WO2016075914A1 - 画像信号処理装置、画像信号処理方法及び画像信号処理プログラム - Google Patents
画像信号処理装置、画像信号処理方法及び画像信号処理プログラム Download PDFInfo
- Publication number
- WO2016075914A1 WO2016075914A1 PCT/JP2015/005569 JP2015005569W WO2016075914A1 WO 2016075914 A1 WO2016075914 A1 WO 2016075914A1 JP 2015005569 W JP2015005569 W JP 2015005569W WO 2016075914 A1 WO2016075914 A1 WO 2016075914A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image signal
- component
- unit
- texture
- frequency component
- Prior art date
Links
- 238000012545 processing Methods 0.000 title claims description 204
- 238000003672 processing method Methods 0.000 title claims description 33
- 238000004364 calculation method Methods 0.000 claims abstract description 101
- 238000000034 method Methods 0.000 claims abstract description 94
- 238000012937 correction Methods 0.000 claims abstract description 24
- 230000015572 biosynthetic process Effects 0.000 claims description 88
- 238000003786 synthesis reaction Methods 0.000 claims description 88
- 230000007850 degeneration Effects 0.000 claims description 78
- 238000000926 separation method Methods 0.000 claims description 38
- 230000008569 process Effects 0.000 claims description 32
- 230000002194 synthesizing effect Effects 0.000 claims description 32
- 230000001131 transforming effect Effects 0.000 claims description 10
- 230000009466 transformation Effects 0.000 abstract description 13
- 238000000354 decomposition reaction Methods 0.000 description 25
- 238000010586 diagram Methods 0.000 description 19
- 230000009467 reduction Effects 0.000 description 15
- 230000014509 gene expression Effects 0.000 description 13
- 230000000694 effects Effects 0.000 description 12
- 230000006870 function Effects 0.000 description 8
- 238000004590 computer program Methods 0.000 description 6
- 230000002146 bilateral effect Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000001629 suppression Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 239000006185 dispersion Substances 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000011946 reduction process Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/10—Image enhancement or restoration using non-spatial domain filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/168—Segmentation; Edge detection involving transform domain methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/40—Picture signal circuits
- H04N1/409—Edge or detail enhancement; Noise or error suppression
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20048—Transform domain processing
- G06T2207/20064—Wavelet transform [DWT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20182—Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20192—Edge enhancement; Edge preservation
Definitions
- the present invention relates to an image signal processing method, an image signal processing device, and an image signal processing program, and more particularly to an image signal processing method, an image signal processing device, and an image signal processing program for removing noise from an image.
- a technology for removing noise contained in an image is an indispensable technology for reproducing a captured image more clearly.
- Patent Documents 1 to 3 The following techniques are disclosed in Patent Documents 1 to 3 and Non-Patent Documents 1 and 2 in relation to a technique for removing noise from captured images.
- Patent Document 1 discloses a method for improving image quality through wavelet transform and inverse wavelet transform.
- Patent Document 2 discloses a method for creating a provisional high-resolution image (a base image) by enlarging an input image to an output size.
- Patent Document 3 wavelet transformation is performed on an original image, image flaws are repaired with respect to low frequency components by interpolation, and inverse wavelet transformation is performed using the restored low frequency components and high frequency components. Is reconstructed to obtain a final repaired image.
- Non-Patent Document 1 discloses a technique of denoising processing (referring to processing for removing noise; the same applies hereinafter) based on image component separation.
- Non-Patent Document 2 discloses a denoising method based on wavelet degeneration.
- Non-Patent Documents 1 and 2 will be briefly described.
- FIG. 15 is a conceptual diagram for explaining the technique of Non-Patent Document 1.
- an original image signal f in is input, the structure component u consisting of the edges of the image component and a flat component, separates the texture component v consisting noise and fine patterns.
- the Total Variation Minimization (TV) method disclosed in Non-Patent Document 3 or the bilateral filter described in Non-Patent Document 4 can be used.
- a TC (Texture Component) reduction unit 1002 applies a process of suppressing a noise component included in the texture component v to generate a texture component v ′ in which the noise is suppressed.
- the soft decision threshold processing shown in the following formula (1) is effective.
- composition unit 1003 synthesizes the structure component u and the texture component v ′ in which the noise is suppressed to generate an output image signal.
- the TV method which is one method for separating the input image signal into the structure component u and the texture component v, will be described.
