JP4239484B2 - Image processing method, image processing program, and image processing apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an image processing method, an image processing program, and an image processing apparatus for processing image data obtained by a color filter having a delta arrangement.
[0002]
[Prior art]
An electronic camera images a subject with an image sensor such as a CCD. In this imaging device, a Bayer arrangement is known in which three color filters of RGB (red, green, and blue) are arranged as shown in FIG. A delta arrangement arranged as shown in FIG. 24B is also known. Furthermore, a honeycomb arrangement arranged as shown in FIG. 24C is also known. Various image processing methods have been proposed for image data obtained by the Bayer array, such as US Pat. No. 5,552,827, US Pat. No. 5,629,734, and Japanese Patent Laid-Open No. 2001-245314.
[0003]
On the other hand, for image data obtained in a delta arrangement, an image processing method has been proposed in Japanese Patent Application Laid-Open No. 8-340455, US Pat. No. 5,805,217, and the like.
[0004]
[Problems to be solved by the invention]
However, in US Pat. No. 5,805,217, image processing according to the characteristics unique to the delta arrangement is not performed. Japanese Patent Application Laid-Open No. 8-340455 proposes only a simple image restoration method. Also, the various image processing methods proposed for the Bayer array are not necessarily applicable to the delta array as they are.
[0005]
The present invention is an image that outputs high-definition interpolated image data or high-definition image data converted to another color system based on image data obtained by a triangular lattice color filter having a delta arrangement or the like A processing method, an image processing program, and an image processing apparatus are provided.
[0006]
[Means for Solving the Problems]
The invention of claim 1 is applied to an image processing method, wherein a plurality of pixels represented by first to nth color components (n ≧ 2) and having color information of one color component in one pixel are in a triangular lattice shape. An image acquisition procedure for acquiring the first image arranged in the image, an interpolation procedure for interpolating the first color component to the pixel lacking the first color component using the color information of the acquired first image, An output procedure for outputting a second image based on the color information of one image and the interpolated color information, and the interpolation procedure includes a first color component for an interpolation target pixel of the first image. using the color information of a region including a pixel adjacent to the second, an average information of the first color component, have row interpolation, further comprising a similarity determining step determines the intensity of similarity to at least three directions, The interpolation procedure uses the average information of the first color component according to the strength of similarity determined in the similarity determination procedure. And requests.
The invention of claim 2 is applied to an image processing method, and is a first image in which a plurality of pixels represented by a plurality of color components and having color information of one color component per pixel are arranged in a triangular lattice pattern. A color for generating color information of a color component different from the color information of the first image by performing weighted addition of the acquired color information of the first image with a variable coefficient value of zero or more by acquiring the color information of the first image An information generation procedure, and an output procedure for outputting the second image using the generated color information. The color information generation procedure includes a color component of the pixel and a processing target pixel of the first image. Color information in an area including a pixel in which different color components are second closest is weighted.
A third aspect of the present invention, in the image processing method according to claim 2, further comprising a similarity determining step determines the intensity of similarity to at least three directions, the color information generation step is determined by the similarity determination procedures The coefficient value of the weighted addition is made variable according to the strength of similarity.
According to a fourth aspect of the present invention, in the image processing method according to the second or third aspect , the first image is represented by first to third color components, and the pixel having the first color component of the first image is provided. In the case of the processing target pixel, the color information generation procedure weights the color information in the region including the processing target pixel, the pixel in which the second color component is second closest, and the pixel in which the third color component is second closest. It is to add.
According to a fifth aspect of the present invention, in the image processing method according to any one of the second to fourth aspects, the color information of the first image generated by the color information generation procedure before the output procedure after the color information generation procedure And a correction procedure for correcting color information of different color components by a filter process including a predetermined fixed filter coefficient.
According to a sixth aspect of the present invention, in the image processing method according to the fifth aspect , the filter coefficient includes positive and negative values.
The invention of claim 7 is applied to an image processing method, wherein a plurality of pixels represented by first to nth color components (n ≧ 2) and having color information of one color component in one pixel are in a triangular lattice shape. An image acquisition procedure for acquiring a first image arranged in the color space, and a color difference generation procedure for generating color information of a color difference component between the first color component and the second color component using the color information of the first image And an output procedure for outputting the second image using the color information of the generated color difference component, and the color difference generation procedure includes at least a second for a pixel having the first color component of the first image. The color information of the color difference component is generated using the color information of the pixel with the second closest color component.
According to an eighth aspect of the present invention, in the image processing method according to the seventh aspect , the color difference generation procedure is performed on the processing target pixel having the first color component of the first image: 1) the first color component of the pixel And 2) generating color information of color difference components based on average information of color information of the second color component in an area including a pixel in which the second color component is second closest to the pixel It is.
According to a ninth aspect of the present invention, in the image processing method according to the seventh or eighth aspect , the color difference generation procedure further generates color information of the color difference component based on the curvature information of the second color component relating to the processing target pixel. Is.
The invention of claim 10 is the image processing method according to any one of claims 7 to 9 , further comprising a similarity determination procedure for determining the strength of similarity in at least three directions, wherein the color difference generation procedure is similar. The color information of the color difference component is generated according to the strength of the property.
The invention according to claim 11 is the image processing method according to any one of claims 1 to 10 , wherein the output procedure outputs the second image at the same pixel position as the first image.
The invention of claim 12 is an image processing program for causing a computer to execute the procedure of the image processing method according to any one of claims 1 to 11 .
The invention of claim 13 is a computer-readable recording medium on which the image processing program of claim 12 is recorded.
A fourteenth aspect of the present invention is an image processing apparatus equipped with the image processing program of the twelfth aspect .
[0007]
DETAILED DESCRIPTION OF THE INVENTION
-First embodiment-
(Configuration of electronic camera)
FIG. 1 is a functional block diagram of the electronic camera according to the first embodiment. The electronic camera 1 includes an A / D conversion unit 10, an image processing unit 11, a control unit 12, a memory 13, a compression / decompression unit 14, and a display image generation unit 15. Further, an interface unit 17 for a memory card that interfaces with a memory card (card-like removable memory) 16 and an external device that interfaces with an external device such as a PC (personal computer) 18 via a predetermined cable or wireless transmission path. An interface unit 19 is provided. Each of these blocks is connected to each other via a bus 29. The image processing unit 11 is composed of, for example, a one-chip microprocessor dedicated to image processing.
[0008]
The electronic camera 1 further includes a photographing optical system 20, an image sensor 21, an analog signal processing unit 22, and a timing control unit 23. An optical image of the subject acquired by the imaging optical system 20 is formed on the image sensor 21, and the output of the image sensor 21 is connected to the analog signal processing unit 22. The output of the analog signal processing unit 22 is connected to the A / D conversion unit 10. The output of the control unit 12 is connected to the timing control unit 23, and the output of the timing control unit 23 is connected to the image sensor 21, the analog signal processing unit 22, the A / D conversion unit 10, and the image processing unit 11. The image sensor 21 is constituted by a CCD, for example.
[0009]
The electronic camera 1 further includes an operation unit 24 and a monitor 25 corresponding to a release button, a mode switching selection button, and the like. The output of the operation unit 24 is connected to the control unit 12, and the output of the display image generation unit 15 is connected to the monitor 25.
[0010]
Note that a monitor 26, a printer 27, and the like are connected to the PC 18, and an application program recorded on the CD-ROM 28 is installed in advance. In addition to a CPU, a memory, and a hard disk (not shown), the PC 18 includes a memory card interface unit (not shown) that interfaces with the memory card 16, an electronic camera 1, etc. An external interface unit (not shown) that interfaces with an external device is provided.
[0011]
In the electronic camera 1 configured as shown in FIG. 1, when the shooting mode is selected by the operator via the operation unit 24 and the release button is pressed, the control unit 12 passes through the timing control unit 23 and the image sensor 21. The timing control is performed on the analog signal processing unit 22 and the A / D conversion unit 10. The image sensor 21 generates an image signal corresponding to the optical image. The image signal is subjected to predetermined signal processing by the analog signal processing unit 22, digitized by the A / D conversion unit 10, and supplied to the image processing unit 11 as image data.
[0012]
In the electronic camera 1 according to the present embodiment, R (red), G (green), and B (blue) color filters are arranged in a delta arrangement (described later) in the image sensor 21, and are supplied to the image processing unit 11. The image data is shown in the RGB color system. Each pixel constituting the image data has color information of any one of RGB color components.
[0013]
The image processing unit 11 performs image processing such as gradation conversion and edge enhancement in addition to performing image data conversion processing described later on such image data. The image data that has undergone such image processing is subjected to predetermined compression processing by the compression / decompression unit 14 as necessary, and is recorded in the memory card 16 via the memory card interface unit 17.
[0014]
The image data for which image processing has been completed is recorded on the memory card 16 without being subjected to compression processing, or converted into a color system adopted by the monitor 26 or the printer 27 on the PC 18 side, and the external interface unit 19. You may supply to PC18 via. When the playback mode is selected by the operator via the operation unit 24, the image data recorded in the memory card 16 is read out via the memory card interface unit 17 and is compressed / decompressed by the compression / decompression unit 12. The decompression process is performed and displayed on the monitor 25 via the display image generation unit 15.
