIG. 6, every time the moving image signal for one frame is sequentially input as a moving image, the moving image noise removal apparatus 1 starts from step S1. Perform the action. Thereby, the moving image noise removing apparatus 1 can remove noise from the moving image signal.

  The configuration and operation of the moving image noise removal device 1 according to the first embodiment have been described above, but the present invention is not limited to this configuration. Here, in the moving image noise removing apparatus 1, the intra-block noise removing unit 9 subtracts the noise level value of each block estimated by the block noise level estimating unit 7 from the transform coefficient of each block, thereby moving the moving image signal. Removed the noise. However, the noise level is not necessarily estimated in units of blocks, and may be estimated in units of frames, for example. Thus, hereinafter, as a second and third embodiment, a moving image noise removing apparatus that performs noise removal by estimating a noise level in units of frames will be described.

[Second Embodiment: Moving Image Noise Removal Device]
The moving image noise removing apparatus 1A according to the second embodiment shown in FIG. 7 includes a block dividing unit 3, an orthogonal transform unit 5, a delay unit 6, a frame noise level estimating unit 7A, and an intra-block noise removing unit 9. And inverse orthogonal transform means 11. Since the configuration other than the delay unit 6 and the frame noise level estimation unit 7A is the same as the configuration of the moving image noise removal apparatus 1 described with reference to FIG. 1, the same reference numerals are given and description thereof is omitted.

  The delay means 6 delays the output of the transform coefficient of the block transformed by the orthogonal transform means 5 by the frame. For example, the delay means 6 is a memory capable of storing a conversion coefficient for one frame. The delay means 6 functions as a buffer until the estimation of the noise level of the frame is completed because the frame noise level estimation means 7A described later estimates the noise level in units of frames.

The frame noise level estimation means (noise level estimation means) 7A estimates a noise level value indicating the magnitude of the noise component superimposed on the moving image signal in units of frames. This frame noise level estimation means 7A searches for a flat portion with little change in pixel value for each predetermined block in a frame (screen) constituting a moving image signal, and a signal superimposed on that portion. The component amount is estimated as a noise level value.
Here, the configuration of the frame noise level estimation means 7A will be described with reference to the block configuration diagram shown in FIG.

  Normally, a moving image screen has a flat portion with little change in pixel value, for example, an infinite region such as the sky, a region where the camera is not in focus, or the like. These areas are areas where the frequency components of the image have few high frequency components. On the other hand, noise is a high-frequency component with fine particles. That is, in the screen, the high frequency component in the region with a low high frequency component can be regarded as being superimposed by noise.

  Therefore, the frame noise level estimation means 7A estimates the noise level value from the high frequency component in the region where the high frequency component is small in the screen. Here, the frame noise level estimating means 7A includes a high frequency component detecting means 21, a high frequency component accumulating means 23, and a noise level determining means 25.

  The high frequency component detection means 21 averages the conversion coefficient of the predetermined high frequency component converted by the orthogonal transform means 5 in the block, and detects it as the level of the high frequency component of the block.

  In the orthogonally transformed frequency component, the low frequency component including the direct current component is a component including a large amount of the frequency component (signal component) of the image signal as described with reference to FIG. Therefore, here, the high-frequency component detection means 21 sets the coefficient of the low-frequency component to “0” and averages the coefficient of the high-frequency component in the other region within the block, thereby obtaining the level of the high-frequency component of the block. Will be considered. This averaging can be calculated by the root mean square (RMS) of each high frequency component coefficient, the absolute value average, or the like.

  Here, with reference to FIG. 9 (refer to FIG. 8 as appropriate), a method in which the high frequency component detection means 21 detects the level of the high frequency component from the frequency component of the block will be described. FIG. 9A shows a frequency component (transform coefficient) of a certain block obtained by the orthogonal transform means 5, and FIG. 9B shows a state in which only a high frequency component is extracted from the frequency component. Here, in order to simplify the description, the block size is assumed to be 4 × 4 pixels.

As shown in FIG. 9A, it is assumed that a certain block is orthogonally transformed by the orthogonal transformation means 5 and the same number of coefficients as the number of pixels of the block (here, 4 × 4) are obtained.
At this time, the high frequency component detecting means 21 regards the DC component and the coefficients of the horizontal frequency component and the vertical frequency component including the DC component as “0” as coefficients near the DC component (FIG. 9B). )), And extract other coefficients as high-frequency component coefficients. That is, the coefficients x ₁₁ , x ₁₂ , x ₁₃ , x ₁₄ , x ₂₁ , x ₃₁ , x ₄₁ are regarded as “0”, and the other coefficients x ₂₂ , x ₂₃ , x ₂₄ , x ₃₂ , x ₃₃ , x _{34 are considered.} , X ₄₂ , x ₄₃ , x ₄₄ are extracted as high frequency component coefficients.

  The high frequency component detection means 21 averages each coefficient extracted as the high frequency component coefficient within the block, thereby obtaining the level of the high frequency component of the block. Here, when averaging is performed using the root mean square, the high frequency component detecting means 21 calculates the level L of the high frequency component by the following equation (2).

