WO2009081709A1

WO2009081709A1 - Image processing apparatus, image processing method, and image processing program

Info

Publication number: WO2009081709A1
Application number: PCT/JP2008/071996
Authority: WO
Inventors: Hideya Aragaki
Original assignee: Olympus Corporation
Priority date: 2007-12-25
Filing date: 2008-11-27
Publication date: 2009-07-02

Abstract

An image processing apparatus comprises an image decomposing part that decomposes an original image signal into images representative of a plurality of components including both a first component image, which is a framework component indicative of a general structure of an image including the edges of the image and the flat areas defined by those edges, and a second component image obtained by calculations based on the original image signal and also based on the first component image; a direction determining part that determines the direction of each edge component for the first component image; and a noise reducing part that reduces noises in accordance with the directions of the edge components for the second component image.

Description

Description w & ^. ^ M and drawing ^ M program mv ^

The present invention relates to, and in particular, to a holding device that eliminates image noise. Background leakage

A noise removal device that removes a noise signal contained in an image signal is widely used by a mouth-pass filter. However, there is a problem that if the noise is reduced by the smoothing key by the low-pass filter, the edge component included in the image signal is also smoothed, and the image power s is reduced.

JP55-133179 Α addresses this problem by detecting the direction of the edge component contained in the image and applying the logic along the direction without reducing the sharpness of the image. «Doing noise; ^ is shown.

JP2001-57677 A discloses noise removal; ^ using multi-resolution conversion. In this noise elimination method, first, the image signal is divided into signals in a plurality of frequency bands using a Laplacian pyramid ^ H or wavelet transform technique. Then, after applying the filter key having the direction as described in JP55-133179A to the divided signals in each band, they are recombined. This has the advantage that it is possible to remove the noise of the bow girl suitable for each of the divided bands while suppressing the conversion of the edge component to! JP2001-57677 A also uses the iS¾ side image in the multiple H ^ g transform to determine the direction of the image used for direction determination; ^ or the result of the direction determination using the low side image. The combination of the results of direction discrimination using the image of the band of interest is shown: ^ :. Disclosure of investigation

However, in JP55-133179A, noise is generated based on the result of direction determination using the image signal, so that it is strongly influenced by the noise included in the image signal. For this reason, it is difficult to obtain an accurate direction discrimination result, and there is a problem that as a result of noise, the weak force ^ ffliit is crushed.

JP2001-57677 A is designed to improve the direction discrimination accuracy by combining the direction discrimination result using the image on the fiber side and the direction discrimination result using the image of the band of interest ^ "; However, since the images on the {^ side in many a ^ f ^ do not include etsu or fine crane structures, it is difficult to obtain accurate direction determination results. In addition, since the direction discrimination for the image in the band of interest is strongly affected by noise, it is difficult to obtain an accurate direction discrimination result, and the direction discrimination processing needs to be calculated twice. Therefore, it is not desirable from the parents of.

The present invention has been made in view of the above problems, and maintains the Etsu and ^ t structure by performing highly accurate edge direction discrimination based on the skeletal components corresponding to the typical edge structure extracted from the image signal. And d.

The image transfer apparatus according to a sickle according to the present invention is a first component that is a skeleton component that indicates the center of an image including a flat region divided by the edge of the image and the edge of the image. For the image, the second component image calculated based on the edited image signal and the first component image, the image ^ part to be an image representing multiple components, and the first component image A direction discriminating unit that discriminates the direction of the edge component, and a noise that observes noise in accordance with the direction of the edge component of the self-second component image are provided.

The thumbtack position according to another sickle of the present invention includes a first component image, which is a skeleton component made up of a flat component extracted from a part other than the image edge and the edge of the image, For the second component image that is calculated based on the image signal and the first component image, the image sequence that is connected to the image that represents multiple components, including the first component image, A direction discriminating unit that discriminates the direction of the edge component, and a noise {g »that detects noise according to the direction of the tiHB edge component for the second component image, are determined.

According to yet another sickle image method of the present invention, the image signal is a skeletal component indicating a general structure of an image including a flat region divided by an image edge and an edge of the image. First

A component image, a second component image calculated based on the tillSITO image signal and the unpleasant first component image, and an image representing a plurality of components, including the direction of the wedge component with respect to the first component image The noise is determined according to the direction of the edge of the printing edge component for the second printing image.

An image program according to still another aspect of the present invention includes a first component image, which is a skeleton component indicating a global structure of an image including a flat region divided by an image edge and a tmiB edge. , Rnm t mm ι The second component image calculated based on the component image, a step suitable for an image representing a plurality of components including, and the direction of the edge component relative to the ffl! S first component image _Bl is calculated by making the computer purple for the step and the step for generating noise according to the direction of the self edge component for the tins second component image. Solve the problem, Brief description of the drawings

FIG. 1 is a configuration diagram of a device according to the first embodiment.

2A, 2B, and 2C are diagrams showing examples of the Mi image signal I, the first component image U, and the second component image V, respectively, and FIG. 2D, FIG. 2E, and FIG. FIG. 5 is a diagram showing an example of a direct destruction component, a harmonic component, and a frequency component obtained by performing separation based on the frequency component of FIG.

FIG. 3 is a diagram for explaining the direction discriminating method.

FIG. 4 is a block diagram of noise in the first state.

FIG. 5A, FIG. 5B, and FIG. 5C are diagrams for explaining the coring process.

Figure 6 is a flowchart corresponding to the key from the image ^? Part to the image compositing part in the first ^? State. FIG. 7 is a configuration diagram of the imaging according to the second actual state.

8A, 8B, and 8C are diagrams for explaining the direction discriminating method.

9A, 9B, and 9C are diagrams for explaining the noise model.

FIG. 10 is a block diagram of the noise iS in the second state.

FIG. 11 is a flowchart corresponding to the processing from the image collar part to the image composition part in the second embodiment.

FIG. 12 is a configuration diagram of a device according to the third mode.

FIG. 13B is a diagram showing an example in which there is no correlation between color components, and FIG. 13B is a diagram showing an example in which there is a correlation between machine components. FIG. 14 is a configuration diagram of the noise iS «in the third mode.

Figure 15 shows the image in the third ¾ »state ^! 5 is a flowchart corresponding to keys from the image to the image composition unit. Best mode for carrying out the invention

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. Implementation of the following? In the ^ state, an explanation will be given by taking an example of the present invention as an example.

First real «state>

FIG. 1 is a system configuration diagram of an imaging according to the first embodiment of the present invention. This arrangement consists of an optical system 101, a solid-state image sensor 10 02, an AZD conversion unit 10 3 (hereinafter referred to as A / D 10 3), a signal key unit 10 04, an image unit 10 05, and an extraction unit 4 0 0, Direction judgment further U part 1 0 6, Noise

07, an image composition unit 1 0 8, 0 9, a storage medium 1 1 0, and a system controller 1 0 0.