[0015]
Note that the decompressed image data is not displayed on the monitor 25 but is converted to the color system adopted by the monitor 26 or printer 27 on the PC 18 side and supplied to the PC 18 via the external interface unit 19. You may do it.
[0016]
(Image processing)
Next, a process for interpolating image data obtained by the delta arrangement on a triangular lattice and restoring it to square lattice data that can be easily handled on a computer will be described. FIG. 25 is a diagram showing the concept of these processes. The triangular lattice means an array in which pixels of the image sensor are arranged with a shift of ½ pixel for each row. This is an array that forms a triangle when the centers of adjacent pixels are connected. The center point of the pixel may be called a grid point. The delta arrangement in FIG. 24B is arranged in a triangular lattice shape. The image obtained by the arrangement as shown in FIG. 24B may be said to be an image in which pixels are arranged in a triangle. A square lattice means an array in which pixels of the image sensor are arranged without being shifted for each row. When the centers of adjacent pixels are connected, a square is formed. The Bayer array in FIG. 24A is arranged in a square lattice pattern. The image obtained by the arrangement as shown in FIG. 24A may be said to be an image in which pixels are arranged in a rectangle (square).
[0017]
Further, as will be described later, when the process of shifting the image data of the delta arrangement by 1/2 pixel every other row, square lattice data is generated. Interpolation processing on a triangular lattice means that interpolation processing is performed in the state of image data obtained in a delta arrangement. That is, it means that the color information of the color component that is missing at the pixel position of the delta arrangement is interpolated.
[0018]
FIG. 2 is a flowchart showing an outline of image processing performed by the image processing unit 11. In step S1, an image obtained by the image sensor 21 having a delta arrangement is input. In step S2, the similarity is calculated. In step S3, the similarity is determined based on the similarity obtained in step S2. In step S4, based on the similarity determination result obtained in step S3, an interpolated value of the missing color component in each pixel is calculated. In step S5, the obtained RGB color image is output. The RGB color image output in step S5 is image data obtained on a triangular lattice.
[0019]
Next, in step S6, one-dimensional displacement processing is performed on the image data obtained on the triangular lattice. As will be described later, the one-dimensional displacement process is performed on every other row of image data. In step S7, square lattice image data is output by combining the image data subjected to the one-dimensional displacement processing and the image data not subjected to the one-dimensional displacement processing. Steps S1 to S5 are interpolation processing on a triangular lattice, and steps S6 and S7 are square processing. Details of these processes will be described below.
[0020]
-Interpolation processing on triangular lattice-
In the following description, the interpolation process at the R pixel position will be described as a representative. FIG. 3 is a diagram illustrating a positional relationship of pixels obtained by the image sensor 21 having a delta arrangement. The pixels of the image sensor 21 are arranged with a shift of ½ pixel for each row, and the color filters are arranged on each pixel at a ratio of RGB color components of 1: 1: 1. That is, the colors are evenly distributed. For reference, RGB color filters are arranged in a ratio of 1: 2: 1 (FIG. 24A).
[0021]
A pixel having R component color information is called an R pixel, a pixel having B component color information is called a B pixel, and a pixel having G component color information is called a G pixel. The image data obtained by the image sensor 21 has color information of only one color component for each pixel. Interpolation processing is processing for obtaining color information of other color components missing in each pixel by calculation. Hereinafter, a case where the color information of the G and B components is interpolated at the R pixel position will be described.
[0022]
In FIG. 3, a processing target pixel (interpolation target pixel) that is an R pixel is referred to as Rctr. In addition, each pixel position existing around the pixel Rctr is expressed using an angle. For example, a B pixel existing in the direction of 60 degrees is expressed as B060, and a G pixel is expressed as G060. However, this angle is not a precise angle but an approximation. Also, the direction connecting 0 ° -180 ° is the 0 ° direction, the direction connecting 120 ° -300 ° is the 120 ° direction, the direction connecting 240 ° -60 ° is the 240 ° direction, and the direction connecting 30 ° -210 ° is The direction connecting 30 degrees, 150 degrees to 330 degrees is referred to as 150 degrees direction, and the direction connecting 270 degrees to 90 degrees is referred to as 270 degrees direction.
[0023]
1. Similarity calculation
Similarities C000, C120, and C240 in the directions of 0 degree, 120 degrees, and 240 degrees are calculated. In the first embodiment, as shown in equations (1), (2), and (3), similarity between different colors constituted by different color components is obtained.
C000 = {| G000-Rctr | + | B180-Rctr | + | (G000 + B180) / 2-Rctr |} / 3 ... (1)
C120 = {| G120-Rctr | + | B300-Rctr | + | (G120 + B300) / 2-Rctr |} / 3 ... (2)
C240 = {| G240-Rctr | + | B060-Rctr | + | (G240 + B060) / 2-Rctr |} / 3 ... (3)
Thus, the similarity between different colors defined between adjacent pixels has the ability to spatially resolve the image structure of the Nyquist frequency defined by the triangular lattice of the delta arrangement.
[0024]
Next, the accuracy of the similarity is increased by performing the peripheral addition of the similarity and considering the continuity with the peripheral pixels. Here, the similarity obtained for each pixel is subjected to peripheral addition using the coefficients shown in FIG. 4 for the R position. The similarity with marginal addition is written in lower case. [] Represents a pixel position on the delta array as viewed from the pixel to be processed. Equation (4) represents the peripheral addition of the degree of similarity in the 0 degree direction.
c000 = {6 * C000 [Rctr]
+ C000 [R030] + C000 [R150] + C000 [R270]
+ C000 [R210] + C000 [R300] + C000 [R090]} / 12 ... (4)
The c120 in the 120 degree direction and the c240 in the 240 degree direction are obtained in the same manner.
[0025]
2. Similarity determination Since the similarity described above shows a greater similarity as the value is smaller, the strength of similarity in each direction is determined so as to continuously change with the reciprocal ratio of the similarity. That is, it is determined by (1 / c000) :( 1 / c120) :( 1 / c240). Specifically, the next weighting coefficient is calculated.
[0026]
Expressing the similarity in each direction as weighting factors w000, w120, w240 normalized by 1,
w000 = (c120 * c240 + Th) / (c120 * c240 + c240 * c000 + c000 * c120 + 3 * Th) ... (5)
w120 = (c240 * c000 + Th) / (c120 * c240 + c240 * c000 + c000 * c120 + 3 * Th) ... (6)
w240 = (c000 * c120 + Th) / (c120 * c240 + c240 * c000 + c000 * c120 + 3 * Th) ... (7)
It is obtained by. However, the threshold value Th is a term for preventing divergence and takes a positive value. Usually, Th = 1 may be set, but this threshold value may be increased for a noisy image such as a high-sensitivity photographed image. The weighting factors w000, w120, and w240 are values corresponding to the strength of similarity.
[0027]
In the Bayer array, as shown in US Pat. No. 5,552,827, US Pat. No. 5,629,734, and Japanese Patent Laid-Open No. 2001-245314, as the direction determination method in G interpolation, a continuous determination method using a weight coefficient and a discrete determination method using a threshold determination are used. There are two ways. In the Bayer array, the nearest neighbor G components are densely present in the four directions with respect to the interpolation target pixel, so that either the continuous determination method or the discrete determination method can be used without any problem. However, in the delta arrangement in which the nearest neighbor G component exists only in the three directions of the 0 degree direction, the 120 degree direction, and the 240 degree direction, continuous direction determination using a weighting factor is important. The nearest neighbor G component is a pixel that is in contact with the interpolation target pixel and has a G component. In FIG. 3, G000, G120, and G240.
[0028]
3. Interpolation value calculation Interpolation values of the G component and the B component are calculated using the weighting coefficients. The interpolation value is composed of two terms, average information and curvature information.
<Average information>
Gave = w000 * Gave000 + w120 * Gave120 + w240 * Gave240 ... (8)
Bave = w000 * Bave000 + w120 * Bave120 + w240 * Bave240 ... (9)
here,
Gave000 = (2 * G000 + G180) / 3 ... (10)
Bave000 = (2 * B180 + B000) / 3 ... (11)
Gave120 = (2 * G120 + G300) / 3 ... (12)
Bave120 = (2 * B300 + B120) / 3 ... (13)
Gave240 = (2 * G240 + G060) / 3 ... (14)
Bave240 = (2 * B060 + B240) / 3 ... (15)
It is.
<Curvature information>
dR = (6 * Rctr-R030-R090-R150-R210-R270-R330) / 12 ... (16)
<Interpolated value>
Gctr = Gave + dR ... (17)
Bctr = Bave + dR ... (18)
[0029]
Usually, in G interpolation of a Bayer array, average information is obtained using the nearest neighbor G component. However, in the case of the delta arrangement, it has been experimentally found that if the same is done, there will be a problem that the vertical line and the boundary line in the direction of 30 degrees and the direction of 150 degrees become rough. For example, in the case of a vertical line boundary, it is assumed that the weighting coefficients are w000≈0 and w120≈w240≈0.5, and this corresponds to a portion that performs an average process between two directions. However, only the nearest neighbor pixel average takes the left two-point average at a point along the vertical line, and the next adjacent diagonal upper right or lower right point takes the right two-point average. it is conceivable that.