In this equation (2), the coefficient 16/9 is a normalization coefficient, and is a coefficient for normalizing nine high-frequency component coefficients to coefficients for the entire block. By normalizing in this way, the level of the high frequency component for the entire block can be obtained. The high-frequency component detection means 21 may calculate the level L of the high-frequency component based on the average (absolute value average) of the absolute values of the coefficients extracted as the high-frequency component coefficients.
Further, the high frequency component detection means 21 does not need to set the coefficient of the high frequency component to a rectangular region as shown in FIG. 9B, for example, various patterns (patterns A to C) as shown in FIG. Can be used. Returning to FIG. 8, the description of the configuration of the frame noise level estimation means 7A will be continued.

  The high frequency component accumulating unit 23 accumulates the number of blocks corresponding to the level of the high frequency component detected by the high frequency component detecting unit 21 for each level. The high-frequency component accumulating unit 23 accumulates (histograms) the number of blocks corresponding to the level of the high-frequency component for each frame (see FIG. 10). As a result, it is analyzed which level of which high-frequency component is distributed in the screen. The accumulated value for each level accumulated by the high frequency component accumulating means 23 is output to the noise level determining means 25.

The noise level determination means 25 determines a low level value of the high frequency component level accumulated by the high frequency component accumulation means 23 or an average value corresponding to a predetermined number of blocks from the low level value as a noise level value. It is. As described with reference to FIG. 9, the coefficients detected by the high-frequency component detection means 21 as the high-frequency components (for example, the coefficients x ₂₂ , x ₂₃ , x ₂₄ , x ₃₂ , x ₃₃ , x ₃₄ , x ₄₂ , x ₄₃ , x ₄₄ ) is almost “0” if it is a flat part with little change in the pixel value of the screen, so if noise is superimposed on the flat part, It can be said that the high frequency component is noise. Therefore, the noise level determination means 25 is a low level value of the level of the high frequency component or a predetermined number (for example, 2 to 5) from the low level value in the total of blocks having the high frequency component accumulated by the high frequency component accumulation means 23. The average value (weighted average value) according to the number of blocks is determined as the noise level value.

  Here, the minimum value of the high frequency component level accumulated by the high frequency component accumulating means 23 is used as the low level value of the high frequency component level. Note that the noise level of an image captured by a CCD or the like may differ depending on the brightness of the screen. When the minimum value of the high-frequency component level is used, the noise level may be determined at a low level. Therefore, the low level value is not necessarily limited to the minimum value, and a value larger than the minimum value within a predetermined range from the minimum value may be used.

  Here, the noise level determination method of the noise level determination means 25 will be described with reference to the graph shown in FIG. Note that the graph of FIG. 10 is a diagram for visually explaining the accumulated value of the block, and does not mean that the high-frequency component accumulating unit 23 generates the graph.

Usually, the high-frequency component includes the high-frequency component of the image signal itself and noise at the same time. At this time, in the graph of FIG. 10, the right histogram from the center of the graph having a large high-frequency component is considered to be the high-frequency component of the image (including noise). On the other hand, several histograms at the left end of the graph are considered to be almost only noise components that do not include high-frequency components of the image.
Therefore, the noise level determination means 25 calculates an average value from cumulative values corresponding to several histograms on the left end of the graph and determines it as a noise level value.

For example, as shown in FIG. 10, when determining the noise level from three histograms, the noise level determination means 25 selects L ₁ , L ₂ , L ₃ with low levels of high frequency components, and the respective frequencies. An average value (weighted average value) is obtained from C ₁ , C ₂ , and C ₃ by the following equation (3), and set as a noise level value N _L.

Here, the noise level determination means 25 calculates the noise level value by the weighted average of the levels of a plurality of high frequency components, but the noise level value is determined by the arithmetic average of the levels of the high frequency components without considering the frequency. May be. Further, the noise level determination means 25 may simply set the minimum high-frequency component level (L _{1 in} the case of FIG. 10) as the noise level value.
Then, the noise level determination unit 25 outputs the determined noise level value to the intra-block noise removal unit 9 (FIG. 7).

The configuration of the frame noise level estimation means 7A has been described above, but the present invention is not limited to this configuration.
Here, the frame noise level estimation means 7A estimates the noise level value from the transform coefficients of all the blocks generated by the orthogonal transform means 5, but the block to be used is 50% or more of the entire screen (more (70% or more is preferable).

  Although the example in which the frame noise level estimation means 7A detects the noise level of the moving image signal in units of frames (spatial direction) has been described here, it is detected for each of a plurality of frames (spatial + time direction). It is good. For example, the frame noise level estimating means 7A accumulates the blocks corresponding to the high-frequency component levels for three frames in the high-frequency component accumulating means 23 together with the preceding and succeeding frames, and the noise level determining means 25 based on the accumulated result. The noise level of the intermediate frame may be determined.

  Also, here, the frame noise level estimation means 7A uses the low level value of the level or the low level value in the block histogram for each level of the high frequency component accumulated by the high frequency component accumulation means 23 by the noise level determination means 25. The average value (weighted average, etc.) corresponding to the predetermined number of blocks is determined as the noise level. However, the noise level determination means 25 is simply a low level value (minimum value) or a predetermined number of times from the low level value among the high frequency component levels detected by the high frequency component detection means 21 regardless of the number of blocks. The average value of the level values may be determined as the noise level value. In this case, the frame noise level estimation means 7A can be configured by omitting the high frequency component accumulating means 23 from the configuration.