The solid-state imaging device 1 0 2 is a signal processing unit 1 0 4 ^^^ via AZD 1 0 3. The signal section 1 0 4 is displayed as the image section 1 0 5 ^. The image continuous section 105 is an extraction section 400 and an image composition section 10. Extraction unit 4 0 0 is a direction determination additional IJ unit 1 0

6, and the noise is 0 7 ^^^. The direction discriminator 1 0 6 has noise {g »1 0 7 ^^^. Noise 0 7 is the image composition part 1 0 8 ^^^. Image composition unit 1 0 8 ί¾ ± β 1 0 9

It has been played.

The unit is bi-directionally connected to the system controller 100, and the operation is controlled by the system controller 100. In Figure 1, dislike is data communication, and 赚 is maintenance. Represents a line.

Figure 1 illustrates ¾fSx of the process. Based on the control of the system controller 100, the fixed element 10 0 2 outputs an optical image formed on the surface of the femoral image element 1 0 2 through the optical system 1 0 1 as an analog 15 image signal. This analog image signal is sent to AZD 100.

In the first difficult form, the thigh image element 102 is a color image sensor having a color filter array arranged on the front surface. The solid element 0 2 may be many;

The analog IS image signal that has been converted to A / D 1 0 3 ^ ¾ is converted to the digital No. W code, and then converted into a predetermined color image signal (hereinafter referred to as the original image signal I) by the signal processing unit 10 4. The The image signal I consists of R, G, and B color signals. Subsequent checks are performed independently of color beliefs.

The converted application image signal I is sent to the image collar portion 105. The image continuous unit 105 transmits the image signal I to the first component image U and the second component image V according to the color confidence.

The first component image U is a skeletal component viewable image structure of the original image signal I that includes a flat ± bandage (a component that changes slightly) and edge components. This first component image U is derived from "a component indicating the structure of an image including the edge included in the image signal I and the smooth flat region of the image change divided by the edge", or the image signal I. It is defined as “a component obtained by removing a fine structural component such as a texture (hereinafter referred to as a texture component)”. The first component image U is obtained from the average sample extracted from the edge of the image and the part other than the edge. Also expressed as a skeleton component.

The second component image V is a component calculated based on the original image signal I and the first component image U. The More specifically, the second component image V is a part of the first component image U that includes the texture component and noise and is equivalent to the image signal I.

As an example, Fig. 2B and Fig. 2C show the first component image U and the second component image V, respectively, which are separated into components of the original image signal I shown in Fig. 2A. In order to achieve this, the signal is displayed as a one-dimensional signal.) The edge that represents a small image included in the image signal I is included in the first component image U, and the anomaly component that is awkward is included in the second component image V. include.

For comparison, examples of DC component, wave component, and Ί¾ frequency component that have been separated by frequency, such as Fourier transform, are shown in Fig. 2D, Fig. 2Ε, Fig. 2 respectively. F 〖. The edge that represents the 赚 in the original image signal I is represented as a linear sum of each component, so it is not included only in a specific component.

Here, for simplicity of explanation, the image signal I is described as being connected to two components, but it is possible to reduce the number to three or more components.

There are three types of image type male and multi-type type, which will be described in detail later. The extraction unit 400 is a rectangular area of a predetermined size centered on the pixel of interest from the first component image U generated by the image ^ unit 1 0 5, for example, in this embodiment, a pixel block of 5 × 5 pixels (hereinafter referred to as a pixel block) , Pixel block of the first component image U) is extracted. Also from the second component image V, a pixel block (hereinafter referred to as a pixel block of the second component image V) corresponding to the pixel block region of the first component image U is extracted with the target pixel at the center. The size of this area can be changed to other sizes as required.

The pixel block of the first component image U extracted by the extraction unit 400 is subjected to a direction determination unit 10 6, and the pixel block of the second component image V is subjected to noise 1 0 7 ^. Based on the pixel block of the first component image U, the direction discriminator 1 0 6 determines the edge direction at the target pixel position. Is determined. The edge direction discrimination result is noise 0. Details of the direction determination key will be described later.

The noise {5 «1 0 7 first applies a filter to the pixel block of the second component image V by determining the filter coefficient or the filter process itself based on the edge direction indicated by the edge direction discrimination result. This is performed for all pixels. The second component image after filtering is referred to as v, hereinafter eir.

Noise 0 7 is further subjected to coring processing to make the fine / hi symbol zero for the finallet processing ν ′, and noise {g »a is completed. Hereafter, the second image after coring ^ a is called V ".

Here, the coring key is a signal obtained by subtracting the «value of the input signal by the threshold value if the input signal value is less than the threshold value, and the signal is defined as a single bit. It seems to be a number. Details of noise will be discussed later.

The image composition unit 10 8 obtains the first component image U from the image sequence 1 0 5 force, and the noise and the second component image V ”after processing at a predetermined ratio, for example, a ratio of 1: 1. As a result, a composite image 1 'in which noise is reduced is obtained with respect to the desired image signal I. The composite image is recorded on a memory medium such as a flash memory via JM ^ l 0 9. 1 1 0 ^ ¾.

Next, the details of the image の of the image signal I performed in the image 舊 part 105 will be described. When the first component image u is extracted from the image signal I force, the skeleton component of the image signal I will be used as the ¾ ^ Use a simple key. As described above, there are Calo arithmetic type separation and Multiplication type separation in the image ^, and each method will be described in turn below.

Add type 爐> The Karo arithmetic type »e is expressed as the sum of the first component image u and the second component image V as shown in the image signal I force s equation (1).

I = + V (1)

Here, first, we will explain the method using a bounded variation function and a norm. In order to perform the dredge, the A 2 B C variational model (Aujol-Aubert-Blanc-Feraud "" Chambolle model) ¾ "described in Reference 1 below is used.

Reference 1: Jean-Francois Aujol, Guy Gi, boa, Tony Chan & Stanley Osher, Structure- Texture Image Decomposition ~ Modeling, Algorithms, and Parameter Selection, International Journal of Computer Vision, Volume 67, Issue 1 (April 2006) Pages: 111-136 Year of Publication: 2006

As the property of the l-th component image u obtained as Bounded Variation Function Space (BV) composed of multiple “small sub-regions of smooth luminance change” divided by unbounded ¾ boundaries Modeled. The energy of the first component image U is defined as the kinetic energy by the TV (Total Variation) norm J (U) in equation (2).

