JP2007310635A - Image processing device and image processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To enhance image quality improvement by adaptive low-pass filtering. <P>SOLUTION: Prior to low-pass filtering, edge information is read in about a plurality of pixels, filter coefficients by which pixel values at edges are to be multiplied are reduced, and the filter coefficient reductions are added to a filter coefficient by which a reference pixel is to be multiplied. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to adaptive low-pass filtering.

  The low-pass filter (low-pass filter) process is a process for attenuating (cutting) high-frequency components (noise) in details included in an image. In this processing, since the low frequency component is emphasized by the amount corresponding to the cut of the high frequency component, the entire image tends to be blurred. Therefore, when the pixel to be processed (reference pixel) is an edge, that is, an image contour, an adaptive low-pass filter process that does not perform a low-pass filter is generally used. As a result, blurring of the edge portion is regulated, and subjective image quality can be improved. As a conventional technique regarding such a low-pass filter adaptively, Patent Document 1 discloses an image processing apparatus that adaptively controls low-pass filter processing in accordance with the direction of block distortion and suppresses block distortion. . Specifically, a low-pass filter process that removes horizontal frequency components is performed for the direction of the vertical block boundary, and a low-pass filter that removes vertical frequency components for the horizontal block boundary. Perform filtering.

Japanese Patent Laid-Open No. 2004-23666

  Incidentally, the adaptive low-pass filter process is intended to improve the image quality by considering whether or not the reference pixel is an edge. However, in the above-described conventional technique, a plurality of pixels (so-called “so-called”) are located around the reference pixel. It is not taken into account whether or not the surrounding pixels are edges. In general, there is a large difference in pixel values between a pixel corresponding to an edge and a pixel not corresponding to an edge adjacent to the pixel. Therefore, for example, when the reference pixel does not correspond to an edge and any of a plurality of peripheral pixels corresponds to an edge, the pixel value of the reference pixel after low-pass filtering is affected by the pixel value of the pixel corresponding to the edge. Is a factor that hinders image quality improvement.

  The present invention has been made in view of such circumstances, and an object thereof is to further improve image quality in adaptive low-pass filter processing.

  In order to solve such a problem, the first invention provides an image processing device that performs low-pass filter processing on a pixel region constituted by a reference pixel and a plurality of peripheral pixels located around the reference pixel. To do. The image processing apparatus includes a filter correction unit and a low-pass filter processing unit. The filter correction unit refers to the edge information indicating the edge of the image, reduces the filter coefficient corresponding to the pixel corresponding to the edge among the plurality of peripheral pixels, and increases the decrease to the filter coefficient corresponding to the reference pixel. By doing so, a filter to be applied to the pixel region is generated. The low-pass filter processing unit performs low-pass filter processing by applying a filter generated by the filter correction unit to the pixel area when the reference pixel does not correspond to an edge, and low-pass filter processing when the reference pixel corresponds to an edge. Do not do.

  Here, in the first invention, the low-pass filter processing unit preferably includes a register, a plurality of multipliers, and a divider. At this time, a register and a plurality of filter coefficients constituting the filter are stored. When the reference pixel does not correspond to an edge, the plurality of multipliers perform multiplication while positionally corresponding the plurality of filter coefficients to the plurality of pixel values constituting the pixel region. The divider performs division processing on the sum of the pixel values of the plurality of pixels subjected to multiplication processing by the sum of the plurality of filter coefficients.

  In the first invention, at least one of the plurality of filter coefficients is a multiplier of 2, and the multiplication process based on the filter coefficient that is a multiplier of 2 is preferably a left bit shift process. The sum of the filter coefficients is preferably a multiplier of 2, and the division process is preferably a right bit shift process.

  In the first invention, it is preferable that the low-pass filter processing unit further includes a plurality of selection units. The plurality of selection units select a plurality of pixel values constituting a pixel region or a pixel value of a reference pixel before the multiplication process is performed, and input the selected pixel values to the plurality of multipliers. In this case, the filter correction unit causes the selection unit corresponding to the pixel corresponding to the edge among the plurality of peripheral pixels to select the pixel value of the reference pixel, and among the plurality of peripheral pixels, the pixel that does not correspond to the edge Is selected by the selection unit corresponding to the pixel value of the pixel not corresponding to the edge. Thereby, a filter to be applied to the pixel region is generated.