[Third Embodiment: Moving Image Noise Removal Device]
The moving image noise removing apparatus 1B according to the third embodiment shown in FIG. 11 includes a block dividing unit 3, an orthogonal transform unit 5, a delay unit 6, a frame noise level estimating unit 7B, and an intra-block noise removing unit 9. And inverse orthogonal transform means 11. Since the configuration other than the frame noise level estimation means 7B is the same as the configuration of the moving image noise removing apparatus 1A described with reference to FIG. 7, the same reference numerals are given and description thereof is omitted.

The frame noise level estimation means (noise level estimation means) 7B estimates the noise level value superimposed in the moving image signal based on the correlation of the lines in the frame (screen) constituting the moving image signal.
Here, the configuration of the frame noise level estimation means 7B will be described with reference to the block configuration diagram shown in FIG.

  Normally, when a difference between adjacent line pixel values in a frame screen of a general moving image is taken, the signal component of the frame image itself is reduced, but the signal component approaches an uncorrelated noise component. Thereby, the smaller the difference between the pixel values, the more the difference value can be estimated as the pixel value generated by the uncorrelated noise.

  Therefore, the frame noise level estimation means 7B estimates the noise level value from the difference between the pixel values of adjacent lines in the screen. Here, the frame noise level estimation means 7B includes a line extraction means 31, an inter-line noise level detection means 33, a level accumulation means 35, and a noise level determination means 37.

  The line extraction unit 31 correlates a target line, which is a line constituting a frame (screen) from a moving image signal, and an adjacent line adjacent to the target line in the spatial direction along the scanning line. Are extracted as pixels for each correlation line group. Here, it is assumed that the line extraction unit 31 partially extracts lines so that the ratio is equal to or higher than a predetermined ratio with respect to the entire screen in the screen. The pixel value of each pixel of the line extracted by the line extraction unit 31 is output to the inter-line noise level detection unit 33.

Here, with reference to FIG. 13 (refer to FIG. 12 as appropriate), the lines (correlation line group) extracted by the line extraction means 31 will be described.
As shown in FIG. 13 (a), the line extraction means 31 has an Nth line (target line) and an (N + 1) th line (adjacent line) in the main scanning unit of the screen D (two-dimensional image). Two adjacent lines are sequentially extracted as a correlation line group. At this time, the line extraction means 31 provides a gap for each adjacent correlation line group, and extracts a plurality of correlation line groups so as to cover an area of 50% or more (more preferably 70% or more) of the entire screen D. . As a result, the line extraction unit 31 can extract a correlation line group in the screen D with at least a small change in pixel value between lines.

13A shows an example in which the line extraction unit 31 extracts a line corresponding to the horizontal width of the screen as a line. However, as shown in FIG. It is good also as extracting a part of short line partially.
Further, as shown in FIG. 13C, the line extraction unit 31 includes a line at the same horizontal position on the screen (frame) continuous in the time direction, the N _tth line of the screen at time t, The _{Nt +} 1th line on the screen at time (t + 1) may be extracted as a correlation line group. In this case as well, as in FIG. 13B, it is not always necessary to use the width of the screen in the horizontal direction, and the line may be partially extracted.

In addition, here, an example is shown in which lines are extracted in the horizontal direction along the scanning lines, but lines may be extracted in the vertical direction.
In this example, the line extraction unit 31 partially extracts the correlation line group from the screen D so as to be equal to or higher than the predetermined ratio. However, the line extraction unit 31 extracts the correlation line group of the entire screen D. Also good. Returning to FIG. 12, the description of the configuration of the frame noise level estimation means 7B will be continued.

  The inter-line noise level detection means 33 averages the pixel value differences of the corresponding pixels (here, the pixels at the same horizontal position) for each adjacent correlation line group in the line extracted by the line extraction means 31. The noise level is detected as the inter-line noise level in the correlation line group of the target line. This averaging can be calculated by the root mean square of the difference between pixel values in each correlation line group, the absolute value average, or the like.

  Here, with reference to FIG. 14 (refer to FIG. 12 as appropriate), a method in which the inter-line noise level detection unit 33 detects the inter-line noise level from the pixel values of adjacent lines will be described.

As illustrated in FIG. 14, it is assumed that the pixel of the Nth line and the pixel of the (N + 1) th line adjacent to the Nth line are extracted by the line extraction unit 31. Each line has M pixels in the horizontal direction. Here, the pixel values of the pixels of the Nth line are p _{N, 1} , p _{N, 2} , p _{N, 3} ,..., P _{N, M,} and the pixel values of the pixels of the (N + 1) th line are pN _{+ 1,1} , pN _{+ 1,2} , pN _{+ 1,3} ,..., pN _{+ 1, M.}

  The inter-line noise level detection means 33 averages the pixel value difference of each pixel at the same horizontal position between the N-th line and the (N + 1) -th line, so that the inter-line noise level is detected. Use noise level. Here, when averaging is performed using the root mean square, the inter-line noise level detection means 33 calculates the inter-line noise level L according to the following equation (4).

  Alternatively, when averaging is performed using absolute value averaging, the inter-line noise level detection means 33 calculates the inter-line noise level L according to the following equation (5).

Here, the method for detecting the inter-line noise level on the same screen (frame) has been described. However, as shown in FIG. 13C, the inter-line noise level may be detected between adjacent frames. Good. In this case, in the above-described equations (4) and (5), the pixel values p _{N, i} , p _{N + 1, i} only become the pixel values p _{Nt, i} , p _{Nt + 1, i} corresponding to the time t, respectively. Similarly, the noise level between lines can be calculated.
Returning to FIG. 12, the description of the configuration of the noise level detection apparatus will be continued.