In Equation (2), X is the pixel position in the horizontal direction of the first component image U, and y is in the vertical direction! ^ Position.

On the other hand, the relation of the second component image V in Eq. (1) is modeled as the relation G. G between «I-function ^^ is the space of the function expressed by Equation (3) by the leakage function gl, g2, and its energy is defined as G norm II v || G in Equation (4) Is done. Ugly inf) ² + (g JL; = + ₂ }

gl.g2 (4) The problem of the image signal I is converted into a variational problem of equation (5) that maximizes the energy generality / J. This variation problem can be solved by Chambolle's Projection method.

inf (hi) + I, -hi- ₂ }

α> 0, μ> 0, Ο _μ = {ν≡Ο \ \\ ν \\ _α ≤μ} ₍₅₎

As a result, the second component image V associated with the application image signal I is affected by noise, but the first component image U is hardly affected by noise and the skeletal component is not dulled.

(^ mam is generated.

As another example of the addition type t ^ method, target smoothing may be performed by a median filter, a morphological filter, or the like. The following is an example of a method based on a bounded variation function.

Example 1: First component image U is the result of median filter ffl applied to image signal I, and second component image V is the wisteria of first component image U in image signal I;

Example 2: Method in which the first component image U is the result of applying a morphological filter to the image signal I, and the second component image V is the image of the first component image U in the i ^ li image signal I

Example 3: The first component image U is the result of applying the reduction / Η0¾ to the image signal I and further expanding it, and the second component image V is the result of the first component image U in the image signal I. Example 4: Method of using the first component image U as the result of applying Bilateral Filter to the image signal I and using the second component image V as the rattan of the first component image U in the image signal I

<Multiplication type 儘> Next, I will explain the multiplication type ^. In the multiplication type, the image signal Ϊ is represented by the product of the first component image U and the second component image V. If the counter iW ^ H image signal obtained by converting the image signal I to the opposite is f, then the equation (6 ) Can be converted to an additive problem.

I = U * V

f = u + v;

f = log /, u = logU, v = logV (g) Here, as in the previous case, we explain the linkage method using a bounded variation function and a norm. Multiplication and separation problems can be solved in the opposite region using the A2BC variational model, as in the case of Karo arithmetic separation. In the following, we will briefly describe the A2BC variation in the region, with the multiplication type ^! As «.

Similar to the first component image U of the additive type separation model described above, the relationship between the anti-crane 1-component image U in Eq. (6) Modeled as a bounded variation function space (BV) composed of small parts ^ '. The energy of the one-component image U vs. crane is the TV (Total Variation) norm J ( Defined in U).

J (u) = || V _M || i & dy

(7) On the other hand, the function space of the difficult two-component image V in Eq. (6) is modeled as the vibration function space G. Sakaiseki 0 is the space of the function expressed by Equation (8) by the gripper functions gl and g2, and its energy is defined as G norm II V II G in Equation (9).

v) = ^d Si ( _x , _y ) +

8I, 82≡L _∞ (R ² ) _(G)

Ha = i gnl.gf2 | fa) ² + and ^v = + (9) Therefore, the problem of the image signal f is expressed as a variational problem of Eq. (1 0) that minimizes the energy functional.

As a result, the image signal I force, the second component image V, which is applied, is affected by noise. The force first component image U is hardly affected by noise. Therefore, it is possible to extract the skeletal component collision ^] image structure it) without dulling the edges.

The image signal obtained by subtracting the corresponding pixel value of the first image U obtained in advance from each pixel value of the image signal I may be used as the second component image V.

Next, details of the operation of the direction determination unit 106 will be described. Based on the pixel block of the first component image U from the extraction unit 400, the direction determination unit 106 calculates a threshold value for determining the edge direction for each direction based on the pixel position of interest.

FIG. 3 is a diagram showing the sharpness of determining the direction from the 5 × 5 pixel block. AO 0 to A 44 indicate each pixel in the pixel block, and the center pixel A 22 corresponds to the target pixel position. In this ¾W mode, the direction of the edge to be determined is the four directions shown in FIG. The number of directions can be increased to 8 directions as necessary.

In direction discrimination, direction 1 force direction 4 is determined, and in order to determine which is the direction along the edge most, the dictation value is calculated for each direction. Various fHffi values are conceivable, but in this embodiment, l ¥ MilE 1 to E 4 calculated based on the equation (11) are used. E 1 to E 4 are evaluation values corresponding to directions 1 to 4, respectively.

E 1 = | A22-A23 | + | A22-A24 | + | A22-A21 | + | A22-A20 | E2 = | A22-A13 | + | A22— A04 | + | A22— A3 l | + | A22_A40 | E3 = | A22-A12 | + | A22_A02 | + | A22— A32 | + | A22-A42 | E4 = | A22-A11 | + | A22_A00 | + | A22— A33 | + | A22-A44 |

(11)

The dragon value has an edge that grows larger when the edge is smaller, and smaller when the edge is smaller.

The direction discriminating unit 106 detects the straight line among tmi 1 force, et al. E 4 and determines that there is no edge in the direction corresponding to the dragon value. That is, the direction corresponding to the smallest threshold is determined as the direction along the edge. The detected edge direction is set to Noise 107. Next, details of the operation of noise 07 will be described.

FIG. 4 is a diagram illustrating an example of the noise 107. The noise 107 includes a direction-specific filter unit 120 and a coring unit 121. The extraction unit 400 and the direction discriminating unit 106 are directional filter units 120 ^^^. The direction-specific filter unit 120 is a coring unit 12. The coring unit 121 is connected to the image composition unit 108.

First, the pixel block of the second component image V obtained from the extraction unit 400 is input to the direction-specific filter unit 120. The direction-specific filter unit 120 determines a filter coefficient based on the edge direction judgment result from the direction determination unit 106, that is, the edge direction, and performs a filtering process based on the determined filter coefficient. Here, the filter coefficient is designed (weighted) so that the pixel positioned in the edge direction with respect to the target pixel greatly contributes to the filter effect.

The finale ^ Μ¾¾ is the coring part 121 ^^. The coring part 121 is The correlating process is performed on the Noleta result to make the signal fine / J and zero. 5A to 5C show an example of processing in the coring unit 1 2 1.

In the coring process, the reference value is B, the upper threshold value is Cl = B + Tl (T1> 0), the lower threshold value is C2 = B + T2 (T2 <0), and the signal A before key and C0 after processing. {The word “D” is a signal ^ 0® represented by the equation (1 2) and the graph in Fig. 5A.