- the structure component u can be obtained by introducing a regularization term into the total variation norm TV (u) represented by the following formula (3) and minimizing the following formula (4).
- u 0 is the original image signal f
- ⁇ is a parameter representing the fidelity with the original image signal.
- DTVF Digital TV Filter
- Equation (4) a filter process based on the local variation is used to solve Equation (4). Assuming that the input image signal is u (0) and the output image signal after being filtered N times is u (N) , the filter output u ⁇ (N) at the pixel position ⁇ is expressed by the following equation (5).
- h ⁇ and h ⁇ are filter coefficients (referred to as filter processing coefficients; the same applies hereinafter) and are represented by the following Expressions (6), (7), and (8).
- is when sufficiently large relative to the noise component, since the h alpha alpha ⁇ 1, it is possible to prevent the blurring of edges.
- time is small, since DTVF becomes h alpha alpha ⁇ 0 considers this region flat and behaves like an ordinary low pass filter.
- FIG. 16 is a conceptual diagram for explaining the technique of Non-Patent Document 2.
- WC degeneration unit 2004 a process (degeneration process) for setting a wavelet coefficient having a small absolute value to 0 is applied to the lowest resolution high-frequency components LH 3 , HL 3 , and HH 3 to generate LH 3 ′ and HL 3. Get ', HH 3 '.
- WC is a wavelet coefficient, that is, a wavelet coefficient.
- the formulas (1) and (2) may be simply used.
- the noise is random noise, the noise component included in the input image is distributed to all the wavelet coefficients. Therefore, the noise can be removed by subtracting the noise component from each wavelet coefficient.
- IWT is Inverse WT, that is, inverse wavelet transform.
- the generation is performed from the low-frequency component LL 3 having the lowest resolution and the high-frequency components LH 3 ′, HL 3 ′, and HH 3 ′ having the lowest resolution to which the degeneration processing is applied.
- the reduction processing is sequentially applied to the high-frequency wavelet coefficients having the resolution other than the lowest resolution (WC reduction sections 2006 and 2008). Then, from the low frequency component of the resolution obtained from the resolution one lower than the resolution and the high frequency component of the resolution to which the reduction process is applied, the low frequency component of the next higher resolution is generated by inverse wavelet transform. (IWT 2007, 2009). Then, the inverse wavelet transform result at the highest resolution is used as the output image.
- FIG. 17 is a diagram illustrating an application example of the technique of Non-Patent Document 2.
- the left side of FIG. 17 is an input image
- the center of FIG. 17 is a result of wavelet transform of one layer
- the right of FIG. 17 is a result of wavelet transform of three layers.
- LL l represents a low frequency component of the l-th layer
- LH l , HL l , and HH l represent high-frequency components of the l-th layer, respectively.
- Non-Patent Documents 3 to 7 are disclosed in relation to the present invention.
- noise removal of an image signal is performed by applying the background technology (in this specification, noise removal means noise removal or reduction)
- an image is constructed contrary to the influence of noise of the image signal.
- the filter coefficient is set so as to blur the edge to be blurred.
- the present invention solves the above-described problems and provides an image signal processing method and the like that rarely set filter coefficients that unintentionally blur edges constituting an image due to the influence of noise. Objective.
- the image signal processing method and the like can suppress an increase in calculation cost.
- An image signal input unit that inputs an image signal of an original image and a wavelet transform unit that generates a low frequency component and a high frequency component by performing wavelet transform on the image signal are provided.
- a first structure / texture separation unit that separates the low frequency component into a first structure component and a first texture component, and corrects the value of the first texture component, and the corrected first texture component Is provided with a texture component degeneration section.
- a first synthesis unit that synthesizes the first structure component and the corrected first texture component and generates a synthesized low frequency component, and the high frequency component and the synthesized low frequency component.
- An inverse wavelet transform unit that generates an image signal after inverse wavelet transform by performing inverse wavelet transform is provided.
- a second structure / texture separation unit that separates the image signal after the inverse wavelet transform into a second structure component and a second texture component; and corrects the value of the second texture component;
- a second texture component degeneration unit that generates a second texture component is provided. Furthermore, the second structure component and the corrected second texture component are combined to generate an image signal processing signal, and an image signal output unit that outputs the corrected image signal is provided. .
- the filter coefficient used for the structure / texture decomposition in the denoising process can be calculated from an image with less noise than the original image generated using the wavelet transform.