  The level accumulating unit 35 accumulates the number of lines corresponding to the inter-line noise level detected by the inter-line noise level detecting unit 33 for each level. The level accumulating unit 35 accumulates (histograms) the number of lines corresponding to the inter-line noise level for each level (see FIG. 15). Thus, it is analyzed how much noise level between lines is distributed in the screen. The accumulated value for each level accumulated by the level accumulating unit 35 is output to the noise level determining unit 37.

  The noise level determination means 37 determines a low level value of the inter-line noise level accumulated by the level accumulation means 35 or an average value corresponding to a predetermined number of lines from the low level value as a noise level. . The inter-line noise level detected by the inter-line noise level detection means 33 has a smaller image component between adjacent lines, so that the smaller the value, the closer to the noise level. Therefore, the noise level determination means 37 is a predetermined number (for example, 2 to 5) from the minimum value or minimum value of the inter-line noise level in the total number of lines for each inter-line noise level accumulated by the level accumulation means 35. The average value (weighted average value) corresponding to the number of lines is determined as the noise level.

  Here, the minimum value of the inter-line noise level accumulated by the level accumulating means 35 is used as the low level value of the inter-line noise level. In addition, since the minimum value of the noise level between lines may be detected by chance regardless of noise depending on the image, the low level value is not necessarily limited to the minimum value. A value larger than the minimum value may be used within a predetermined range.

  Here, the noise level determination method of the noise level determination means 37 will be described with reference to the graph shown in FIG. The graph of FIG. 15 is a diagram for visually explaining the cumulative value of the number of lines, and does not mean that the level cumulative means 35 generates the graph.

The pixel value difference between the lines includes the image component difference and the noise component at the same time. At this time, in the graph of FIG. 15, it is considered that the histogram from the center to the right of the graph where the inter-line noise level is large includes a lot of signal component differences of the moving image (including noise) itself. On the other hand, several histograms at the left end of the graph are considered to be substantially only noise components that do not include differences in signal components of moving images.
Therefore, the noise level determination means 37 calculates an average value from the cumulative values corresponding to several histograms on the left end of the graph and determines it as a noise level. In order to obtain the noise level from this histogram, as described with reference to FIG. 10, for example, an average value (weighted average value) can be obtained by the equation (3) to obtain the noise level _NL .

Here, the noise level determination unit 37 calculates the noise level by weighted average of a plurality of inter-line noise levels. However, the noise level may be determined by arithmetic average of the inter-line noise levels without considering the frequency. Good. Further, the noise level determination means 37 may simply set the minimum inter-line noise level (L _{1 in} the case of FIG. 15) as the noise level.
Then, the noise level determination unit 37 outputs the determined noise level value to the intra-block noise removal unit 9 (FIG. 11).

The configuration of the frame noise level estimation unit 7B has been described above, but the present invention is not limited to this configuration.
Here, the frame noise level estimation means 7B sequentially extracts a total of two lines of the target line and one adjacent line as a correlation line group by the line extraction means 31, and the two lines by the noise level detection means 33 between lines. The inter-line noise level is detected from the pixel value of. However, the correlation line group is not limited to two lines. For example, the inter-line noise level may be detected by a total of three lines including the target line and two adjacent lines.

  Here, with reference to FIG. 16 (refer to FIG. 12 as appropriate), a method for detecting the noise level between three lines will be described.

  As shown in FIG. 16A, for example, the line extraction means 31 is the Nth line (target line), (N−1) and (N + 1) in the main scanning unit of the screen D (two-dimensional image). Three adjacent lines of the first line (adjacent line) are sequentially extracted as a correlation line group. At this time, the line extraction means 31 provides a gap for each adjacent correlation line group, and extracts a plurality of correlation line groups so as to cover an area of 50% or more (more preferably 70% or more) of the entire screen D. .

Further, as shown in FIG. 16B, a part of lines shorter than the horizontal width of the screen may be partially extracted as three lines.
Further, as shown in FIG. 16C, the line extracting unit 31 is the line at the same horizontal position on the screen (frame) continuous in the time direction, and is the N _tth line (target) at the time t. Line) and N _t-1 and N _{t + 1} second lines (adjacent lines) on the screen at times (t−1) and (t + 1) may be extracted as a correlation line group. Also in this case, as in FIG. 16B, the line need not necessarily use the horizontal width of the screen, and may be partially extracted.

  As described above, when the line extraction unit 31 sequentially extracts three correlation line groups, the frame noise level estimation unit 7B uses the inter-line noise level detection unit 33 to calculate the target line for each corresponding pixel of the correlation line group. Find the difference between the pixel value of the sum of the pixel values of the line adjacent to one (first adjacent line) and the line adjacent to the other (second adjacent line) and the pixel value of the target line. A value obtained by averaging the differences is defined as an inter-line noise level.

That is, the inter-line noise level detection means 33 multiplies the pixel values of the corresponding pixels of the three lines by (−1/2, 1, −1/2) or (1/2, −1, 1/2). Then, the difference between the target line and the adjacent line is obtained. Then, the inter-line noise level detection means 33 calculates the inter-line noise level in the correlation line group from the difference by the above-described equation (4) or (5).
In this way, by detecting the noise level with three lines, the image components are dispersed and the noise level can be detected with higher accuracy than when the noise level is detected with two lines.