  In the first invention, it is preferable that the image processing apparatus further includes an edge detection unit that detects edge information by applying a filter process for detecting an edge to a plurality of pixels constituting the image data. At this time, the edge detection unit preferably uses any one of the Sobel filter process, the Previt filter process, and the Laplacian filter process as the filter process for detecting the edge to be applied.

  The second invention provides an image processing method for performing a low-pass filter process on a pixel region constituted by a reference pixel and a plurality of peripheral pixels located around the reference pixel. This image processing method includes a correction step and a processing step. The correction step refers to edge information indicating an edge of the image, reduces a filter coefficient corresponding to a pixel corresponding to the edge among a plurality of peripheral pixels, and increases a decrease to a filter coefficient corresponding to the reference pixel. Thus, a filter to be applied to the pixel region is generated. Further, the processing step performs low pass filter processing by applying a filter generated by the filter correction unit to the pixel area when the reference pixel does not correspond to an edge, and low pass filter processing when the reference pixel corresponds to an edge. Do not do.

  According to the present invention, prior to performing the low-pass filter process, the filter correction process to be applied to the pixel region is performed with reference to edge information indicating the edge of the image. That is, among the plurality of peripheral pixels, the filter coefficient corresponding to the pixel corresponding to the edge is decreased, and the decrease is increased to the filter coefficient corresponding to the reference pixel. Thereby, since the influence of the pixel which is an edge can be suppressed in a low-pass filter process, an image quality improvement can be aimed at.

(First embodiment)
FIG. 1 is a configuration diagram of an image processing apparatus according to the present embodiment. This image processing apparatus performs low-pass filter processing based on edge information indicating an edge (contour) of an image on a pixel region constituted by a reference pixel and a plurality of peripheral pixels located around the reference pixel. Apply. Here, the reference pixel is a pixel to be processed by the low-pass filter process, and is sequentially shifted every time each process is completed. In addition, as an example, the plurality of peripheral pixels in the present embodiment are eight pixels that are diagonally adjacent to the reference pixel, and the pixel region has a 3 × 3 size (9 × 9 including the reference pixel and the plurality of peripheral pixels (9 Pixel). Note that the image data is stored in a storage unit (not shown) or the like provided outside the image processing apparatus, and the image processing apparatus sequentially or collectively stores the image data stored in the storage unit. Read and perform low-pass filter processing for each pixel. The image processing apparatus includes a low-pass filter processing unit 1, an edge detection unit 2, and a filter correction unit 3.

  When the reference pixel does not correspond to an edge, the low-pass filter processing unit 1 performs a low-pass filter process by applying a filter generated by the filter correction unit to the pixel region. When the reference pixel corresponds to an edge, the low-pass filter processing unit 1 Do not process. Whether it corresponds to an edge is determined by referring to the edge information.

  The edge detection unit 2 detects edge information by using a filter process for detecting an edge for a plurality of pixels constituting image data. The edge information is detected for each pixel. For example, prior to performing the low-pass filter process, the edge information is detected based on the pixel that is the edge detection target and its surrounding pixels.

  Prior to performing the low-pass filter process, the filter correction unit 3 refers to the edge information and generates a filter to be applied to the pixel region. Specifically, the filter correction unit 3 decreases the filter coefficient corresponding to the pixel corresponding to the edge among the plurality of peripheral pixels, and increases the decrease to the filter coefficient corresponding to the reference pixel. In the present embodiment, since the filter to be applied is in the low-pass filter processing unit 1, the filter correction unit 3 outputs a control signal CTL instructing the low-pass filter processing unit 1 to update (or change) the filter coefficient. Output.

  FIG. 2 is a configuration diagram of the low-pass filter processing unit 1 as an example. The low-pass filter processing unit 1 includes a register 1a, a multiplier 1b, an adder 1c, and a divider 1d. The register 1a stores a plurality of filter coefficients. In the present embodiment, the plurality of filter coefficients are nine filter coefficients corresponding to each of the nine pixels centered on the reference pixel (i, j) (shaded portion). Each filter coefficient is changed by a control signal CTL from the filter correction unit 3. In the case of the configuration shown in the figure, one filter coefficient corresponding to the reference pixel (i, j) is “4”, and eight filter coefficients corresponding to the peripheral pixels are each “1”.

  The multiplier 1b performs multiplication processing based on each of the plurality of filter coefficients and each of the pixel values of the plurality of pixels. In the present embodiment, the pixel values of the plurality of pixels are nine pixel values centered on the reference pixel (i, j) (shaded portion) of the low-pass filter processing. There are a plurality of multipliers 1b. Each of the multipliers 1b multiplies each of the nine pixel values and each of the nine filter coefficients stored in the register 1a while corresponding to each other, thereby obtaining nine new values. The pixel value of is calculated. The adder 1c adds these nine new pixel values.