- the low frequency component in the wavelet transform is an image signal having a lower resolution than the image signal of the original image.
- a filter coefficient calculated from an image signal having a resolution lower than that of the original image is generated with reference to a wider range of pixel values than a filter coefficient calculated from an image signal having the same resolution as that of the original image signal.
- the influence of noise when calculating the filter coefficient is less than that of an image signal having the same resolution as the original image signal.
- the filter coefficient used for structure / texture decomposition is obtained based on this low frequency component, it is less likely to set a filter coefficient that blurs the edges constituting the image due to the influence of noise. Therefore, it is possible to effectively suppress the noise of the low frequency component while maintaining the edge component included in the low frequency component. Furthermore, if an image with the same resolution as the original image is generated from the low-frequency component with suppressed noise using the inverse wavelet transform, the noise in this image is suppressed from the original image. Less than the original image signal. Therefore, it is less likely to set a filter coefficient that blurs the edges constituting the image due to the influence of noise, and noise can be effectively suppressed while maintaining edge components. In addition, the image signal processing apparatus according to the present embodiment does not need to correct the filter coefficient using the calculation results of a wide range of image feature amounts, and thus can suppress an increase in calculation cost.
- the first embodiment is an embodiment related to an image signal processing apparatus having a minimum configuration according to the present invention.
- FIG. 1 is a conceptual diagram showing an image signal processing apparatus according to the first embodiment.
- the configuration shown in FIG. 4B is the minimum configuration of the present invention.
- the image signal processing apparatus of the present embodiment includes an image signal input unit 001, an image signal processing unit 002, and an image signal output unit 003, as shown in FIG.
- the image signal input unit 001 outputs the input image signal of the original image to the image signal processing unit 002.
- the image signal processing unit 002 performs denoising processing on the input image signal of the original image, and outputs the corrected image signal subjected to denoising processing to the image signal output unit 003.
- the image signal output unit 003 outputs the image signal subjected to the denoising process to the outside.
- the image signal processing unit 002 includes a WT (Wavelet Transform (wavelet transform)) unit 101 as shown in FIG.
- the image signal processing unit 002 further includes an IWT (Inverse Wavelet Transform (inverse wavelet transform)) unit 201.
- the image signal processing unit 002 further includes a first STD unit (Structure Texture Decomposition) 301.
- the image signal processing unit 002 further includes a second STD unit 302 and a first TC (Texture Component (texture component)) reduction unit 401.
- the image signal processing unit 002 further includes a second TC degeneration unit 402, a first combining unit 501, and a second combining unit 502.
- the original image signal f in is input from the image signal input unit 001, one level of applying the wavelet transform, the low-frequency components of the wavelet transform LL 1 and the high-frequency component LH 1, HL 1, HH 1 is calculated. Then, a low-frequency component LL 1 to the first STD unit 301, and outputs the high frequency component LH 1, HL 1, HH 1 to IWT portion 201.
- the low frequency component LL 1 is an image signal having a resolution one layer lower than the original image signal fin in terms of the wavelet transform.
- the first STD unit 301 separates the low frequency component LL 1 of the wavelet transform input from the WT unit 101 into a structure component u 1 and a texture component v 1 .
- the structure component u 1 is output to the first synthesis unit 501, and the texture component v 1 is output to the first TC degeneration unit 401.
- the first TC degeneration unit 401 applies a process of suppressing the noise component to the texture component v 1 input from the first STD unit 301 to generate a texture component v 1 ′ in which the noise is suppressed,
- the data is output to the synthesis unit 501.
- the first synthesizing unit 501 synthesizes the structure component u 1 input from the first STD unit 301 and the texture component v 1 ′ with suppressed noise input from the first TC degeneration unit 401, and corrected low A frequency component LL 1 ′ is generated.
- the corrected low frequency component LL 1 ′ is output to the IWT unit 201.
- the IWT unit 201 uses the inverse wavelet transform from the corrected low frequency component LL 1 ′ input from the first synthesis unit 501 and the high frequency components LH 1 , HL 1 , HH 1 input from the WT unit 101.
- the reconstructed image signal LL 0 ′ from which noise is removed is generated.
- the reconstructed image signal LL 0 ' is an image signal having the same resolution as the input image signal f in. Then, the reconstructed image signal LL 0 ′ from which noise has been removed is output to the second STD unit 302.