  Also, here, the frame noise level estimation means 7B uses the low level value of the level or the low level value in the histogram of the number of lines for each inter-line noise level accumulated by the level accumulation means 35 by the noise level determination means 37. Therefore, the average value (weighted average or the like) corresponding to the predetermined number of lines is determined as the noise level. However, the noise level determination means 37 is simply a low level value (minimum value) or a predetermined number from the low level value among the noise levels between lines detected by the noise level detection means 33 between lines regardless of the number of lines. The average value of the minute level values may be determined as the noise level value. In this case, the frame noise level estimation means 7B can be configured by omitting the level accumulating means 35 from the configuration.

[Fourth Embodiment: Moving Image Noise Removal Device]
Next, the configuration of the moving image noise removing device according to the fourth embodiment of the present invention will be described with reference to the block configuration diagram shown in FIG.

  In general, the appearance frequency of noise varies depending on the brightness of the screen. This is due to the nonlinearity of the imaging device that captures moving images. Even if the same level of noise is mixed in the video, the bright and dark portions of the screen look different. This is due to the non-linearity of human vision. That is, when performing noise removal, it is desirable to detect the noise level adaptively according to the brightness (luminance) of the pixel.

  Therefore, in the moving image noise removing device 1C according to the fourth embodiment, the moving image noise removing device 1A (see FIG. 7) detects the noise level for each screen of the moving image signal. A function for detecting the noise level every time and changing the amount of noise to be removed for each luminance is added.

  Here, the moving image noise removing apparatus 1C includes a block dividing unit 3, an orthogonal transforming unit 5, a delay unit 6, a luminance-specific noise level estimating unit 7C, a luminance-specific intra-block noise removing unit 9A, and an inverse orthogonal transform. Means 11. Since the configuration other than the luminance level-specific noise level estimation means 7C and the luminance-specific intra-block noise removal means 9A is the same as the configuration of the moving image noise removal apparatus 1A described with reference to FIG. Description is omitted.

  Similarly to the frame noise level estimation means 7A described with reference to FIG. 8, the luminance-specific noise level estimation means (noise level estimation means) 7C at least for each predetermined block in the frame (screen) constituting the moving image signal. A flat portion with a small change in pixel value is searched, and the amount of signal components superimposed on that portion is estimated as a noise level value. However, the luminance level-specific noise level estimation unit 7C is different from the frame noise level estimation unit 7A in that the noise level value is estimated for each luminance level in the frame.

Here, with reference to the block diagram shown in FIG. 18, the configuration of the luminance-specific noise level estimation means 7C will be described.
The luminance level-specific noise level estimation means 7C in FIG. 18 includes a high-frequency component detection means 21, a luminance detection means 22, a luminance-specific high-frequency component accumulation means (high-frequency component accumulation means) 23A, and a luminance-specific noise level determination means (noise level determination). Means) 25A. The high-frequency component detection means 21 is the same as the frame noise level estimation means 7A described with reference to FIG.

  The luminance detection unit 22 detects the luminance from the conversion coefficient converted by the orthogonal transformation unit 5. Here, the luminance detection unit 22 extracts only a direct current component from the frequency components orthogonally transformed for each block by the orthogonal transformation unit 5 and obtains a luminance value. As another method for detecting the luminance of the block, the luminance detecting unit 22 obtains an average of pixel values of several pixels for each block before orthogonal transformation, that is, for each block output from the block dividing unit 3. Thus, the luminance of the block may be detected.

  The high-frequency component accumulation means (high-frequency component accumulation means) 23A for each luminance accumulates the number of blocks for each level of the high-frequency components detected by the high-frequency component detection means 21 for each luminance of the blocks detected by the luminance detection means 22. It is. The high frequency component accumulating unit 23A by luminance has the same function as the high frequency component accumulating unit 23 described with reference to FIG. 8 except that the number of blocks for each level of the high frequency component is accumulated by luminance. That is, the luminance-specific high-frequency component accumulating unit 23A generates the histogram described in FIG. 10 for each luminance.

  The luminance level-specific noise level determination unit (noise level determination unit) 25A sets the low-level value of the high-frequency component level for each luminance accumulated by the luminance-specific high-frequency component totaling unit 23A or a predetermined number of blocks from the low level value. The corresponding average value is determined as a noise level value corresponding to luminance (noise level value by luminance). The brightness-specific noise level determination means 25A has the same function as the noise level determination means 25 described with reference to FIG. 8 except that the noise level value is determined for each brightness. That is, the luminance-specific noise level determination means 25A calculates a low-level value in the high-frequency component level or an average value corresponding to a predetermined number of blocks from the low-level value from the histogram for each luminance shown in FIG. The average value is obtained as the noise level for each luminance.

Note that the luminance-specific noise level determination means 25A can obtain the luminance noise level value that has not been accumulated in the frame by interpolating from the already obtained luminance noise level value by interpolation or extrapolation. it can.
Then, the luminance level-specific noise level determination unit 25A outputs the determined noise level value to the luminance-specific block noise removal unit 9A (FIG. 17). Returning to FIG. 17, the description of the configuration of the moving image noise removing apparatus 1C will be continued.