[0030]
Therefore, if the averaging process is performed according to the distance ratio from the interpolation target pixel in consideration of the second adjacent pixel, for example, in the case of a vertical line, the average is always taken from both the right side and the left side. , 150 degrees direction boundary line drastically improve. Therefore, as shown in Expressions (10) to (15), an averaging process including up to the second adjacent pixel is performed. Thereby, the spatial resolving power in directions of 30 degrees, 150 degrees, and 270 degrees is improved. For example, in Expression (10), G000 is the most adjacent pixel or the first adjacent pixel, and G180 is the second adjacent pixel. In Expression (12), G120 is the most adjacent pixel or the first adjacent pixel, and G300 is the second adjacent pixel. The same applies to other equations. It can be said that the first adjacent pixel is a pixel separated by about 1 pixel pitch, and the second adjacent pixel is a pixel separated by about 2 pixel pitch. It can be said that Rctr and G120 are about 1 pixel apart, and Rctr and G300 are about 2 pixels apart. FIG. 22 is a diagram for defining adjacent pixels. center is a pixel to be processed, nearest is the closest or nearest neighbor or first adjacent pixel, and 2nd is a second adjacent pixel.
[0031]
In the prior art of Bayer array interpolation, the term corresponding to the curvature information dR is usually calculated in consideration of the same directionality as the average information. However, in the delta arrangement, the directionality of the average information (0 degree, 120 degree, 240 degree direction) and the directionality of the extractable curvature information (30 degree, 150 degree, 270 degree direction) do not match. In such a case, the curvature information in the 0 degree direction is defined by averaging the curvature information in the 30 degree direction and the 150 degree direction, and the interpolation value may be calculated in consideration of the directionality of the curvature information in the same manner as the Bayer array. Although it is possible, it has been experimentally found that it is more effective to extract curvature information uniformly without considering directionality. Therefore, in the present embodiment, average information considering directionality is corrected with curvature information having no directionality. Thereby, it is possible to improve the gradation sharpness up to the high frequency region in all directions.
[0032]
The equation (16) for obtaining the curvature information dR uses the coefficient values shown in FIG. Expression (16) obtains the difference between the interpolation target pixel Rctr and the peripheral pixel and the difference between the opposite peripheral pixel and the interpolation target pixel Rctr, and further obtains the difference between them. Therefore, it is obtained by secondary differential calculation.
[0033]
The curvature information dR is information indicating the degree of change of color information of a certain color component. When the value of the color information of a certain color component is plotted and represented by a curve, this is information indicating a change in the curve of the curve and the degree of the curve. That is, the amount reflects the structural information regarding the unevenness of the color information change of the color component. In the present embodiment, the curvature information dR at the R pixel position is obtained by using the Rctr of the interpolation target pixel and the color information of the surrounding R pixels. In the case of the G pixel position, the color information of the G component is used, and in the case of the B pixel position, the color information of the B component is used.
[0034]
As described above, “interpolation processing of G and B components at the R pixel position” could be performed. “Interpolation processing of B and R components at G pixel position” is performed by cyclically replacing the symbol of “G and B component interpolation processing at R pixel position” with R as G, G as B, and B as R. The same process may be performed. In addition, “interpolation processing of R and G components at the B position” cyclically replaces the symbol of “interpolation processing of B and R components at the G position” with G as B, B as R, and R as G. The same process may be performed. That is, the same calculation routine (subroutine) can be used in each interpolation process.
[0035]
The image restored as described above can bring out all the limit resolution performance of the delta arrangement in the spatial direction. That is, when viewed in the frequency space (k space), as shown in FIG. 6, all hexagonal achromatic color reproduction regions of the delta arrangement are resolved. Also, a clear image can be obtained with respect to the gradation direction. This is an extremely effective technique especially for images with many achromatic parts.
[0036]
-Square processing-
Next, a process for restoring the image data interpolated on the triangular lattice as described above to square lattice data that can be easily handled on a computer will be described. As conventional techniques for restoring the data of a single-plate image sensor shifted by 1/2 pixel in a row unit to square lattice data having three colors, JP-A-8-340455 for the delta arrangement and JP-A No. JP-A-2001-103295 and JP-A-2000-194386.
[0037]
Japanese Patent Application Laid-Open No. 8-340455 shows an example in which restoration data is generated at a pixel position different from that of the triangular lattice, or restoration data is generated in a virtual square lattice having a pixel density twice that of the triangular lattice. In addition, an example is shown in which the triangular grid and half rows are restored to the same pixel position, and the remaining half rows are restored to pixel positions shifted by 1/2 pixel from the triangular grid. However, this directly implements the interpolation processing at the square lattice position, and applies different interpolation processing when the distance from the adjacent pixel of the triangular lattice changes. On the other hand, in Japanese Patent Laid-Open No. 2001-103295, it is squared by two-dimensional cubic interpolation, and in Japanese Patent Laid-Open No. 2000-194386, virtual double dense square lattice data is generated.
[0038]
There are various methods as described above. In the first embodiment, a method for maintaining the performance of the delta arrangement to the maximum is shown.
[0039]
In the first embodiment, as described above, first, the interpolation data is restored with the triangular lattice at the same pixel position as that of the delta arrangement. As a result, the spatial frequency limit resolution performance of the delta arrangement can be extracted. In addition, if the color difference correction process and the edge enhancement process are performed in a triangular arrangement, the effect works isotropically. Therefore, once the image is restored on the triangular lattice, RGB values are generated for each pixel.
[0040]
Next, the image data generated on the triangular lattice is square-transformed. It has been experimentally found that it is important to retain the original data as much as possible in order to maintain the resolution performance of the triangular lattice. Accordingly, a displacement process is performed by shifting half a line by half a line every other line, and the remaining half lines are not processed, and the Nyquist resolution of the vertical lines of the triangular arrangement is maintained. It was confirmed by experiments that the displacement processing can be maintained with the vertical line resolution of the triangular lattice almost without any problem, although it is somewhat affected by the Nyquist frequency of the square lattice when self-estimation is performed by cubic interpolation in one dimension of the processing target row. .
[0041]
That is, the processing of the following formula (19) is performed for even rows, and then the substitution processing of formula (20) is performed. Thereby, the color information of the R component at the pixel position [x, y] is obtained.
tmp_R [x, y] = {-1 * R [x- (3/2) pixel, y] + 5 * R [x- (1/2) pixel, y]
+ 5 * R [x + (1/2) pixel, y] -1 * R [x + (3/2) pixel, y]} / 8 ... (19)
R [x, y] = tmp_R [xy] ... (20)
The G component and B component are obtained in the same manner. The process of Expression (19) is referred to as a cubic displacement process. Further, since the displacement processing is performed on the position in the one-dimensional direction, it is also referred to as one-dimensional displacement processing. FIG. 7 is a diagram showing coefficient values used in the equation (19). The cubic one-dimensional displacement processing can be said to be processing for applying a one-dimensional filter composed of positive and negative coefficient values.
[0042]
The image restoration method as described above can not only maintain the limit resolution of the triangular arrangement to the maximum, but also the image restoration process on the triangular lattice can use the same processing routine for all pixels, and finally the half of Since only a simple one-dimensional process is required for the rows, a simpler algorithm can be achieved compared to the prior art without increasing the amount of data. The effect is particularly great when implemented in hardware.
[0043]
-Second Embodiment-
In the first embodiment, the similarity between different colors is determined and the similarity is determined. In the second embodiment, the similarity is determined by determining the similarity between the same colors. The second embodiment is effective for an image including a lot of coloring portions. Since the configuration of the electronic camera 1 of the second embodiment is the same as that of FIG. 1 of the first embodiment, the description thereof is omitted. The image processing will be described with a focus on differences from the first embodiment. Unless otherwise specified, the process is the same as in the first embodiment.
[0044]
(Image processing)
-Interpolation processing on triangular lattice-
1. Similarity calculation
Similarities C000, C120, and C240 in the directions of 0 degree, 120 degrees, and 240 degrees are calculated. In the second embodiment, as shown in equations (20), (21), and (22), the similarity between the same colors constituted by the same color components is obtained. A calculation is performed between pixels separated by 3 pixel pitches. FIG. 8 is a diagram illustrating pixel positions used in the equations (20), (21), and (22).
C000 = {(| R000-Rctr | + | R180-Rctr |) / 2 + | G000-G180 | + | B000-B180 |} / 3 ... (20)
C120 = {(| R120-Rctr | + | R300-Rctr |) / 2 + | G120-G300 | + | B120-B300 |} / 3 ... (21)
C240 = {(| R240-Rctr | + | R060-Rctr |) / 2 + | G240-G060 | + | B240-B060 |} / 3 ... (22)
The similarity between the same colors defined in this way has the ability to spatially resolve the horizontal line in the chromatic part and the image structure in the 120 degree direction and the 240 degree direction, and for systems with chromatic aberration. Have the ability to discriminate image structures without being affected.
[0045]
Next, peripheral addition of similarity is performed in the same manner as in the first embodiment.
[0046]
2. Similarity determination is the same as in the first embodiment.
[0047]
3. Interpolation value calculation Interpolation values of the G component and B component are calculated. The interpolation value consists of two terms, average information and curvature information.