B + T1 A A D = A-Tl-B

B + T2 A A A + T1 龄 D = B

A ^ B + T2 ^^ D = A-T2-B (1 2) where T l and Τ 2 are the parameters that determine the upper and lower thresholds, respectively. In the first ¾6 mode, A fixed value set in advance is used.

Fig. 5B and Fig. 5C show that the coring force S is performed. Fig. 5 B [^ su One-dimensional signal

If coring is performed on ^ A, Fig. 5 C 〖^ "信" ^ 直 D is obtained. For box B, the moving average force S of faith A is generally used.

However, it is the second component image V that has been filtered by the i-direction filter unit 120, and the second component image V is obtained by removing the first component image U, which is the image signal I force skeleton component. It is. Therefore, Naoto B is set to 0 in this QI ^ state.

Coring 匁! The result is the image composition unit 10.

It is also possible to replace the fixed ft¾element 102 with a monochrome image pickup element and configure the 基づき based on the monochrome image signal. Therefore, the image signal I becomes a monochrome signal, and each component U, V input from the image signal becomes a] signal.

Fig. 6 shows the flow of software processing from the image decomposition unit 105 to the image composition unit 10 8. The program used in software processing is, for example, For example, store it in the computer storage. C PU computer reads the storage device program from, ¾ItT¾ ₀ Note that KiHanare置, for example, a magneto-optical disk, CD- ROM, DVD-ROM, a body memory. It is also possible to distribute a program to a computer via a communication line, and the computer receiving this distribution can HfrT the program.

In step S O 1, the image signal I is suitable for the first component image U and the second component image V. In step S 0 2, the edge direction is calculated based on the first component image U, and the edge direction is determined based on the evaluation value.

In step S 0 3, direction-specific filtering is performed on the second component image V based on the edge direction determined in step S 0 2. In step S O 4, coring processing is performed on the second component image V, after filtering by direction. In step S O 5, the first component image U and the second component images V,, after keying in step S O 4 are synthesized to obtain a synthesized image. As described above, according to the first male form, the image signal is divided into the first component image U which is a skeleton component and the second component image V obtained from the wisteria of the first component image U which is similar to the original image signal. In the image representing multiple components, the direction of the edge component is determined with respect to the first component image U, and the noise is input according to the direction of the edge component with respect to the second component image V. . By discriminating the direction of the wedge component based on the first component image U including the skeletal component, it becomes possible to accurately determine the direction without being affected by noise. In addition, by applying noise to the second component image V containing noise according to the direction of the edge component, it is possible to make noise ^!

In particular, the pixel block of the first component image U and the pixel block of the second component image V corresponding to the pixel block of the first component image U are extracted. Of the pixel block of the second component image V, the edge component direction and the pixel position on the second component image V corresponding to the target pixel position Based on the pixel values of the pixel group located in the orthogonal direction, the fine wave component in the region is calculated. First

When noise is applied to the two-component image V, by calculating the wave component considering the direction of the edge component, noise {g »ia with an edge component added can be obtained.

In addition, since the direction of the edge component from the target pixel position is determined using the pixel block of the first component image u, a stable direction determination result can be obtained.

In addition, since the first component image U and the second component image V after noise are synthesized, a noise image signal can be obtained with respect to the image signal I.

Since the first component image U includes the flat portion edge component of the customer image, the edge direction discrimination accuracy from the first component image U can be improved.

The image part 1 0 5 is the addition type defined by I = U + V ^! Or, because the multiplication type 爐 defined by I = U X V is performed, the first component image u including the skeletal component and the second component image V including the minute fluctuation component ^? This makes it possible to perform suitable for growth.

In addition, the image edge portion 105 calculates a total variation norm with respect to the image signal I and minimizes the calculated total variation norm to thereby obtain a first component including an edge component and a flat component of the image signal I force. Since the one-component image U is extracted, the first component image U can be extracted with high accuracy.

The original image signal I is composed of a plurality of machine parts, and the image connection part 1 0 5, direction discriminating part 1 0 6, and noise 0 7 are arranged in order of the decoration of the image signal I. It is possible to reverse the noise.

FIG. 7 is a system configuration diagram of the imaging apparatus according to the second embodiment of the present invention. Basic Φ «Nariha It is equivalent to the first state, and the same name and number are assigned to the same configuration. Only the differences will be described below.

In the second ¾ »state, from the ^ state of 1 in FIG. 1, the direction discriminating unit 1 0 6 force S direction discriminating unit 2 3 4 is noisy, {¾« 1 0 7 is noisy, {5 ^ 2 0 1 Replaced with no. The parameter configuration 2 0 0 is i ^.

The image ^ 1 part 1 0 5 includes an image composition part 1 0 8, a parameter setting 2 0 0, and an extraction part 4 0 0 ^^^. The extraction unit 400 is obscured by the direction discrimination unit 2 3 4 and noise 2 0 1. The direction discriminating unit 2 3 4 and the parameter setting 2 0 0 are noise 2 0. Noise ί® »2 0 1 is the image composition part 1 0 8 ^^^.

In the second embodiment, the upper threshold value and the lower threshold value in the coring process are set at a predetermined fixed value. In the second mode, the signal level of the first component image U is set. Based on this, the amount of noise in the second component image V is estimated, and the upper limit threshold and the lower limit threshold are set based on the noise amount. With such a configuration, it is possible to control the effect of the coring process by the amount of noise.

The process »will be described with reference to FIG. Similar to the first ¾6 ^ mode, the extraction unit 4 0 0 extracts the pixel block of the first component image U and the pixel block of the second component image V. The pixel block of the first component image U extracted by the extraction unit 400 is sent to the direction discriminating unit 23, and the pixel block of the second component image V is sent to the direction discriminating unit 2 34 and noise {¾¾¾ 2 0 1 ^^^

The direction discriminating unit 2 3 4 discriminates the edge direction based on the pixel block of the first component image U and the second component image V. The edge direction discrimination result is noise {g «2 0 1 ^^. We will go back to the details of the direction discriminator. The parameter setting ^ p ₂₀₀ is obtained by first obtaining the signal level m of the first component image U from the image section 05, and then using the noise amount model set in advance,

The amount of noise in the second component image V corresponding to the U signal level is estimated for each pixel. The details of the estimation of the amount of noise will be delayed.

Next, parameters Tl, Τ 2 are set based on the amount of noise σ. The parameters Tl and Τ2 are set to values proportional to the noise amount σ by, for example, Equation (13). However, in Equation (1 3), k is a predetermined coefficient, for example, 1/2.

Tl = k σ

T2 = -k σ (13)

Parameters Tl and Τ2 are set to noise {© ¾Ρ201. Noise iS ¾ ¾ 201 is the same as in the first ¾ 6 state; first, based on the edge direction indicated by the edge direction determination result obtained from the direction determination unit 234 for the pixel block of the second component image V by ^ Determine the filter coefficient and apply the filter key.