The intra-brightness noise removal unit 9A performs transform coefficients other than the DC component of the block obtained by orthogonally transforming the block-divided moving image signal in the orthogonal transform unit 5, delaying in the delay unit 6 in units of frames, and performing orthogonal transform. The maximum value of zero and a value obtained by subtracting the noise level value estimated by the luminance level-specific noise level estimation means 7C for each luminance from the DCT conversion coefficient) is calculated.
That is, for each block, the intra-brightness noise removal unit 9A determines the noise level value corresponding to the same luminance as the luminance estimated by the luminance-specific noise level estimation unit 7C using the direct current component of the block as the luminance. Subtract from the conversion coefficient in a range that does not become negative. Thereby, the moving image noise removing apparatus 1C can remove noise according to the brightness of the screen by adaptively detecting the noise level superimposed on the moving image signal according to the brightness of the screen. It becomes possible.

  The configuration of the luminance level-specific noise level estimation means 7C has been described above, but the present invention is not limited to this configuration. Here, the noise level estimation means 7C for each luminance accumulates the number of blocks for each level of the high frequency component for all luminances, and estimates the noise level value for each luminance. However, the luminance level-specific noise level estimation means 7C classifies the luminance in advance, performs the total of the high-frequency component levels using one luminance in the segment as the representative luminance, and from the accumulated value of the representative luminance of each category, The noise level value may be estimated for each luminance by obtaining a cumulative value of other luminance by interpolation or the like. Alternatively, the noise level value of the representative luminance may be set as the noise level value of all luminances in the section.

[Fifth Embodiment: Moving Image Noise Removal Device]
Next, with reference to the block diagram shown in FIG. 19, the structure of the moving image noise removal device according to the fifth embodiment of the present invention will be described.

  The moving image noise removing apparatus 1D according to the fifth embodiment includes a block dividing unit 3, an orthogonal transform unit 5, a delay unit 6, a luminance-specific noise level estimating unit 7D, a luminance-specific intra-block noise removing unit 9A, And an inverse orthogonal transform means 11. Since the configuration other than the luminance level-specific noise level estimation means 7D is the same as that of the moving image noise removing device 1B described in FIG. 11 and the moving image noise removing device 1C described in FIG. The description is omitted.

  Similarly to the frame noise level estimation means 7B described with reference to FIG. 11, the luminance-specific noise level estimation means (noise level estimation means) 7D is based on the correlation of the lines in the frame (screen) constituting the moving image signal. The noise level value superimposed inside is estimated. However, the brightness level noise level estimation means 7D differs from the frame noise level estimation means 7B in that the noise level value is estimated for each brightness level in the frame.

Here, the configuration of the noise level estimation means 7D for each luminance will be described with reference to the block configuration diagram shown in FIG.
20 includes a line extraction unit 31, a luminance detection unit 32, an inter-line noise level detection unit 33, a luminance-based level accumulation unit (level accumulation unit) 35A, and a luminance-specific noise level. Determination means (noise level determination means) 37A. Since the configuration other than the luminance detection unit 32, the luminance level totalization unit 35A, and the luminance level noise level determination unit 37A is the same as that of the frame noise level estimation unit 7B described in FIG. Is omitted.

  The luminance detection unit 32 detects the luminance of the line extracted by the line extraction unit 31. Here, the luminance detection unit 32 detects the luminance of the line by obtaining the average of the pixel values for each line output from the line extraction unit 31. The luminance detecting means 32 may obtain the luminance by averaging all pixel values of the line, or select a predetermined number of pixels (for example, 50% of the entire line) according to the number of pixels of the line. Then, the average of the pixel values of the selected pixels may be set as the luminance.

  The level-by-brightness accumulation means (level accumulation means) 35A accumulates the number of lines for each inter-line noise level detected by the inter-line noise level detection means 33 for each line luminance detected by the luminance detection means 32. It is. The level-by-brightness accumulation unit 35A has the same function as the level accumulation unit 35 described with reference to FIG. 12 except that the number of lines for each inter-line noise level is accumulated for each luminance. That is, the luminance-based level accumulating unit 35A generates the histogram described in FIG. 15 for each luminance.

  The luminance level-specific noise level determination unit (noise level determination unit) 37A corresponds to a low level value of the inter-line noise level for each luminance accumulated by the luminance level accumulation unit 35A or a predetermined number of lines from the low level value. The average value is determined as a noise level corresponding to luminance (noise level for each luminance). The brightness-specific noise level determination unit 37A has the same function as the noise level determination unit 37 described with reference to FIG. 12 except that the noise level is determined for each brightness. That is, the noise level determination unit 37A for each luminance obtains a low level value of the noise level between lines or an average value corresponding to a predetermined number of lines from the low level value from the histogram for each luminance shown in FIG. The noise level value for each luminance.

  The configuration of the luminance-specific noise level estimation unit 7D has been described above, but the present invention is not limited to this configuration. Here, the noise level estimation means 7D for each luminance accumulates the number of blocks for each level of the high frequency component for all luminances, and estimates the noise level value for each luminance. However, the luminance level-specific noise level estimation means 7D classifies the luminance in advance, performs the accumulation of the high-frequency component levels using one luminance in the classification as the representative luminance, and from the accumulated value of the representative luminance of each division, The noise level value may be estimated for each luminance by obtaining a cumulative value of other luminance by interpolation or the like. Alternatively, the noise level value of the representative luminance may be set as the noise level value of all luminances in the section.