  The divider 1d performs a division process on the sum of the pixel values of the plurality of pixels that have been multiplied by the sum of the plurality of filter coefficients. In the case of the example shown in FIG. 2, the total sum of the filter coefficients is “12 (“ 4 ”× 1, 1” × 8) ”, so that the divider 1 d has a total of 9 pixel values, Divide by the total "12" of filter coefficients. The divided pixel value is output as a corrected pixel value in the reference pixel (i, j).

  FIG. 3 is an explanatory diagram relating to edge detection processing. The edge detection unit 2 uses Sobel filter processing as edge extraction processing. The Sobel filter process is a basic technique used when extracting an edge from image data, and is a filter process that weights pixels near a pixel of interest (target pixel) when performing first-order differentiation. This figure is a coefficient matrix ((a): horizontal coefficient matrix, (b): vertical coefficient matrix) used for the Sobel filter processing. The edge detection unit 2 adds (A) a result of multiplying each pixel value of 3 × 3 size centered on a pixel targeted for processing (edge detection processing) by the coefficient matrix of FIG. A product obtained by multiplying the coefficient matrix of 3 (b) is added (B). Then, the edge detection unit 2 calculates the sum (A + B) of both, and detects the target pixel as an edge when the sum is a predetermined threshold value. Pixels outside the image data are detected as edges without exception.

  In the present embodiment, the Sobel filter processing described above is used as the edge detection processing. However, the present invention is not limited to this, and whether or not the target pixel is an edge based on the target pixel and its surrounding pixels. May be processing for obtaining binary (mask image) image data indicating. Therefore, it may be a Prewitt filter process using the same primary differentiation as the Sobel filter process, an edge detection process using a second derivative such as a Laplacian filter process, or the like. For the same reason, the size of the pixel block used for detecting the edge is not limited to 3 × 3.

  In the present embodiment, the edge detection unit 2 is provided in the image processing apparatus. However, the present invention is not limited to this. As described above, if there is an image compression method in which edge information is already held at the decoding stage, the edge detector 2 may not be provided. In this case, the image data compressed by these image compression methods is stored in a storage unit (not shown) outside the image processing apparatus, and the data relating to the image data (pixel value) is stored in the low-pass filter processing unit. 1 is directly input to the filter correction unit 3.

  FIG. 4 is an explanatory diagram relating to the filter coefficient, and is a coefficient matrix of the filter coefficient corresponding to the edge information of the reference pixel (i, j) and its surrounding pixels. Prior to performing the low-pass filter processing, the filter correction unit 3 reads edge information of the reference pixel (i, j) and its surrounding pixels in advance. Then, the filter correction unit 3 refers to the edge information and changes the reference pixel (i, j) and its peripheral pixel filter coefficients stored in the register 1a in the low-pass filter processing unit 1. In the present embodiment, the sum of the filter coefficients is “12”, and the filter correction unit 3 apportions this.

  For example, when the reference pixel (i, j) is not an edge and none of the peripheral pixels is an edge, the filter coefficient corresponding to the reference pixel (i, j) is set to “4”, and the filter coefficient corresponding to the peripheral pixel is set to “4”. "1" is set ((a) in the figure). Further, for example, the reference pixel (i, j) is not an edge, but three pixels (i−1, j + 1), (i−1, j), (i−1, j−1) among peripheral pixels. ) Is an edge (shaded area), the filter coefficients corresponding to these three pixels are set to “0”. Accordingly, the remaining filter coefficient “3 (“ 1 ”× 3”) ”is added to the filter coefficient“ 4 ”corresponding to the reference pixel (i, j), and the filter coefficient is set to“ 7 ”( b)).

  According to this embodiment, prior to performing the low-pass filter process, the filter correction unit 3 reduces the filter coefficient corresponding to the pixel corresponding to the edge among the plurality of peripheral pixels to “0”, and reduces the decrease. By increasing the filter coefficient to “7” corresponding to the reference pixel, a filter to be applied to the pixel region is generated. Thereby, in the low-pass filter process, the influence of the pixel that is the edge can be suppressed, and the image quality can be improved.