<Average information>: The same as in the first embodiment.
<Curvature information>
dR = (6 * Rctr-R030-R090-R150-R210-R270-R330) / 12 ... (23)
dG = ((G000 + G120 + G240)-(G060 + G180 + G300)) / 6 ... (24)
dB = ((B060 + B180 + B300)-(B000 + B120 + B240)) / 6 ... (25)
<Interpolated value>
Gctr = Gave + (dR + dG + dB) / 3 ... (26)
Bctr = Bave + (dR + dG + dB) / 3 ... (27)
[0048]
In the second embodiment, unlike the first embodiment, the average information is corrected with the curvature information of three colors. This is because, as disclosed in Japanese Patent Application Laid-Open No. 2001-245314 of the prior art, in an image including a chromatic portion and chromatic aberration, overcorrection only by dR is prevented, and the clearness in the gradation direction is maintained. However, JP 2001-245314 A relates to a Bayer array. Even when three colors of curvature information are used in this way, unlike the case of the Bayer arrangement, correction using curvature information having no directionality for any of the RGB three components is performed for the most all directions, as in the first embodiment. Experiments have revealed that the effect of increasing the tone clarity is high. In addition, since the delta arrangement is composed of R: G: B = 1: 1: 1, it was found that the curvature information of the three components should be used at 1: 1: 1. FIG. 9 illustrates coefficient values used in equations (23), (24), and (25).
[0049]
As described above, “interpolation processing of G and B components at the R pixel position” could be performed. “Interpolation processing of B and R components at the G pixel position” and “Interpolation processing of R and G components at the B position” are also obtained by cyclically replacing RGB, as in the first embodiment. The square processing can also be performed in the same manner as in the first embodiment.
[0050]
As described above, the restoration processing according to the second embodiment can discriminate and resolve the color boundary even for an image including the chromatic part, and can be sharpened without causing overcorrection in the gradation direction. Can be obtained. It is also strong against systems that include chromatic aberration.
[0051]
-Third embodiment-
In image processing of image data obtained with a Bayer array, the RGB signal is usually converted to YCbCr consisting of luminance and color difference to reduce the false color remaining in the image after interpolation processing, and the color difference low-pass on the Cb and Cr planes. Post processing is performed to remove the false color by applying a filter or a color difference median filter, and to return to the RGB color system. In addition, edge enhancement processing is performed on the luminance component Y plane in order to compensate for a decrease in sharpness due to an optical low-pass filter or the like. In the case of a delta arrangement, exactly the same post processing may be applied.
[0052]
However, the third embodiment shows a post-processing method suitable for the delta arrangement. Since the configuration of the electronic camera 1 of the third embodiment is the same as that of FIG. 1 of the first embodiment, the description thereof is omitted.
[0053]
FIG. 10 is a flowchart illustrating an overview of image processing performed by the image processing unit 11 in the third embodiment. It is assumed that interpolation processing is performed on a triangular lattice as in the first embodiment. FIG. 10 starts from input of an RGB color image after interpolation processing. That is, steps S1 to S5 in FIG. 2 of the first embodiment are finished, and then the flowchart in FIG. 10 is started.
[0054]
In step S11, RGB color image data after interpolation processing is input. In step S12, the RGB color system is converted to the YCrCgCb color system peculiar to the third embodiment. In step S13, a low-pass filter process is performed on the color difference surface (CrCgCb surface). In step S14, edge enhancement processing is performed on the luminance surface (Y surface). In step S15, the YCrCgCb color system is converted back to the original RGB color system at the stage where the false color on the color difference surface is removed. In step S16, the obtained RGB color image data is output. The RGB color image data output in step S16 is image data obtained on a triangular lattice.
[0055]
When the square processing is performed on the image data obtained on the triangular lattice, the processing in steps S6 and S7 in FIG. 2 is performed as in the first embodiment. Hereinafter, the details of the processing in steps S12 to S15 will be described.
[0056]
1. Color system conversion Conversion from RGB color system to YCrCgCb color system. However, the YCrCgCb color system is defined as represented by the formulas (28) to (31).
Y [i, j] = (R [i, j] + G [i, j] + B [i, j]) / 3 ... (28)
Cr [i, j] = R [i, j] -Y [i, j] ... (29)
Cg [i, j] = G [i, j] -Y [i, j] ... (30)
Cb [i, j] = B [i, j] -Y [i, j] ... (31)
[0057]
After the circular zone plate is imaged and interpolated on a triangular lattice, conversion is performed to treat RGB evenly in Y plane generation as shown in equation (28), with the hexagonal corner portion of FIG. 6 as the center. The generated color moire disappears from the luminance surface Y completely. As a result, all false color elements can be included in the color difference components Cr, Cg, and Cb. In addition, since the three types of color difference components are prepared and handled in the third embodiment instead of the two types of color difference components conventionally handled, all the false color elements unique to the delta arrangement can be extracted. .
[0058]
2. Color Difference Correction Here, as an example of color difference correction processing, an example is shown in which low-pass processing is performed on the color difference surface using Equation (32). When a color difference low-pass filter is applied on a triangular lattice (triangular arrangement), the false color reduction effect can be obtained with good compatibility in all directions. Further, median filter processing on a triangular lattice may be used. Here, [] represents the pixel position on the triangular lattice as viewed from the processing target pixel, as shown in FIG. FIG. 11 illustrates the coefficient values used in equation (32). Note that the low-pass filter is not limited to that shown here, and other filters may be used. 12 and 13 show other examples of the low-pass filter.
<Low pass processing>
tmp_Cr [center] = {9 * Cr [center]
+ 2 * (Cr [nearest000] + Cr [nearest120] + Cr [nearest240]
+ Cr [nearest180] + Cr [nearest300] + Cr [nearest060])
+ Cr [2nd000] + Cr [2nd120] + Cr [2nd240]
+ Cr [2nd180] + Cr [2nd300] + Cr [2nd060]} / 27 ... (32)
The same processing is performed for tmp_Cg and tmp_Cb.
<Substitution process>
Cr [i, j] = tmp_Cr [i, j] ... (33)
Cg [i, j] = tmp_Cg [i, j] ... (34)
Cb [i, j] = tmp_Cb [i, j] ... (35)
[0059]
3. Edge Enhancement Processing Next, if the edge enhancement processing for the luminance surface Y is necessary, the following processing is performed on the triangular lattice.
<Band pass processing>
Edge extraction is performed by Laplacian processing on a triangular lattice. FIG. 14 illustrates the coefficient values used for the Laplacian process of Equation (36). Note that the Laplacian is not limited to that shown here, and other types may be used. 15 and 16 show other examples of Laplacian.
YH [center] = {6 * Y [center]
-(Y [nearest000] + Y [nearest120] + Y [nearest240]
+ Y [nearest180] + Y [nearest300] + Y [nearest060])} / 12 ... (36)
<Edge enhancement processing>
Y [center] = Y [center] + K * YH [center] ... (37)
Here, K is a value that is greater than or equal to zero and is a parameter for adjusting the degree of edge enhancement.
[0060]
4). Color system conversion As described above, the color difference correction process is performed, and when the false color on the color difference surface is removed, the original RGB color system is restored.
R [i, j] = Cr [i, j] + Y [i, j] ... (38)
G [i, j] = Cg [i, j] + Y [i, j] ... (39)
B [i, j] = Cb [i, j] + Y [i, j] ... (40)
[0061]
In this way, the luminance resolution is saved in the Y component generated at a ratio of R: G: B = 1: 1: 1, and by introducing three equally treated color difference components Cr, Cg, Cb, Color difference correction with extremely high false color suppression capability is possible. Further, by performing color difference correction processing and luminance correction processing on a triangular lattice, correction processing suitable for the directionality of the delta arrangement can be performed.
[0062]
-Fourth embodiment-
In the first embodiment, the process of interpolating the color information of the color component missing in each pixel on the triangular lattice has been described. In the fourth embodiment, an example of image restoration processing of a system different from the interpolation processing in the RGB plane of the first embodiment is shown. This is a method in which a luminance component and a color difference component are directly generated from a delta arrangement without using interpolation processing in the RGB plane. Inheriting the usefulness of separation into one luminance plane and three chrominance planes conceptually shown in the third embodiment, the luminance plane is responsible for maximizing the achromatic luminance resolution, The three color difference planes are responsible for maximizing the color resolution of the three primary colors.
[0063]
Since the configuration of the electronic camera 1 of the fourth embodiment is the same as that of FIG. 1 of the first embodiment, the description thereof is omitted.
[0064]
(Image processing)
FIG. 17 is a diagram showing a concept of generating a luminance plane (Y) and three color difference planes (Cgb, Cbr, Crg) directly from the delta plane of the delta arrangement, and then converting it to the original RGB color system. It is. FIG. 18 is a flowchart illustrating an overview of image processing performed by the image processing unit 11 in the fourth embodiment.
[0065]
In step S21, an image obtained by the image sensor 21 having a delta arrangement is input. In step S22, the similarity is calculated. In step S23, the similarity is determined based on the similarity obtained in step S22. In step S24, a luminance plane (Y0 plane) is generated based on the similarity determination result obtained in step S23 and the image data of the delta arrangement obtained in step S21. In step S25, correction processing is performed on the luminance surface (Y0 surface) obtained in step S24.