Next, unlike the first ¾Sfe $ state, by applying the coring ffi based on the parameters Tl and Τ2 obtained from the parameter settings to the second component image V, after the filter key, Complete Noise iS ^! Details of noise will be discussed later.

Next, details of the operation of the direction determination unit 234 will be described. The direction in the first »^ state is similar to the operation of the determination unit 106, except that the pixel values at the corresponding pixel positions of the pixel block of the first component image U and the pixel block of the second component image V are predetermined. The difference is that the direction is discriminated based on the reference image I dir generated by weighted averaging according to the ratio. This point will be described with reference to FIG.

In the direction discriminating unit 234, first, the first component as shown in FIG. Obtain the pixel block of the image U and the pixel block of the second component image V shown in Fig. 8B (see Fig. 8A. In Fig. 8A, A 0 0 to A4 4 indicate each pixel in the pixel block, and the center The pixel A 2 2 corresponds to the target pixel position In Fig. 8B, B 0 0 to B 44 indicate each pixel in the pixel block, and the center pixel B 2 2 corresponds to the target pixel position. A new pixel value is calculated from, and the pixel block of the reference image I dir is subdivided to determine the direction.

Each pixel of C 00 to C 44 in FIG. 8C is a pixel of the reference image I dir to be flfled for direction discrimination. Each pixel C 0 0 to C 4 4 includes pixels A 0 0 to A 4 4 included in the pixel block of the first component image and pixels BO 0 to B 4 4 included in the pixel block of the second component image. From the above, it is constructed by weighted addition such as equation (1 4); 1 ^ 1 ~.

Cij = Aij + wij * Bij (i, j = 0, ..., 4) (1 4)

In equation (1 4), wij is a weight of about 0 to 0.5. When wij is close to 0, the first component image U-force S is given priority, so it is resistant to noise and the direction is determined based on the ^^-like character, ^^ structure. On the other hand, the larger the wij is, the more accurate direction discrimination can be performed in consideration of the influence of minute fluctuation components such as texture.

Note that wij can have a different value depending on the pixel position in the pixel block. In addition, a configuration in which wij can be input from the outside to the direction discriminating unit can be configured so that wij can be arbitrarily set.

Next, based on the pixel block of the reference image I dir, direction determination is performed by the same method as in the first exemplary embodiment, and the direction determination result is set to noise.

Next, the noise amount estimation method in the parameter setting 200 will be described. In the parameter setting 200, a signal level-one noise carrying model (hereinafter referred to as a noise model) force S for the second component image V is recognized. No. Parameter setting 2 0 0 is By referring to the model, the amount of noise corrected in the second component image V is estimated. The noise model will be described below.

The amount of noise increases to the second order excitation with respect to the word level immediately after conversion by A / D 103. As disclosed in JP2005—175718 A, when the signal level-noisy-noise variance model is expressed by a quadratic function, it is expressed by equation (15). However, in Equation (15), α,

) 3, γ is a subterm.

σ = ∑ ² ₀ + β∑ ₀ + γ

However, the amount of noise σ varies not only with the signal level but also with the gain of the element. As an example, Fig. 9 プロット plots the amount of noise σ at 3 ISO ISO sensitivities (gains) 100, 200 and 400 related to the gain at ^^ t. Each curve has the form shown in Eq. (15), but its coefficient depends on the ISO sensitivity related to the gain. If t is set to t, the gain is set to g, and the noise model is converted to ^: in consideration of the above, it can be expressed by Equation (16).

In Equation (16), ο ^, _ygt i is a constant term determined according to humidity t and gain g. Color! ^ Signal ^ \ This noise model can be applied independently to each color signal.

However, since the signal processing unit 104 performs signal processing after AZD conversion, the above model cannot be used as it is. In other words, considering the difficulty of the noise model, it is necessary to consider the characteristics 14 of the signal section 104.

For example, the Knee process that converts a 12-bit input signal into a 8-bit output signal performs a γ transformation: ^, and the signal ^ ® unit 104 has input / output reliability as shown in Figure 9B. is doing. In the figure, L (12) represents the signal level immediately after A and D conversion, and L (8) represents the signal level after signal processing.

Therefore, if the characteristics of the signal level and the amount of noise immediately after AZD conversion in Equation (15) or Equation (16) and the characteristics of the signal processing unit 104 as shown in Fig. 9B are considered, The signal level L (8) and the amount of noise are the values in the curve shown in Fig. 9C.

The parameter setting ^ 2 0 0 is used to obtain the signal level of the 1-component image U from the image ^? Part 1 0 5 force. Next, based on the signal level of the first component image U, the noise amount is obtained by referring to the noise model shown in Fig. 9C. For example, according to Equation (1 3), depending on the amount of noise The parameters Tl and Τ2 are set.

Next, details of the operation of noise 2 0 1 will be described. ¾Φ «is the same as the noise ig« 1 07 in the first ¾ embodiment, and the same configuration and the same name and number are assigned. Only the differences will be described below.

The operation of noise 10 07 in the first male form is the same as that of the noise 10 07 except that the coring process is controlled based on the parameters T 1 and T 2 obtained from the parameter settings ¾¾20. Different.

FIG. 10 is a diagram showing an example of a noise iS »20 1. The noise 2 0 1 has a configuration in which the coring part 1 2 1 of noise 0 7 shown in FIG. 4 is replaced with a coring part 2 2 1. The direction discriminating unit 2 3 4 is a direction-specific filter unit 1 2 0 ^^^. The parameter setting example 2 0 0 and the direction-specific filter section 1 2 0 are coring sections 2 2 1 The coring unit 2 2 1 is the image composition unit 1 0 8 ^^^.

First, the pixel block of the second component image V output from the extraction unit 400 is input to the direction-specific filter unit 120. The direction-specific filter unit 1 2 0 is the same as the direction discriminator 2 3 4 The filter direction is determined based on the edge direction, that is, the edge direction, and the filter processing is performed. The filter result is the coring part 2 2 1 ^ ¾.

The coring unit 2 2 1 performs coring so that a minute signal is used as the opening for the filter processing result. The operation is the same as that of the coring portion 1 2 1 in the noise 07 in the first embodiment. However, since the parameters T 1 and T 2 that specify the coring upper limit threshold and the lower limit threshold are obtained from the parameter setting 2 0 0 and used, depending on the estimated noise amount, It is possible to control the noise level of the core link. The coring result is generated by an image composition unit 1 0 8 ^ ¾.