  In the present embodiment, the low-pass filter processing unit 1 performs processing based on a 3 × 3 pixel region (pixel block) centered on the reference pixel (i, j), but is not limited thereto. For example, the processing may be performed based on a 4 × 4, 5 × 5 pixel region. In this embodiment, the reference pixel is handled as one pixel. However, the present invention is not limited to this. For example, the reference pixel may be processed as a plurality of pixels (for example, 2 × 2 pixels). In that case, low-pass fill processing is performed for each of the plurality of pixels.

(Second Embodiment)
In the first embodiment, the total sum of the filter coefficients used for the low-pass filter processing is “12”, but the present invention is not limited to this, and the sum of different filter coefficients can be used. In particular, the multiplication process applied to the multiplier and the division process applied to the divider can be replaced with the bit shift process by setting the filter coefficient and the sum of the filter coefficients to a multiplier of 2. Further, the pixel value (or edge pixel) used for the low-pass filter processing is a pixel included in the pixel block defined by the rectangular pixel region, but is not limited to this, and depends on a trade-off relationship with the desired image quality. Different numbers and regions can be defined.

  FIG. 5 shows the low-pass filter processing unit 1 according to the present embodiment. Note that the image processing apparatus according to the present embodiment also includes the edge detection unit 2 and the filter correction unit 3 described in the first embodiment, but the configurations thereof are omitted because they are substantially the same. That is, this embodiment corresponds to a modification of the first embodiment. In the present embodiment, the sum of the filter coefficients of the low-pass filter process is “8”, and the value of the filter coefficient is “1” or “4”. In this embodiment, the peripheral pixels are four pixels on the top, bottom, left, and right of the reference pixel (i, j), and the pixel region used for the low-pass filter process is five pixels.

  The register 1a stores five filter coefficients. These five filter coefficients are stored as fixed values in the register 1a. In the example shown in FIG. 5, one filter coefficient corresponding to the pixel (i, j) to be processed is “4”, and four filter coefficients corresponding to the peripheral pixels are “1”. . As a result, the sum of the filter coefficients becomes 8 (“4” × 1, “1” × 4).

  The switch 1e is a selection unit that selects a pixel value input to the multiplier 1b based on the control signal CTL of the filter correction unit 3. There are a plurality of switches 1e, each of which has a circuit configuration with two inputs and one output, and selects either the pixel value corresponding to each of the peripheral pixels or the pixel value of the reference pixel (i, j). The selected pixel value is input to the multiplier 1b. For example, when the filter correction unit 3 determines that any of the surrounding pixels is an edge, the switch 1e that receives one of the pixels corresponding to the edge instructs the selection of the pixel value of the reference pixel. Further, when it is determined that a part of the peripheral pixels is not an edge, the switch 1e that inputs one of the pixels corresponding to the edge instructs to select the pixel value of the peripheral pixel. As a result, the filter coefficient for the pixel that is substantially an edge is subtracted from “1” to “0”, and the filter coefficient for the reference pixel is added from “4” to “5”.

  That is, the change to the filter coefficient stored in the register 1a (in the first embodiment) by the filter correction unit 3 is replaced with the control of the switch 1e. Thereby, the same function (subtraction of the filter coefficient for the pixel corresponding to the edge and addition for the reference pixel) as in the first embodiment can be realized.

  The multiplier 1b performs multiplication processing based on each of the five filter coefficients and each of the pixel values of the five pixels. At this time, for the pixel value of the reference pixel (i, j), the multiplier 1b corresponding to the filter coefficient “4” can perform a 2-bit left bit shift process (× 4). Since each of the plurality of filter coefficients stored in the register 1a is “1”, the result calculated by the multiplication process remains the pixel value of one pixel input to the multiplier 1b. Therefore, in the present embodiment, the low-pass filter processing unit 1 does not have to provide the register 1a and the multiplier 1b related to the filter coefficient “1”.

  The divider 1d performs division processing that divides the five pixel values calculated by the adder 1c. At this time, since the total sum of the filter coefficients is “8”, the divider 1d can perform a 3-bit right bit shift process (1/8 times) on the calculated sum. The pixel value subjected to the bit shift process is output as a corrected pixel value of the reference pixel (i, j).

  For example, consider a case where only the left pixel (i−1, j) of the peripheral pixels is a pixel corresponding to an edge. In that case, the switch 1e having the left pixel (i−1, j) as one input by the control signal CTL of the filter correction unit 3 selects the pixel value of the reference pixel (i, j), and the rest The three switches 1e having one pixel as an input select the pixel value of the pixel. As a result, the adder 1c has a value obtained by shifting the pixel value of the reference pixel (i, j) to the left by 2 bits (doubled) and four pixels (i, j), (i, j +) including the reference pixel. 1), (i, j-1), (i + 1, j) pixel values are multiplied by the filter coefficient (in this case, the filter coefficient is “1”, and is output as it is)) and input. . Then, the added value is shifted by 3 bits to the right by the divider 1d, and the corrected pixel value at the reference pixel (i, j) is output.