[0066]
On the other hand, in step S26, color difference components Cgb, Cbr, and Crg are generated based on the similarity determination result obtained in step S23 and the delta arrangement image data obtained in step S21. In step S26, not all color difference components Cgb, Cbr, and Crg have been generated yet for all pixels. In step S27, interpolation processing is performed on the color difference components that have not been generated based on the surrounding color difference components. As a result, Cgb, Cbr, and Crg color difference planes are completed.
[0067]
In step S28, the generated Y, Cgb, Cbr, Crg color system is converted to the RGB color system. In step S29, the converted RGB color image data is output. Steps S21 to S29 are all processing on a triangular lattice. Therefore, the RGB color image data output in step S29 is triangular lattice image data. Details of these processes will be described below. If square processing is necessary, processing similar to steps S6 and S7 in FIG. 2 of the first embodiment is performed.
[0068]
1. Calculation of similarity First, the similarity is calculated. Here, the degree of similarity obtained by an arbitrary method may be used. However, use as accurate as possible. For example, the similarity between different colors shown in the first embodiment, the similarity between the same colors shown in the second embodiment, a combination of them, a color index, etc. It may be used by switching the similarity between the same colors.
[0069]
2. Similarity determination Determined in the same manner as in the first embodiment.
[0070]
3. Calculation of restoration value 1) Generation of luminance component First, a luminance plane is generated by weighted addition so that the delta plane is always R: G: B = 1: 1: 1 with a positive coefficient that changes according to the direction similarity. . The range of the weighted addition is the same as in the first embodiment, and the G component and B component are taken up to the second adjacent pixel. That is, when the equations (8) and (9) are rewritten once again using the definition equations of equations (10) to (15) of the first embodiment, equations (41) and (42) are obtained.
<G> _2nd = Gave = w000 * Gave000 + w120 * Gave120 + w240 * Gave240 ... (41)
<B> _2nd = Bave = w000 * Bave000 + w120 * Bave120 + w240 * Bave240 ... (42)
Using these and the central R component, the luminance component Y0 is generated by the equation (43).
Y0 = Rctr + <G> _2nd + <B> _2nd (43)
[0071]
Since the luminance component generated in this way is generated while always including the central pixel with a constant color component ratio using a positive directional weighting factor, the potential of gradation sharpness is extremely high, and the spatial resolving power is high. A high image is obtained, and an image that is connected to peripheral pixels very smoothly without being affected by chromatic aberration is obtained. For example, when the similarity between different colors is adopted as the similarity for an achromatic circular zone plate, the spatial resolution reaches the limit resolution of FIG. 6 as in the first embodiment.
[0072]
2) Luminance component correction Since the above-described luminance surface is generated using only positive coefficients, Laplacian correction processing is performed in order to extract the potential of gradation clarity contained therein. Luminance plane Y0 is designed to be connected to the surrounding pixels very smoothly in consideration of directionality, so there is no need to calculate correction terms that need to be newly calculated according to directionality. Thus, a single process using a fixed bandpass filter is sufficient. There are several methods for taking the Laplacian in the triangular arrangement as shown in FIGS. 14 to 16 of the third embodiment. However, if the degree of optimization is increased slightly, the luminance plane Y0 can only be generated by collecting the G and B components in the directions of 0, 120, and 240 degrees, so to compensate for the gap direction, Then, a case where correction is performed with the Laplacian of FIG. 15 in directions of 30 degrees, 150 degrees, and 270 degrees independent of that (formula (44)) is shown. Let Y be the corrected luminance component.
[0073]
<Band pass processing>
YH [center] = {6 * Y0 [center]
-(Y0 [2nd030] + Y0 [2nd150] + Y0 [2nd270]
+ Y0 [2nd210] + Y0 [2nd330] + Y0 [2nd090])} / 12 ... (44)
<Correction process>
Y [center] = Y [center] + k * YH [center] ... (45)
Here, k is a positive value and is normally 1. However, by setting it to a value greater than 1, the edge enhancement processing as shown in the third embodiment can also be performed here.
[0074]
3) Color difference component generation Unlike the definition of the third embodiment, the three color difference planes are generated directly from the delta plane independently of the luminance plane Y. Three color difference components Cgb, Cbr, and Crg are obtained. However, Cgb = GB, Cbr = BR, and Crg = RG are defined.
[0075]
First, Crg and Cbr are obtained at the R position. The difference value between the R component of the central pixel and the G component or B component of the peripheral pixels is calculated in consideration of the directionality.
Crg = Rctr-{<G> _2nd + dG} ... (46)
Cbr = {<B> _2nd + dB} -Rctr ... (47)
Here, dG and dB are the same as those defined in the equations (24) and (25) of the second embodiment, and <G> _2nd and <B> _2nd are the same as the equations (41) and (42). It is. Also in the generation of the color difference component, as in the first embodiment, the average information including up to the second adjacent pixel is calculated, and the consistency with the luminance component is taken, so that the directions in the directions of 30 degrees, 150 degrees, and 270 degrees are obtained. Increases resolution. DG and dB are not necessarily required, but they are added because they have the effect of increasing color resolution and vividness.
[0076]
The color difference components Cgb and Crg at the G position and the color difference components Cbr and Cgb at the B position are obtained in the same manner. At this point, the Crg and Cbr components are obtained at the R position, and the Cgb component is obtained at the nearest pixel. FIG. 19 is a diagram showing this state.
[0077]
Next, a Cgb component is obtained at the R position from Equation (48) using the Cgb component of the pixels around the R position (interpolation process). At this time, the calculation is performed using the direction determination result obtained at the R position.
Cgb [center] = w000 * (Cgb [nearest000] + Cgb [nearest180]) / 2
+ w120 * (Cgb [nearest120] + Cgb [nearest300]) / 2
+ w240 * (Cgb [nearest240] + Cgb [nearest060]) / 2 ... (48)
[0078]
As described above, four components of YCgbCbrCrg are obtained for all pixels. If necessary, the color difference surfaces Cgb, Cbr, and Crg may be subjected to correction processing such as a color difference low-pass filter similar to that of the third embodiment to suppress false color.
[0079]
4) Color system conversion YCgbCbrCrg generated for each pixel is converted to the RGB color system. Y = (R + G + B) / 3, Cgb = GB, Cbr = BR, and Crg = RG satisfy the relationship of Cgb + Cbr + Crg = 0 instead of the independent four formulas. In other words, the conversion method is not unique because it is a 4-to-3 conversion, but in order to suppress color moire and maximize the luminance resolution and color resolution, all the conversion formulas for R, G, B are all Y, Cgb, Cbr. , Crg components are included, and the mutual moire canceling effect is used. In this way, all of the maximum performances generated by the respective roles of the Y, Cgb, Cbr, and Crg planes can be reflected in each of R, G, and B.
R [i, j] = (9 * Y [i, j] + Cgb [i, j] -2 * Cbr [i, j] + 4 * Crg [i, j]) / 9 ... (49)
G [i, j] = (9 * Y [i, j] + 4 * Cgb [i, j] + Cbr [i, j] -2 * Crg [i, j]) / 9 ... (50)
B [i, j] = (9 * Y [i, j] -2 * Cgb [i, j] + 4 * Cbr [i, j] + Crg [i, j]) / 9 ... (51)
[0080]
As described above, the image restoration method according to the fourth embodiment has extremely high gradation sharpness and excellent luminance resolution performance and color resolution performance in the spatial direction at the same time. Even if it demonstrates the effect of being strong. When square processing is necessary, it can be performed in the same manner as in the first embodiment.
[0081]
-Fifth embodiment-
In the second embodiment, the similarity is determined by obtaining the similarity between the same colors. At this time, only the similarities in the 0 degree, 120 degree, and 240 degree directions were obtained. However, in the fifth embodiment, similarities in directions of 30 degrees, 150 degrees, and 270 degrees are also obtained. Since the configuration of the electronic camera 1 of the fifth embodiment is the same as that of FIG. 1 of the first embodiment, the description thereof is omitted.
[0082]
A case where G and B components are interpolated at the R pixel position will be mainly described. Reference is also made to FIG. 8 of the second embodiment. In the R pixel position, there are three adjacent G components adjacent to the positions indicating 0, 120, and 240 degrees, and the second adjacent pixels indicate 60 degrees, 180 degrees, and 300 degrees separated by two pixels. There are three in position. Similarly, for the B component, there are three nearest neighbor pixels at positions indicating 60 degrees, 180 degrees, and 300 degrees, and the second adjacent pixels are positions indicating 0 degrees, 120 degrees, and 240 degrees separated by two pixels. There are three. In addition, the R component nearest neighbor pixels are six at positions indicating 30 degrees, 90 degrees, 150 degrees, 210 degrees, 270 degrees, and 330 degrees separated by two pixels, and the second adjacent pixels are separated by three pixels. There are six positions that indicate degrees, 60 degrees, 120 degrees, 180 degrees, 240 degrees, and 300 degrees.