Note that it is also possible to replace the fixed element 102 with a monochrome image pickup element and to configure based on a monochrome image signal. Then, the image signal I becomes a monochromatic mouth signal, and becomes each component u, v «¾¾ signal transmitted from the J¾S image signal.

Fig. 11 shows the flow of the processing from the image ^ 1 part 10 5 to the image composition part 10 8 by software processing. The program used in the software is written in 1t¾ of the computer's ROM, RAM, etc. and executed by the computer's CPU. The first

Then, assign the same step number.

In step S O 1, the image signal I is converted into a first component image U and a second component image V. Step S 1. Then, after calculating the edge direction based on the first component image U and the second component image V, the edge direction is determined based on the evaluation value. In step S11, based on the noise model, the noise amount of the second component image V is estimated for each pixel from the signal level of the first component image U, and parameters Tl and Τ2 are set based on the noise amount. To do.

In step SO 3, the edge direction determined in step S 1 0 for the second component image V Based on, filter by direction. In step S 12, the second component image V ′ after the direction-specific filtering is subjected to the coring 麵 based on the parameters T 1 and Τ 2 set in step S 11. In step S 0 5, the first component image U and the second component images V,, after being subjected to the coring key in step S 12 are synthesized to obtain a synthesized image Γ.

As described above, according to the second embodiment, the noise parameters Τ 1 and Τ 2 are set based on the first component image U, and the noise S ^ parameter Τ 1, に対して is set for the second component image V. 2. Perform noise according to the direction of the edge component. Neutral estimated from the first component image U

As a result, the degree of noise can be controlled navigationally according to the amount of noise, and more accurate noise iS ^ W becomes possible.

In particular, the pixel block of the first component image U and the pixel block of the second component image V corresponding to the pixel block of the first component image U are extracted, and the target pixel of the pixel blocks of the second component image V is extracted. For the pixel position on the second component image V corresponding to the position, the extinction amount in the region is calculated based on each pixel value of the pixel group that is aligned in the direction orthogonal to the direction of the edge component. The calculated harmonic component is corrected based on the noise parameters T 1 and T 2. When noise {S ^ is applied to the second component image V, the Sl wave component is calculated considering the direction of the edge component, and the noise parameter T l, Τ 2 set based on the noise level By correcting the components, high-precision noise can be achieved.

The direction discriminator 2 3 4 discriminates the direction of the edge component that is the target pixel using the pixel block of the first component image U and the pixel block of the second component image V, so that the texture not included in the skeleton component Thus, it is possible to obtain a highly accurate direction discrimination result in consideration of such a small ¾¾¾ component. The direction discriminating unit 2 3 4 also weights the pixel block of the first component image U and the second component image V. Since the edge component direction is determined using the image signal calculated by averaging, a highly accurate direction that simultaneously considers the skeletal component and minute wrinkle components such as textures not included in the skeletal component A discrimination result can be obtained.

Furthermore, by providing a setting for setting the weighted average ratio to an arbitrary value, the first component image including the skeletal component ¾r ^, the second component image including the minute viewing component, and the force s direction discrimination accuracy are affected. The degree can be freely changed. This makes it possible to flexibly improve the direction discrimination accuracy.

The image signal I is composed of a plurality of components, and the image resolution unit 1 0 5, the direction discriminating unit 2 3 4, and the noise 2 0 1 force S are ordered by ^ of the image signal I. The noise iS «2 0 1 is applied to the first component image U and the second component image V that have been generated by the image ^? The noise is input to the second component image V according to the direction of the noise parameters T1, T2 and the edge component. As a result, the effect of the noise can be controlled, and more accurate noise can be achieved.

FIG. 12 is a system configuration diagram of an imaging apparatus according to the third embodiment of the present invention. The composition is the same as the second »^ example, and the same composition is assigned the same name and number. Only the differences will be described below.

The configuration of the third embodiment is different from the configuration of the second embodiment shown in FIG. 7 in that the buffer 3 0 1, the correlation calculation unit 3 0 2, the parameter correction unit 3 0 3, and the noise 1ffi¾ The configuration is such that 2 0 1 is replaced with noise low 3 4.

Image ^ 1 part 1 0 5 is extracted part 4 0 0 Parameter setting ¾¾ 2 0 0 and image composition unit 1 0 8 ^^ The extraction unit 400 includes a direction discriminating unit 2 3 4, noise iS «3 4, and a buffer 3 0. The buffer 3 0 1 has a correlation operation unit 3 0 2 ^^^. The direction discriminator 2 3 4 has noise ffi ^ 3 0 4 ^^. The parameter setting 2 0 0 and the correlation calculation unit 3 0 2 are a parameter correction unit 3 0 3 ^^^. The parameter correction unit 3 0 3 is noise, iS¾¾ 3 0 4 ^^. Noise, image 3 0 4, is included in the image composition unit 10.

In the second mode, the noise amount in the second component image V is estimated based on the first component image U, and the threshold value is set based on the noise amount. Here, in a color image that does not contain noise, the cross-correlation coefficient between stations evaluated in the 3 X 3 near ^! Region (hereinafter, the correlation is a positive value close to 1. On the other hand, noise ¾r ^ color image In this case, the number of pixels that are close to each other decreases, so the number of close contacts decreases, so it can be said that the smaller the value of the number of awakens, the more noise is included in the image signal. .

Therefore, in the third embodiment, the color correlation between the color signals of the second component image V after the decomposition is obtained, and the parameters T l and Τ 2 for controlling the coring key are corrected according to the obtained color correlation. To do. As a result, in the region where the correlation is high, the coring width force S at the coring 麵 is reduced so that the original signal acts as i¾m ". On the other hand, in the region where the correlation is low, the corin A wider corrugation width and flattening of the era makes it possible to use a more accurate coring process.

That is, in the third embodiment, the fine relationship between each color signal of the succeeding second component image V is obtained, and the threshold value of the core link is set based on the noise amount and the color correlation. Explain the key »with reference to Fig. 12.

The pixel block of the second component image V extracted by the extraction unit 400 is stored in the buffer 300. Said.

No. In the parameter setting 200, after estimating the noise amount of the color confidence by the same method as in the second mode, the parameters T 1 and Τ 2 are set based on, for example, Eq. (1 3). The parameters T 1 and Τ 2 are the parameters ¾Ε part 3 0 3 ^^.

The correlation calculation unit 30 2 obtains the pixel block of the second component image V from the buffer 3 0 1 for all the color signals, and then obtains each pixel in the pixel block near the pixel block including the target pixel position. Calculates the correlation number of the touch and outputs the most / J @r 蕭, which is the most / J, and the number r. Details of the correlation calculation will be described later. The maximum / J number r is parameter ¾E part 3 0 3 ^ ¾.