  In the above, the structure of the moving image noise removal apparatus 1A-1D which concerns on 2nd-5th embodiment was demonstrated. Similar to the moving image noise removing device 1, the moving image noise removing devices 1A to 1D according to the second to fifth embodiments are operated by a moving image noise removing program that causes a general computer to function as each of the above-described means. Can be made.

The operations of the moving image noise removing apparatuses 1A to 1D are basically the same as those of the moving image noise removing apparatus 1 described with reference to FIG. 6, except for the following points.
That is, as shown in FIG. 21, the moving image noise removal apparatuses 1A to 1D take steps S3 and S4 of FIG. Is estimated.

  Specifically, the moving image noise removing apparatuses 1A to 1D perform orthogonal transform on the I × J blocks by the block dividing unit 3 after the operation of step S1 (step S3A). Then, the moving image noise removal apparatuses 1A to 1D delay the transform coefficient generated in step S3A by the delay unit 6 by the amount of the frame (step S3B).

The moving image noise removal device 1A~1D is by the frame noise level estimation means 7A, estimates a noise level value _{N L} in the frame (step S4A). About the operation | movement after this, about the operation | movement similar to the moving image noise removal apparatus 1 demonstrated in FIG. 6, the same code | symbol is attached | subjected and description is abbreviate | omitted.

In step S7A, the moving image noise removal apparatuses 1A to 1D use the intra-block noise removal unit 9 (intensity-specific intra-block noise removal unit 9A) to convert coefficients of the blocks B _{i, j} specified by the variables x and y. The noise level value N _L of the noise of the block estimated in step S4A is subtracted from B _{i, j} (x, y) in a range that does not become negative.

Further, in the moving image noise removing apparatuses 1C and 1D, in step S4A, the noise level value in the frame is estimated for each luminance by the luminance level estimating means 7C and 7D for each luminance. In step S7A, the moving image noise removing apparatuses 1C and 1D subtract the noise level value corresponding to the luminance of the block from the conversion coefficient for each block by the luminance-in-block noise removing unit 9A.
Through the above operation, the moving image noise removal apparatuses 1A to 1D estimate the noise level value for each frame and perform noise removal in units of blocks.

1 (1A, 1B, 1C, 1D) Moving image noise removing device 3 Block dividing means 5 Orthogonal transform means 6 Delay means 7 Block noise level estimating means (noise level estimating means)
7A, 7B Frame noise level estimation means (noise level estimation means)
7C, 7D Brightness-specific noise level estimation means (noise level estimation means)
9 Block noise removal means 9A Intensity block noise removal means (intra-block noise removal means)
11 Inverse orthogonal transformation means 21 High-frequency component detection means 22 Luminance detection means 23 High-frequency component accumulation means 23A High-frequency component accumulation means by luminance (high-frequency component accumulation means)
25 Noise level determination means 25A Brightness-specific noise level determination means (noise level determination means)
31 Line extraction means 32 Luminance detection means 33 Noise level detection means between lines 35 Level accumulation means 35A Level accumulation means by level (level accumulation means)
37 Noise level determination means 37A Brightness-specific noise level determination means (noise level determination means)

Claims (10)