[0083]
1. Calculation of similarity 1) Calculation of similarity between three pixels at the same color
Similarities C000, C120, and C240 in the directions of 0 degree, 120 degrees, and 240 degrees are calculated. The similarity between same colors formed between the same color components in these directions cannot be defined shorter than the interval of three pixels.
C000 = {(| R000-Rctr | + | R180-Rctr |) / 2 + | G000-G180 | + | B000-B180 |} / 3 ... (52)
C120 = {(| R120-Rctr | + | R300-Rctr |) / 2 + | G120-G300 | + | B120-B300 |} / 3 ... (53)
C240 = {(| R240-Rctr | + | R060-Rctr |) / 2 + | G240-G060 | + | B240-B060 |} / 3 ... (54)
[0084]
The similarity between the same colors defined in this way is examined for the directionality that coincides with the direction in which the missing G component and B component exist at the R pixel position, so the horizontal line in the chromatic portion and the 120 degree direction It is thought that it has the ability to spatially resolve the image structure in the 240 degree direction. However, it should be noted that, unlike the same-color similarity between two pixels in the Bayer array, it is information between pixels that are very far from the three-pixel interval.
[0085]
2) Calculation of similarity between two pixels and same color Next, similarities C030, C150, and C270 in directions of 30 degrees, 150 degrees, and 270 degrees are calculated. Unlike the 0 degree, 120 degree, and 240 degree directions, the similarity between the same colors in these directions can be defined by a shorter two-pixel interval.
C030 = {(| R030-Rctr | + | R210-Rctr |) / 2 + | G240-G000 | + | B060-B180 |} / 3 ... (55)
C150 = {(| R150-Rctr | + | R330-Rctr |) / 2 + | G000-G120 | + | B180-B300 |} / 3 ... (56)
C270 = {(| R270-Rctr | + | R090-Rctr |) / 2 + | G120-G240 | + | B300-B060 |} / 3 ... (57)
However, since the similarity in the direction that does not coincide with the direction in which the G component and the B component that are missing at the R pixel position exist is examined, a technique is required to effectively use them.
[0086]
3) Peripheral addition of similarities Further, the accuracy of similarity is increased by performing peripheral addition for each of the above-described similarities and considering continuity with surrounding pixels. Here, peripheral addition is performed for the R pixel position for each of the six similarities obtained for each pixel. The similarity with marginal addition is written in lower case. However, [] represents a pixel position on the delta array as viewed from the processing target pixel. Expression (58) is the same as Expression (4) in the first embodiment.
c000 = {6 * C000 [Rctr]
+ C000 [R030] + C000 [R150] + C000 [R270]
+ C000 [R210] + C000 [R300] + C000 [R090]} / 12 ... (58)
c120, c240, c030, c150, c270 are obtained in the same manner.
FIG. 26 shows the azimuth relationship of the similarity described above.
[0087]
2. Similarity determination The smaller the degree of similarity described above, the greater the similarity in that direction. However, it is meaningful to determine the similarity in the direction in which the G component and the B component that do not exist in the processing target pixel exist, that is, the directions of 0 degrees, 120 degrees, and 240 degrees, and 30 degrees, 150 degrees, and 270. Even if the similarity in the degree direction can be determined, it does not make much sense. Therefore, the first possibility is to continuously determine the strength of similarity in the 0, 120, and 240 degrees directions using the reciprocal ratio of the degrees of similarity in the 0, 120, and 240 degrees directions. It is. That is, it is determined by (1 / c000) :( 1 / c120) :( 1 / c240).
[0088]
For example, since the similarity c000 has the ability to resolve the chromatic color horizontal line, the ky axis direction of the frequency reproduction range of each RGB color component in the delta arrangement of FIG. 20 can be extended to the limit resolution. In other words, such a determination method can extend the color resolution of a chromatic image to the limit resolution of the hexagonal vertex in FIG. However, since it is a similarity between long distances of 3 pixel intervals, the directionality cannot be discriminated due to the influence of the aliasing frequency component at an angle between them, especially in the directions of 30, 150, and 270 degrees. On the other hand, the most adverse effect is caused, and only a color resolving power that allows the vicinity of the midpoint of each side of the hexagon of FIG. 20 to be obtained can be exhibited.
[0089]
Therefore, in order to prevent such an adverse effect of the long distance correlation, the similarity c030, c150, c270 of the short distance correlation is effectively utilized. However, taking the reciprocal only makes it possible to determine the similarity in the directions of 30 degrees, 150 degrees, and 270 degrees, so in order to convert it into similarities in the directions of 0 degrees, 120 degrees, and 240 degrees, In other words, the similarity value itself is interpreted as representing the similarity in the directions orthogonal to the directions of 30 degrees, 150 degrees, and 270 degrees, that is, 120 degrees, 240 degrees, and 0 degrees. Therefore, the similarity in the 0 degree, 120 degree, and 240 degree directions is determined by the following ratio.
(c270 / c000) :( c030 / c120) :( c150 / c240)
[0090]
When the similarity in the 0 degree, 120 degree, and 240 degree directions is expressed as weighting coefficients w000, w120, and w240 normalized by 1,
w000 = (c120 * c240 * c270 + Th)
/ (c120 * c240 * c270 + c240 * c000 * c030 + c000 * c120 * c150 + 3 * Th) ... (59)
w120 = (c240 * c000 * c030 + Th)
/ (c120 * c240 * c270 + c240 * c000 * c030 + c000 * c120 * c150 + 3 * Th) ... (60)
w240 = (c000 * c120 * c150 + Th)
/ (c120 * c240 * c270 + c240 * c000 * c030 + c000 * c120 * c150 + 3 * Th) ... (61)
It is obtained by. However, the constant Th is a term for preventing divergence and takes a positive value. Usually, Th = 1 may be set, but this threshold value may be increased for a noisy image such as a high-sensitivity photographed image.
[0091]
The 0 degree direction and the 270 degree direction, the 120 degree direction and the 30 degree direction, and the 240 degree direction and the 150 degree direction are orthogonal to each other. This orthogonal relationship is expressed as a 0 degree direction ⊥ 270 degree direction, a 120 degree direction ⊥ 30 degree direction, and a 240 degree direction ⊥ 150 degree direction.
[0092]
Thus, the similarity determined continuously using the similarity in the six directions has a spatial resolving power that accurately reproduces all of the hexagons in FIG. 20 with respect to the chromatic image. Further, the spatial resolving power based on the similarity between the same colors can always be achieved without being affected by the chromatic aberration included in the optical system because the similarity is observed between the same color components.
[0093]
3. Interpolation value calculation Interpolation values are obtained in the same manner as in the second embodiment.
[0094]
As described above, in the fifth embodiment, it is possible to extract all the RGB single-color spatial color resolving power originally possessed by the delta arrangement for any image. In addition, a clear image can be restored in the gradation direction, and strong performance is exhibited even for a system including chromatic aberration.
[0095]
Note that the similarity calculation and similarity determination of the fifth embodiment can be applied to the similarity calculation and similarity determination of the fourth embodiment. Such an image restoration method exhibits extremely high gradation sharpness and an effect of being strong against chromatic aberration while simultaneously achieving excellent luminance resolution performance and color resolution performance in the spatial direction. In particular, it is possible to derive the highest color resolution performance of the delta arrangement while achieving a clear gradation. In normal image processing, there is a trade-off in which edge enhancement processing is performed to improve gradation sharpness, and the color resolving power decreases accordingly, fading, or stereoscopic effect is impaired. By introducing four-component restoration values, the conflicting problems can be solved simultaneously.
[0096]
-Sixth embodiment-
In a Bayer array, in order to reduce false colors remaining in an image after interpolation processing, RGB signals are usually converted to YCbCr consisting of luminance and color difference, and a color difference low-pass filter is applied to the Cb and Cr planes, or a color difference median filter is used. Is applied to remove false colors and return to the RGB color system. Even in the case of the delta arrangement, if the Nyquist frequency cannot be lowered completely with an optical low-pass filter, an appropriate false color reduction process is required to improve the appearance. In the sixth embodiment, a post-processing method that does not impair the color resolution performance that is an excellent feature of the delta arrangement as much as possible is shown. Since the configuration of the electronic camera 1 of the sixth embodiment is the same as that of FIG. 1 of the first embodiment, the description thereof is omitted.
[0097]
FIG. 21 is a flowchart illustrating an overview of image processing performed by the image processing unit 11 in the sixth embodiment. As in the first, second, fourth, and fifth embodiments, it is assumed that interpolation processing is performed on a triangular lattice. FIG. 21 starts from input of an RGB color image after interpolation processing. For example, steps S1 to S5 in FIG. 2 of the first embodiment are finished, and then the flowchart in FIG. 21 is started.
[0098]
In step S31, RGB color image data after interpolation processing is input. In step S32, the RGB color system is converted to the YCgbCbrCrg color system unique to the sixth embodiment. In step S33, a color determination image is generated. In step S34, a color index is calculated using the color determination image generated in step S33. In step S35, a color determination of whether the color saturation is low or high is performed based on the color index in step S34.
[0099]
In step S36, color difference correction is performed by switching the low-pass filter to be used based on the color determination result in step S35. The color difference data to be corrected is generated in step S32. In step S37, the YCgbCbrCrg color system is converted back to the original RGB color system when the false color on the color difference surface is removed. In step S38, the obtained RGB color image data is output. The RGB color image data output in step S38 is image data obtained on a triangular lattice.