The parameter correction unit 303 corrects the parameters T 1 and T 2 based on the maximum / J §relation number r. The corrected parameters T 1 and D 2 are hereinafter referred to as D 1 and T 2. Set parameter T l, Τ 2 to ¾E.

Noise 0 4 is a color signal that represents the filter processing based on the edge direction obtained from the direction discriminator 2 3 4 and the coring based on the parameters T l ′ and T 2 ″ obtained from the parameter corrector 3 0 3. The noise is done by doing independently.

Here, if the direction-specific filter and coring are performed independently for each color signal, there will be artifacts caused by the correlation between the color signals being lost. Therefore, the second component image v,, and the second component image V of noise {S ^ iir after coring processing are weighted and added based on a predetermined ratio to suppress visually noticeable artifacts. The second component images V,, after weighted addition are sent to the image composition unit 10. Details of noise processing will be described later.

Next, the details of the correlation calculation processing and the correction of the parameters T 1 and Τ 2 according to the correlation between the color signals will be described. Here, for simplicity of explanation, it is assumed that the second component image V is one-dimensional. I will explain. Since the second component image V is the fine component like texture, ^^ component and noise, the first component image U in the m image signal I ^^ component, so the second component image V is zero. It is the component that rises as' βτ as the center.

Fig. 13 Α and 1313B show »between the pixel position of the image signal I and the signal level of each machine. In the example shown in Figure 13A, there is no correlation between machine minutes. The signal level for each ^^ at each pixel position is as shown in Table 1.

[table 1]

Table 2 shows the number of phases between R and G, between G and B, and between B and R.

[Table 2]

this ,

Becomes ~ 0.75148.

On the other hand, in the example of FIG. 13B, there is a correlation between the color components. Table 3 shows the signal level of each color component at each pixel position.

[Table 3] Pixel position RGB

1 1 0 -1

2 4 3 3

3 5 3 4

4 -2 -1 0

5 1 1 0 -1

Each phase between R ~ G, G ~ B, B-R Γ 蕭, the number in Table 4 ^ 1 ~.

[Table 4]

In this case, the maximum / J ^ number r is 0.873891.

In the third cat form, the maximum / J 湘 relation number r is used, and parameters T 1 and T 2 are ffi! E according to equation (17).

Tl, — Ck X (1-r) ² XT1

T2, — Ck X (1-r) ² XT 2 (17)

In equation (17), Ck is a predetermined coefficient. If Ck = l, parameters Tl,

Τ 2 will be multiplied by 3.0677 times of Fig. 13 ：: 6 \ 0.01590 times of Fig. 13B, respectively.

Accordingly, as in FIG. 13 A, Le a correlation between械分, since the width between the upper limit threshold value and the lower threshold of the coring threshold is larger trees this, act to more ⁵ the signal To do. On the other hand, as shown in Fig. 13B, when there is a correlation between touching minutes, the correction is made so that the width between the upper and lower coring thresholds becomes smaller, and there is a slight correction that preserves the signal. Done.

Thus, according to the third embodiment, the coring processing threshold value is corrected in accordance with the correlation between the color components, so that noise among the fluctuation components included in the second component image V is corrected. It is possible to discriminate between the component due to the texture and the component due to the original structure of the image, such as texture, and it is possible to suppress the deterioration of the latter due to coring.

Next, details of the operation of the noise iS »304 will be described. The play is the same as the noise iSS ¾201 in the second ¾ state, and the same name and number are assigned to the same configuration. However, the difference is that the second component image V,, after the coring key and the second component image V immediately after the image sequence are weighted and output.

FIG. 14 is a diagram illustrating an example of the configuration of the noise ig »304. The noise 1S ^ 304 is configured as shown in Fig. 4 〖; ΐ ^ {noise, {S »201 with an adder 321 i ¾.

No. The parameter fIE section 303 has a coring section 221 ^^^. The coring unit 22 1 is a calorie calculating unit 32. Image part 105¾¾Πcalculation part 321 ^^^ The adder 321 has an image total of 08 ^^^.

Based on the parameters Tl, Τ 2, corrected by the parameter correction unit 303, the direction-specific finer processing 120 and the coring unit 221 perform direction-specific finisher processing and coring 麵 respectively. . The subsequent second component image V, ′ is added to the adder 321.

The calorie calculation unit 321 obtains the second component image V immediately after the image sequence from the image sequence unit 105, and based on a predetermined ratio set in advance, the second component image V and the second component image after the coring key. Component Image V '' is weighted and added. This suppresses artifacts that occur when noise {S ^ W is excessively affected by the direction-specific filter key and coring processing. be able to.

FIG. 15 shows a flow when the processes from the image decomposition unit 105 to the image composition unit 108 are performed by software processing. Programs used in the software can be stored in the ROM, RAM, etc. of the computer! ^ And executed by the computer's CPU. The same step number is assigned to the same key as the signal key flow in the first difficult form.

In step SO 1, the image signal I is input to the first component image U and the second component image V. In step S 10, after the threshold value indicating the edge direction is calculated based on the first component image U and the second component image V, the edge direction is determined based on M. In step S 13, the correlation of each color signal of the second component image V is calculated to calculate the maximum coefficient _r .

In step S 11, the noise amount of the second component image V is estimated for each pixel from the signal level of the first component image U based on the noise model, and parameters T 1 and T 2 are calculated based on the noise amount. Set. In step S14, the parameters Tl, Τ2 obtained in step S11 are obtained based on the maximum / J 湘隱 r r obtained in step S13 to obtain parameters Tl,, Τ2,. In step S 03, direction-specific filtering is performed on the second component image V based on the edge direction determined in step S 10. In step S12, coring 麵 is performed on the second component image V, after the direction-specific filter processing, based on the parameters T1 ′ and T2 ′ set in step S14.

In step S15, the second component image V,, after coring and the second component image V of noise ^ are weighted and added at a predetermined ratio. In step S 05, the first component image U and the second component image V, 'after weighted addition in step S 15 are synthesized to obtain a synthesized image Γ. As described above, according to the third aspect, the image signal I is composed of a plurality of tactile segments, and the first component image U and the second component image V that have been converted to “^!” Of the original image signal I are buffered. 3 0 1 ^ ^ In addition, based on the second component image V of ^ ¾ ^ that was fined in the buffer 3 0 1, calculate the phase and number for each ^ ^ minute, and ^ ^ Correction based on the correlation coefficient is performed on the noise {¾¾ parameters D 1 and T 2 obtained by sequentially setting the parameter setting 2 0 0 to U. Then, the first component image U and the first For the two-component image V, the direction discriminator 2 3 4 and noise 3 0 4 are used in order for coloration.Estimated by the first component image U containing the skeletal component for each mechanical component of the power error image signal The noise した and ° parameters set based on the noise level were corrected based on the number of awakenings for each touch, and the noise ffi relationship was used. More accurate noise ie ^ a is possible.