A moving image noise removing apparatus for removing noise superimposed on a digital moving image signal,
Block dividing means for dividing the digital moving image signal into blocks of a predetermined size for each screen;
An orthogonal transform unit that orthogonally transforms the block divided by the block dividing unit for each block;
Noise level estimation means for estimating a noise level value indicating the magnitude of the noise component superimposed on the block based on the size of a predetermined high frequency component conversion coefficient of the block transformed by the orthogonal transformation means; ,
By calculating the maximum value of zero and the value obtained by subtracting the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means, In-block noise removal means for generating a noise removal conversion coefficient, which is a conversion coefficient from which the noise component is removed,
An inverse orthogonal transform unit that regenerates the digital moving image signal by performing an inverse orthogonal transform for each block of the noise removal transform coefficient generated by the intra-block noise removing unit;
A moving image noise removing apparatus comprising: In order to remove the noise superimposed on the digital video signal,
Block dividing means for dividing the digital moving image signal into blocks of a predetermined size for each screen;
Orthogonal transform means for orthogonally transforming the blocks divided by the block dividing means for each block;
A noise level estimation means for estimating a noise level value indicating the magnitude of the noise component superimposed on the block based on the size of a predetermined high frequency component conversion coefficient of the block transformed by the orthogonal transformation means;
By calculating the maximum value of zero and the value obtained by subtracting the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means, In-block noise removal means for generating a noise removal conversion coefficient that is a conversion coefficient from which the noise component is removed,
An inverse orthogonal transform unit that regenerates the digital moving image signal by performing an inverse orthogonal transform on each block of the noise removal transform coefficient generated by the intra-block noise removing unit;
A moving image denoising program characterized by functioning as A moving image noise removing apparatus for removing noise superimposed on a digital moving image signal,
Block dividing means for dividing the digital moving image signal into blocks of a predetermined size for each screen;
An orthogonal transform unit that orthogonally transforms the block divided by the block dividing unit for each block;
Among the transform coefficients that have been orthogonally transformed by the orthogonal transform means, the level values of the high frequency components of a predetermined number of transform coefficients for each screen are averaged to superimpose on the digital moving image signal. Noise level estimation means for estimating a noise level value indicating the magnitude of the noise component for each screen;
By calculating the maximum value of zero and the value obtained by subtracting the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means, In-block noise removal means for generating a noise removal conversion coefficient, which is a conversion coefficient from which the noise component is removed,
An inverse orthogonal transform unit that regenerates the digital moving image signal by performing an inverse orthogonal transform for each block of the noise removal transform coefficient generated by the intra-block noise removing unit;
A moving image noise removing apparatus comprising: The noise level estimation means includes
High-frequency component detection means for averaging the conversion coefficient of the predetermined high-frequency component converted by the orthogonal transform means within the block and detecting it as a level value of the high-frequency component of the block;
A noise level determination unit that determines a low level value of a level value of the high frequency component detected by the high frequency component detection unit or an average value for a predetermined number from the low level value as the noise level value;
The moving image noise removing apparatus according to claim 3, further comprising: The noise level estimation means includes
Further comprising a luminance detecting means for detecting the luminance for each block;
The noise level determination unit calculates a low level value of the high-frequency component level for each luminance detected by the luminance detection unit or an average value for a predetermined number of times from the low level value, and a size of the noise component for each luminance. Is determined as a noise level value indicating
The intra-block noise removing means includes:
From the conversion coefficient of the block, for each luminance corresponding to the direct current component of the block, a value obtained by subtracting the noise level value determined for each luminance by the noise level determination unit and the maximum value of zero is the noise removal conversion. The moving image noise removing device according to claim 4, wherein the moving image noise removing device is calculated as a coefficient. In order to remove the noise superimposed on the digital video signal,
Block dividing means for dividing the digital moving image signal into blocks of a predetermined size for each screen;
Orthogonal transform means for orthogonally transforming the blocks divided by the block dividing means for each block;
Among the transform coefficients that have been orthogonally transformed by the orthogonal transform means, the level values of the high frequency components of a predetermined number of transform coefficients for each screen are averaged to superimpose on the digital moving image signal. Noise level estimation means for estimating a noise level value indicating the magnitude of the noise component for each screen;
By calculating the maximum value of zero and the value obtained by subtracting the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means, In-block noise removal means for generating a noise removal conversion coefficient that is a conversion coefficient from which the noise component is removed,
An inverse orthogonal transform unit that regenerates the digital moving image signal by performing an inverse orthogonal transform on each block of the noise removal transform coefficient generated by the intra-block noise removing unit;
A moving image denoising program characterized by functioning as A moving image noise removing apparatus for removing noise superimposed on a digital moving image signal,
Block dividing means for dividing the digital moving image signal into blocks of a predetermined size for each screen;
An orthogonal transform unit that orthogonally transforms the block divided by the block dividing unit for each block;
The digital moving image signal is averaged within the screen by averaging pixel value differences between a horizontal or vertical line constituting the screen of the digital moving image signal and an adjacent line adjacent to the line in a spatial direction or a time direction. Noise level estimation means for estimating a noise level value indicating the magnitude of the noise component for each screen superimposed on the screen;
By calculating the maximum value of zero and the value obtained by subtracting the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means, In-block noise removal means for generating a noise removal conversion coefficient, which is a conversion coefficient from which the noise component is removed,
An inverse orthogonal transform unit that regenerates the digital moving image signal by performing an inverse orthogonal transform for each block of the noise removal transform coefficient generated by the intra-block noise removing unit;
A moving image noise removing apparatus comprising: The noise level estimation means includes
Line extraction means for sequentially extracting, as the correlation line group, the target line that is the line constituting the screen and the adjacent line adjacent to the target line from the digital moving image signal;
In the correlation line group extracted by the line extraction means, the difference between the pixel values of the corresponding pixels of the target line and the adjacent line is averaged and detected as an inter-line noise level in the correlation line group of the target line. Noise level detection means between lines;
Noise level determination means for determining an average value corresponding to the number of lines corresponding to a predetermined number from the low level value of the noise level between lines detected by the noise level detection means between the lines as the noise level value;
The moving image noise removing apparatus according to claim 7, further comprising: The noise level estimation means includes
Further comprising a luminance detecting means for detecting the luminance for each target line;
The noise level determination means uses a low level value of the inter-line noise level for each luminance detected by the luminance detection means or an average value for a predetermined number of times from the low level value, and the magnitude of the noise component for each luminance. Is determined as a noise level value indicating
The intra-block noise removing means includes:
9. The moving image according to claim 8, wherein the noise level value determined for each luminance by the noise level determination means is subtracted from the transform coefficient of the block for each luminance corresponding to the DC component of the block. Noise removal device. In order to remove the noise superimposed on the digital video signal,
Block dividing means for dividing the digital moving image signal into blocks of a predetermined size for each screen;
Orthogonal transform means for orthogonally transforming the blocks divided by the block dividing means for each block;
By averaging the pixel value difference between the line constituting the screen of the digital moving image signal and the adjacent line adjacent to the line in the screen, each of the screens superimposed on the digital moving image signal Noise level estimation means for estimating a noise level value indicating the magnitude of the noise component;
By calculating the maximum value of zero and the value obtained by subtracting the noise level value estimated by the noise level estimating means from the transform coefficient other than the DC component of the block transformed by the orthogonal transform means, In-block noise removal means for generating a noise removal conversion coefficient that is a conversion coefficient from which the noise component is removed,
An inverse orthogonal transform unit that regenerates the digital moving image signal by performing an inverse orthogonal transform on each block of the noise removal transform coefficient generated by the intra-block noise removing unit;
A moving image denoising program characterized by functioning as

Images