[0100]
When the square processing is performed on the image data obtained on the triangular lattice, the processing in steps S6 and S7 in FIG. 2 is performed as in the first embodiment. Hereinafter, the details of the processing in steps S32 to S37 described above will be described.
[0101]
1. Color system conversion
Convert from RGB color system to YCgbCbrCrg color system. However, the YCgbCbrCrg color system is defined by the following equations (62) to (65).
Y [i, j] = (R [i, j] + G [i, j] + B [i, j]) / 3 ... (62)
Cgb [i, j] = G [i, j] -B [i, j] ... (63)
Cbr [i, j] = B [i, j] -R [i, j] ... (64)
Crg [i, j] = R [i, j] -G [i, j] ... (65)
In this way, when conversion is performed to treat RGB evenly, for example, in the circular zone plate, the color moire that occurs around the corner of the hexagon in FIG. 6 disappears completely from the luminance plane Y, and all False color elements can be included in the color difference components Cgb, Cbr, and Crg.
[0102]
2. Color Evaluation 1) Generation of Color Determination Image In order to evaluate the color by distinguishing the false color in the achromatic part as much as possible from the chromatic part, the false color is temporarily reduced by applying a strong color difference low-pass filter on the entire surface. However, since this is a temporary color determination image, it does not affect the actual image.
TCgb [center] = {9 * Cgb [center]
+ 2 * (Cgb [nearest000] + Cgb [nearest120] + Cgb [nearest240]
+ Cgb [nearest180] + Cgb [nearest300] + Cgb [nearest060])
+ Cgb [2nd000] + Cgb [2nd120] + Cgb [2nd240]
+ Cgb [2nd180] + Cgb [2nd300] + Cgb [2nd060]} / 27 ... (66)
The same applies to TCbr and TCrg.
[0103]
2) Calculation of color index Next, the color index Cdiff is calculated using the image for color determination with reduced false colors, and color evaluation is performed in units of pixels.
Cdiff [i, j] = (| Cgb [i, j] | + | Cbr [i, j] | + | Crg [i, j] |) / 3 ... (67)
[0104]
3) Color determination The continuous color index Cdiff is determined as a threshold and converted to a discrete color index BW.
if Cdiff [i, j] ≦ Th then BW [i, j] = 'a' (low saturation)
else then BW [i, j] = 'c' (High chroma)
Here, the threshold value Th is preferably set to about 30 in the case of 256 gradations.
[0105]
3. Since the image can be divided into two regions by adaptive color difference correction color determination, the false color of the achromatic color portion is strongly erased, while the color difference correction processing is applied to preserve the color resolution of the chromatic color portion as much as possible. Here, the low-pass processing of Equation (68) and Equation (69) is performed, but median filter processing may be used. [] Represents the pixel position on the triangular lattice as viewed from the processing target pixel. The low pass filter of equation (68) uses the coefficient values shown in FIG. 11 of the third embodiment, and the low pass filter of equation (69) uses the coefficient values shown in FIG.
[0106]
<Low pass processing>
if BW [center] = 'a'
tmp_Cgb [center] = {9 * Cgb [center]
+ 2 * (Cgb [nearest000] + Cgb [nearest120] + Cgb [nearest240]
+ Cgb [nearest180] + Cgb [nearest300] + Cgb [nearest060])
+ Cgb [2nd000] + Cgb [2nd120] + Cgb [2nd240]
+ Cgb [2nd180] + Cgb [2nd300] + Cgb [2nd060]} / 27 ... (68)
Calculate similarly for tmp_Cbr and tmp_Crg.
if BW [center] = 'c'
tmp_Cgb [center] = {6 * Cgb [center]
+ Cgb [nearest000] + Cgb [nearest120] + Cgb [nearest240]
+ Cgb [nearest180] + Cgb [nearest300] + Cgb [nearest060]} / 12 ... (69)
Calculate similarly for tmp_Cbr and tmp_Crg.
[0107]
<Substitution process>
Cgb [i, j] = tmp_Cgb [i, j]
Cbr [i, j] = tmp_Cbr [i, j]
Crg [i, j] = tmp_Crg [i, j]
[0108]
4). Color system conversion processing At the stage where the false color on the color difference surface is removed, the original RGB color system is restored using the same equations (70) to (72) as in the fourth embodiment.
R [i, j] = (9 * Y [i, j] + Cgb [i, j] -2 * Cbr [i, j] + 4 * Crg [i, j]) / 9 ... (70)
G [i, j] = (9 * Y [i, j] + 4 * Cgb [i, j] + Cbr [i, j] -2 * Crg [i, j]) / 9 ... (71)
B [i, j] = (9 * Y [i, j] -2 * Cgb [i, j] + 4 * Cbr [i, j] + Crg [i, j]) / 9 ... (72)
[0109]
In this way, in the sixth embodiment, it is possible to reduce the false color of the entire image while maintaining the color resolution performance of the delta arrangement as much as possible. If this post-processing is incorporated into the image data generated in the fourth embodiment, the YCgbCbrCrg color system is already in the state, so the first color system conversion is unnecessary and the last color system is not necessary. Before performing system conversion, 2.3. It is only necessary to add the process.
[0110]
It should be noted that the image restoration methods described in the first to sixth embodiments can be applied to any level regardless of whether the input image of the delta arrangement remains in the linear gradation of the output of the image sensor or has a gamma-corrected gradation. It can also be applied to those with tonal characteristics. Although there is no way to deal with input images that have already undergone gamma correction in 8 bits, today's digital still camera imaging signals are generally 12-bit linear gradation output. The relationship between the image restoration process suitable for such a case (including the RGB system as in the first and second embodiments and the YCCC system as in the fourth embodiment) and the gradation processing is here. Shown in FIG. 27 is a flowchart showing the processing.
1) Linear gradation delta arrangement data input (step S41)
2) Gamma correction processing (step S42)
3) Image restoration processing (step S43)
4) Inverse gamma correction process (step S44)
5) User gamma correction process (step S45)
[0111]
The user gamma correction process is a process of converting a linear gradation to an 8-bit gradation suitable for display output, that is, a process of compressing the dynamic range of an image within the range of an output display device. Independently of this, when the gradation is once converted into a certain gamma space and the image restoration processing corresponding to the first to sixth embodiments is performed, a better restoration result can be obtained. The following gradation conversion methods are available.
<Gamma correction processing>
Input signal x (0 ≤ x ≤ xmax), output signal y (0 ≤ x ≤ ymax), input image is delta plane
y = ymax√ (x / xmax)
If the input signal is 12 bits, xmax = 4095, and the output signal may be set to 16 bits ymax = 65535, for example.
<Reverse gamma correction processing>
Input signal y (0 ≤ x ≤ ymax), output signal x (0 ≤ x ≤ xmax), input image is RGB
x = xmax (y / ymax) ²
[0112]
When the image restoration process is performed in the square root type gamma space as described above, the following advantages can be obtained simultaneously from a viewpoint different from the technique of the image restoration process.
1) Color borders can be sharpened (for example, suppression of red / white border color blur and suppression of black borders)
2) Suppression of false colors around bright spots (very bright areas) 3) Improved direction determination accuracy
1) and 2) are effects produced by involving the “interpolation value calculation unit” in the RGB method of image restoration processing and the “restored value calculation unit” in the Y method. Further, 3) is an effect generated by involving the “similarity calculation unit” and the “similarity determination unit” of the image restoration process. In other words, the input signal x contains an error of quantum fluctuation dx = k√x (k is a constant determined by the ISO sensitivity). Since uniform error dy = constant can be handled over 0 ≦ y ≦ ymax, the direction determination accuracy is improved. This technique can be applied not only to the delta arrangement but also to the interpolation process of the Bayer arrangement and various other filter arrangements. In the above embodiment, since the delta arrangement originally has a higher color resolution performance than the Bayer arrangement, this gradation conversion process can be performed before and after the image restoration process to produce even better color clarity. Is possible.
[0114]
In the first embodiment, the example in which the image data subjected to the interpolation processing on the triangular lattice is squared has been described. In other embodiments, it is indicated that when a square process is necessary, the process is performed. However, the image data itself that has undergone interpolation processing or color system conversion on the triangular lattice may be used as it is.
[0115]
In the third embodiment, an example in which correction processing is performed on RGB color image data generated in the same manner as in the first embodiment has been described. However, the present invention is not necessarily limited to this content. The present invention can be similarly applied to RGB color image data generated in other embodiments. Thus, in the said 1st-6th embodiment, it is possible to combine suitably, respectively. That is, in the first to sixth embodiments, similar direction determination processing, interpolation processing or color difference plane direct generation processing, post-processing such as correction, and square processing are described. It is possible to realize an optimum image processing method and processing apparatus by appropriately combining the processes in the embodiments.
[0116]
In the above embodiment, the description has been made on the assumption that a single-plate image sensor is used, but it is not necessary to limit to this content. The present invention can also be applied to a two-plate image sensor. In the case of the two-plate type, for example, one color component is missing in each pixel. However, the content of the above embodiment can be applied to the interpolation processing of one missing color component, and if necessary, square The same processing can be performed. Further, the method of directly generating the luminance component and the color difference component from the delta arrangement without using the interpolation process of the fourth embodiment can be similarly applied to a two-plate type image pickup device.