Further, according to the third ^ state, the second component image V of noise and the second component image V,, after noise are added based on a predetermined ratio. There is a noise ig ^ in the direction filter, coring processing, etc., and artifacts may be generated ^^. It is possible to dislike the artifact that the artifact is visually conspicuous.

The present invention is not limited to the first to third embodiments J. For example, in the first to third male forms described above, an example in which the image according to the present invention is used as a device has been described. However, the device can be applied to a device of device: ^, orchid as a single unit. «It can also be set.

Claims priority based on Japanese Patent Application 2007-331972, filed with the Japan Patent Office on February 25, 2007. All the contents of Japanese Patent Application 2007-331972 are hereby incorporated by reference. Incorporated in the description.

Claims

The scope of the claims

1. The application image signal is a flat area divided by the edge of the image and υ¾ί を ^ wedge. The general structure of the image is the 1-component image which is the 1-skeleton component, the tiftara image signal and the first component. A second component image calculated on the basis of the split image, an image sequence that represents a plurality of components, and

A direction discriminating unit for discriminating the direction of the wedge component with respect to the first component image of disgust,

For the second component image, a noise part that displays noise according to the direction of the edge component,

Picture with MS

2. The image signal is calculated based on the first component image, which is a skeleton component consisting of the flat band extracted from the edge of the image and the υΐϋΐΕ edge: ^, the ΙΕίίΙΕβ image signal, and the tiHB first component image. A second component image to be issued, and ¾: ^ an image continuous portion connected to an image representing a plurality of components;

方向 ίίΙΞ direction discriminating unit that discriminates the direction of the edge component from the first component image,

E For the second component image, a noise part that generates noise according to the direction of the unpleasant edge component,

Picture with l ^ go

3. Further equipped with parameter setting to set the noise parameter based on the first component image of tfHH,

tins noise is »indicates that noise is applied to the second component image of tin itself according to the direction of the noise component.

4. The target pixel of the first component image ^ h ¾ ^ The first pixel group located in the peripheral region and the second component image located in the second component image corresponding to the pixel position of the first pixel group tiff An extraction unit for extracting the pixel group of 2;

Self-noise is a pixel in the second pixel group that is located in the direction perpendicular to the direction of the tiffS edge component with respect to the pixel position of the second component image that is associated with the pixel position of interest. 3. A face segment calculation unit that calculates an iSii wave component in the tins region based on each pixel value of the group.

5. Disgusting first component image ^: The pixel of interest ¾r ^ The first pixel group located in the peripheral area and the pixel position of the first pixel group corresponding to the pixel position of the first pixel group An extraction unit for extracting the pixel group of 2;

Noise is a pixel in the second pixel group that is positioned in the direction orthogonal to the direction of the edge component of the tiff with respect to the pixel position of the second component image of the tiff corresponding to the pixel position of the tiH5 target pixel. A minute calculation unit for calculating iSli ^ in the Stif self region based on each pixel value of the group;

A correction unit that detects the hate bandits based on the noise,

The image of claim 3 備える comprising

6. The tiiia noise is provided with an adder that performs calorie calculation on the second component image of Neuss and the second component image after noise based on a predetermined ratio.纖 in one section »3¾ 氍

7. The direction discriminating unit uses the first pixel group to discriminate the direction of the hate component that goes to the tiilE target pixel position.

8. The selfish direction determining unit uses the first pixel group and the second pixel group to determine the direction of the self edge component with respect to the MIS target pixel position. Image of

9. The IE direction discriminating unit discriminates the direction of the tiff & wedge component by using the image signal calculated by performing a weighted average of the first component image and the tins second component image. Description picture

1 0. 9ίίΕΛ Further comprising a setting unit for setting the ratio of the weight average to an arbitrary value. M ^ Po

1 1. Further comprising a synthesis unit for synthesizing the 1st component image of disgust and the 2nd component image after noise is applied. oa¾

1 2. The second image can be obtained by converting the first component image from the ffiiara image signal ^ 力 claim 1 force, etc. 1 claim 1 1 or 1 ^! ^ 'S!

1 3. Edited second component image is obtained by decimation of the source image signal with the first component image. Claim 1 Power Obtained Claim 1 1

1 4.

The energy calculation unit that calculates the Total Variation norm for the first component image and the first ffjia Total Variation norm / WtT to extract the first tiff self-component image from the / / ^ Ιί image signal. 1-component image ^ ¾

Claim 1 with power Claim 1 Claim 3 1

1 5. The image signal consists of multiple "^ components"

The tara image signal machine has a knitted self image part, a self direction discriminating part, and a noise part.

1 6. The image signal is composed of multiple machine segments,

The image of the disliked flrai image signal is divided into a disgusting image buttock, a ttilB direction discriminating portion, and a noise viewing portion ¾¾¾1 ^

Hate noise, {5 «is the first component image and second component image created by the MIE image ^ section on the lself ^ ¾ ^ According to claim 3, noise is applied to the IS second component image according to the direction of the noise °, ° parameter, and «self edge component of the decoration.

1 7. The tina ^ M image signal consists of multiple parts,

For each color component of the MIS original image signal, the MI component first component image applied by applying the sfrt self-image part Image, and miB second component image | ^ Section,

Based on the second component image of the instrument decorations described by the language, a correlation calculation unit that calculates the number and phase of each touch,

A parameter correction unit that makes detailed adjustments to the tffia first component image and corrects the set noise and angle parameters based on the number of tifl¾¾ 隱. In preparation for

For the first component image and the second component image of the knitting self, the tins direction discriminating section and the noise ig «are l¾½ ffl on the machine. ^ MS 1 8. The first image, which is a skeletal component showing the basic structure of the image including the flat region divided by the image edge and the υ¾! Ξ edge, and the image signal And the second component image calculated based on the tiflB first component image and an image representing a plurality of components, determine the direction of the edge component with respect to the first component image,

firfEH For two-component images, an image that suppresses noise according to the direction of the edge component ^ &

1 9. Calculate the application signal based on the first component image, which is the skeletal component that shows the typical structure of the image including the flat region divided by the image edge and mis edge, and the ^ tm ι component image A second component image to be generated, a step to make an image representing a plurality of components,

tfjiam 1 component image, the step of determining the direction of the edge component,

赚 ίίΙΕ Step for entering noise in the second component image according to the direction of the tins wedge component An image program that allows a computer to execute and.