  According to the present embodiment, the same effect (image quality improvement) as that of the first embodiment is obtained, and the filter coefficient and the sum of the filter coefficients are set to values (multiplier of 2) that can be implemented only by addition / subtraction and bit shift. For this reason, the multiplication processing / division processing constituting the low-pass filter processing can be replaced with the bit shift processing, so that the arithmetic processing can be speeded up, and the arithmetic processing related to the entire low-pass filter processing can be speeded up. .

1 is a configuration diagram of an image processing apparatus according to a first embodiment. Configuration diagram of low-pass filter processing unit Illustration about edge detection Illustration about filter coefficients The block diagram of the low-pass filter process part concerning 2nd Embodiment

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Low pass filter process part 1a Register 1b Multiplier 1c Adder 1d Divider 1e Switch 2 Edge detection part 3 Filter correction part

Claims (8)

In an image processing apparatus that performs a low-pass filter process on a pixel region formed by a reference pixel and a plurality of peripheral pixels located around the reference pixel.
Referring to edge information indicating an edge of an image, a filter coefficient corresponding to a pixel corresponding to the edge among the plurality of peripheral pixels is decreased, and the decrease is increased to a filter coefficient corresponding to the reference pixel. A filter correction unit for generating a filter to be applied to the pixel region;
When the reference pixel does not correspond to an edge, low-pass filter processing is performed by applying the filter generated by the filter correction unit to the pixel region, and when the reference pixel corresponds to an edge, the low-pass filter processing An image processing apparatus comprising: a low-pass filter processing unit that does not perform processing. The low-pass filter processing unit
A register for storing a plurality of filter coefficients constituting the filter;
When the reference pixel does not correspond to an edge, a plurality of multipliers that perform a multiplication process while positionally corresponding the plurality of filter coefficients to a plurality of pixel values constituting the pixel region;
The image processing apparatus according to claim 1, further comprising a divider that performs a division process on a sum of pixel values of the plurality of pixels subjected to the multiplication process by a sum of the plurality of filter coefficients.   3. The image according to claim 2, wherein at least one of the plurality of filter coefficients is a multiplier of 2, and the multiplication process based on the filter coefficient that is the multiplier of 2 is a left bit shift process. Processing equipment.   4. The image processing apparatus according to claim 2, wherein the sum of the filter coefficients is a multiplier of 2, and the division processing is right bit shift processing. The low-pass filter processing unit
Prior to performing the multiplication process, a plurality of pixel values constituting the pixel region or a pixel value of the reference pixel are selected, and the selected pixel values are input to the plurality of multipliers. Further comprising
The filter correction unit is
A selection unit corresponding to a pixel that does not correspond to an edge among the plurality of peripheral pixels, wherein a selection unit corresponding to a pixel corresponding to an edge among the plurality of peripheral pixels selects a pixel value of the reference pixel. The image processing apparatus according to claim 2, wherein a filter to be applied to the pixel region is generated by selecting a pixel value of a pixel not corresponding to the edge. The image processing apparatus includes:
6. The image processing apparatus according to claim 1, further comprising an edge detection unit that detects the edge information by applying a filter process that detects an edge to a plurality of pixels constituting the image data. The image processing apparatus described in 1.   The image processing apparatus according to claim 6, wherein the edge detection unit uses any one of a Sobel filter process, a Previt filter process, and a Laplacian filter process as a filter process for detecting an edge to be applied. . In an image processing method for performing a low-pass filter process on a pixel region constituted by a reference pixel and a plurality of peripheral pixels located around the reference pixel,
Referring to edge information indicating an edge of an image, a filter coefficient corresponding to a pixel corresponding to the edge among the plurality of peripheral pixels is decreased, and the decrease is increased to a filter coefficient corresponding to the reference pixel. A correction step for generating a filter to be applied to the pixel region;
When the reference pixel does not correspond to an edge, low-pass filter processing is performed by applying the filter generated by the filter correction unit to the pixel region, and when the reference pixel corresponds to an edge, the low-pass filter processing An image processing method comprising: a processing step that does not perform processing.

Images