[0117]
In the above embodiment, various calculation formulas are shown for determining the similarity, but it is not necessarily limited to the contents shown in the embodiment. You may make it determine similarity by another suitable calculation formula. Also, although various calculation formulas are shown in the calculation of luminance information, it is not necessarily limited to the contents shown in the embodiment. Luminance information may be generated by another appropriate calculation formula.
[0118]
In the above-described embodiment, an example of a low-pass filter for color difference correction and the like and a band-pass filter for edge enhancement are shown, but it is not necessarily limited to the contents shown in the embodiment. A low-pass filter or a band-pass filter having another configuration may be used.
[0119]
In the above embodiment, an example of an electronic camera is shown, but it is not necessarily limited to this content. It may be a video camera that captures moving images, a personal computer with an image sensor, a mobile phone, or the like. That is, the present invention can be applied to any device that generates color image data using an image sensor.
[0120]
When applied to a personal computer or the like, a program related to the above-described processing can be provided through a recording medium such as a CD-ROM or a data signal such as the Internet. FIG. 23 is a diagram showing this state. The personal computer 100 receives a program via the CD-ROM 104. The personal computer 100 has a connection function with the communication line 101. The computer 102 is a server computer that provides the program, and stores the program in a recording medium such as the hard disk 103. The communication line 101 is a communication line such as the Internet or personal computer communication, or a dedicated communication line. The computer 102 reads the program using the hard disk 103 and transmits the program to the personal computer 100 via the communication line 101. That is, the program is transmitted as a data signal on a carrier wave via the communication line 101. Thus, the program can be supplied as a computer-readable computer program product in various forms such as a recording medium and a carrier wave.
[0121]
【The invention's effect】
When interpolating a first color component to a pixel lacking the first color component, the present invention interpolates using color information of a region including a pixel in which the first color component is second closest to the interpolation target pixel Therefore, for example, the images of the boundary lines in the directions of 30 degrees, 150 degrees, and 270 degrees (vertical lines) in FIG. 3 are dramatically improved. That is, the spatial resolution in the directions of 30, 150, and 270 degrees is improved.
[Brief description of the drawings]
FIG. 1 is a functional block diagram of an electronic camera according to a first embodiment.
FIG. 2 is a flowchart showing an outline of image processing performed by an image processing unit in the first embodiment;
FIG. 3 is a diagram illustrating a positional relationship of pixels obtained by an image sensor with a delta arrangement.
FIG. 4 is a diagram illustrating coefficients used for peripheral addition.
FIG. 5 is a diagram showing coefficient values used when obtaining curvature information dR.
FIG. 6 is a diagram illustrating an achromatic spatial frequency reproduction region in a delta arrangement.
FIG. 7 is a diagram illustrating coefficient values used for one-dimensional displacement processing.
FIG. 8 is a diagram illustrating pixel positions used in the calculation according to the second embodiment.
FIG. 9 is a diagram illustrating coefficient values used when obtaining curvature information dR, dG, and dB.
FIG. 10 is a flowchart illustrating an outline of image processing performed by an image processing unit in the third embodiment;
FIG. 11 is a diagram illustrating coefficient values of a low-pass filter.
FIG. 12 is a diagram illustrating coefficient values of other low-pass filters.
FIG. 13 is a diagram illustrating coefficient values of other low-pass filters.
FIG. 14 is a diagram illustrating coefficient values of Laplacian.
FIG. 15 is a diagram illustrating coefficient values of other Laplacians.
FIG. 16 is a diagram illustrating coefficient values of other Laplacians.
FIG. 17 is a diagram showing a concept of generating a luminance plane (Y) and three color difference planes (Cgb, Cbr, Crg) directly from a delta plane in a delta arrangement, and then converting them to the original RGB color system. It is.
FIG. 18 is a flowchart illustrating an outline of image processing performed by an image processing unit in the fourth embodiment;
FIG. 19 is a diagram illustrating a state in which a Crg component and a Cbr component are obtained at the R position and a Cgb component is obtained in the nearest pixel.
FIG. 20 is a diagram illustrating a spatial frequency reproduction region of RGB color components in a delta arrangement.
FIG. 21 is a flowchart illustrating an outline of image processing performed by an image processing unit in the sixth embodiment;
FIG. 22 is a diagram defining adjacent pixels.
FIG. 23 is a diagram illustrating a state in which a program is provided through a recording medium such as a CD-ROM or a data signal such as the Internet.
FIG. 24 is a diagram showing a Bayer array, a delta array, and a honeycomb array of RGB color filters.
FIG. 25 is a diagram showing a concept of processing for interpolating image data obtained by a delta arrangement on a triangular lattice and restoring it to square lattice data.
FIG. 26 is a diagram showing a azimuth relationship of similarity in the fifth embodiment.
FIG. 27 is a flowchart showing image restoration processing and gradation processing.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Electronic camera 10 A / D conversion part 11 Image processing part 12 Control part 13 Memory 14 Compression / decompression part 15 Display image generation part 16 Memory card 17 Memory card interface part 18 PC (personal computer)
19 External interface unit 20 Imaging optical system 21 Image sensor 22 Analog signal processing unit 23 Timing control unit 24 Operation units 25 and 26 Monitor 27 Printer 28 CD-ROM

Claims (14)

Image acquisition procedure for acquiring a first image in which a plurality of pixels represented by first to nth color components (n ≧ 2) and having color information of one color component per pixel are arranged in a triangular lattice shape When,
An interpolation procedure for interpolating the first color component to the pixel lacking the first color component using the acquired color information of the first image;
An image processing method comprising: an output procedure for outputting a second image based on the color information of the first image and the interpolated color information,
The interpolation procedure obtains average information of the first color component using color information of an area including a pixel in which the first color component is second closest to the interpolation target pixel of the first image, and There line interpolation,
A similarity determination procedure for determining the strength of similarity in at least three directions;
The image processing method characterized in that the interpolation procedure obtains average information of the first color component in accordance with the strength of similarity determined in the similarity determination procedure . An image acquisition procedure for acquiring a first image represented by a plurality of color components and having a plurality of pixels having color information of one color component per pixel arranged in a triangular lattice pattern;
A color information generation procedure for generating color information of a color component different from the color information of the first image by weighted addition of the acquired color information of the first image with a variable coefficient value of zero or more;
An image processing method comprising an output procedure for outputting a second image using the generated color information,
The color information generation procedure weights and adds color information in a region including a pixel in which a color component different from the color component of the pixel is second closest to the processing target pixel of the first image. An image processing method. The image processing method according to claim 2 ,
A similarity determination procedure for determining the strength of similarity in at least three directions;
The color information generating procedure makes the coefficient value of the weighted addition variable according to the strength of similarity determined in the similarity determining procedure. The image processing method according to claim 2 or 3 ,
When the first image is represented by first to third color components and the pixel having the first color component of the first image is a processing target pixel,
The color information generation procedure weights and adds color information in a region including the processing target pixel, a pixel in which the second color component is second closest, and a pixel in which the third color component is second closest. A featured image processing method. In the image processing method according to any one of claims 2 to 4 ,
After the color information generation procedure and before the output procedure, filter processing including color information of a color component different from the color information of the first image generated in the color information generation procedure, using predetermined fixed filter coefficients An image processing method, further comprising: a correction procedure for correcting. The image processing method according to claim 5 ,
An image processing method, wherein the filter coefficient includes positive and negative values. Image acquisition procedure for acquiring a first image in which a plurality of pixels represented by first to nth color components (n ≧ 2) and having color information of one color component per pixel are arranged in a triangular lattice shape When,
A color difference generation procedure for generating color information of a color difference component between the first color component and the second color component using the color information of the first image;
An image processing method comprising an output procedure for outputting a second image using color information of the generated color difference component,
The color difference generation procedure generates color information of the color difference component by using color information of a pixel having at least a second color component closest to the pixel having the first color component of the first image. An image processing method. The image processing method according to claim 7 .
The color difference generation procedure is performed on a processing target pixel having a first color component of the first image.
1) based on the color information of the first color component of the pixel and 2) the average information of the color information of the second color component in the region including the pixel in which the second color component is second closest to the pixel. An image processing method comprising generating color information of the color difference component. The image processing method according to claim 7 or 8 ,
The color difference generation procedure further includes generating color information of the color difference component based on curvature information of the second color component regarding the processing target pixel. The image processing method according to any one of claims 7 to 9 ,
A similarity determination procedure for determining the strength of similarity in at least three directions;
The image processing method according to claim 1, wherein the color difference generation procedure generates color information of the color difference component in accordance with the strength of the similarity. The image processing method according to any one of claims 1-10,
In the image processing method, the output procedure outputs the second image at the same pixel position as the first image. An image processing program for causing a computer to execute the procedure of the image processing method according to any one of claims 1 to 11 . A computer-readable recording medium on which the image processing program according to claim 12 is recorded. An image processing apparatus equipped with the image processing program according to claim 12 .

Images