- the second STD unit 302 separates the corrected reconstructed image LL 0 ′ input from the IWT unit 201 into a structure component u 0 and a texture component v 0 . Then, the structure component u 0 is output to the synthesis unit 502 and the texture component v 0 is output to the second TC degeneration unit 402.
- Second TC degeneracy unit 402 like the first TC degeneracy unit 401, with respect to the texture component v 0 which is input from the second STD unit 302 applies the processing for suppressing a noise component, the noise is suppressed texture
- the component v 0 ′ is generated.
- the generated texture component v 0 ′ is output to the second synthesis unit 502.
- the second synthesis unit 502 has a structure component u 0 input from the second STD unit 302 and a texture component v 0 in which noise input from the second TC degeneration unit 402 is suppressed. 'Is synthesized to generate an output image signal f out . The generated f out is output to the image signal output unit 003. [Image signal processing method of this embodiment] Next, the image signal processing method of this embodiment will be described.
- FIG. 2 is a flowchart showing the operation of the image signal processing method of the first embodiment.
- the WT unit 101 applies wavelet transform to an input image signal to obtain a low frequency component and a high frequency component of the input image (S001).
- the first STD unit 301 applies structure / texture decomposition to the low-frequency component obtained by the wavelet transform according to the above-described procedure with respect to the same part to obtain a low-frequency component and a texture component (S002). Further, from the texture component of the low frequency component, the first TC degeneration unit 401 obtains a texture component in which noise is suppressed by the above-described procedure described for the same part (S003).
- the first synthesizing unit 501 synthesizes the low-frequency component structure component and the noise-suppressed texture component in accordance with the above-described procedure, and obtains a corrected low-frequency component (S004).
- An inverse wavelet transform is applied using the corrected low frequency component and high frequency component to obtain a reconstructed image signal (S005).
- the second STD unit 302 applies structure / texture decomposition to the reconstructed image signal in the above-described procedure with respect to the reconstructed image signal to obtain a structure component and a texture component of the reconstructed image signal (S006).
- the second TC degeneration unit 402 applies the above-described procedure with respect to the texture component of the reconstructed image signal to obtain a texture component in which noise is suppressed (S007). Then, the second synthesis unit 502 synthesizes the structure component of the reconstructed image signal and the texture component in which noise is suppressed to generate an output image signal (S008).
- the filter coefficient used for the structure / texture decomposition in the denoising process can be calculated from an image with less noise than the original image generated using the wavelet transform.
- the low frequency component in the wavelet transform is an image signal having a lower resolution than the image signal of the original image.
- a filter coefficient calculated from an image signal having a resolution lower than that of the original image is generated with reference to a wider range of pixel values than a filter coefficient calculated from an image signal having the same resolution as that of the original image signal.
- the influence of noise when calculating the filter coefficient is less than that of an image signal having the same resolution as that of the original image signal. Therefore, when the filter coefficient used for structure / texture decomposition is obtained based on this low frequency component, it is less likely to set a filter coefficient that blurs the edges constituting the image due to the influence of noise.
- the image signal processing apparatus does not need to correct the filter coefficient using the calculation results of a wide range of image feature amounts, and thus can suppress an increase in calculation cost.
- the second embodiment is an embodiment of an image signal processing apparatus that calculates filter coefficients used for structure / texture decomposition for an image signal of a certain resolution.
- the wavelet transform layer is arbitrary, but here, the case of one layer will be described.
- FIG. 3 is a conceptual diagram showing an image signal processing apparatus according to the second embodiment.
- the image signal processing apparatus of the present embodiment includes an image signal input unit 011, an image signal processing unit 012, and an image signal output unit 013 as shown in FIG.
- the image signal input unit 011 outputs the input image signal of the original image to the image signal processing unit 012.
- the image signal input unit 011 is connected to, for example, an imaging device such as a camera or a scanner, an image database in which image data is captured and stored, or a network to which the image data is connected. An original image signal is input.
- the image signal processing unit 012 performs denoising processing on the input image signal of the original image, and outputs the corrected image signal subjected to denoising processing to the image signal output unit 013.
- the image signal output unit 013 outputs the image signal subjected to the denoising process to the outside.
- the outside is, for example, a display, a printer, a storage medium such as a hard disk or memory card that holds image data, or a network to which they are connected, and these display, store, or transmit images.
- the image signal processing unit 012 includes a WT unit 111, an IWT unit 211, a first STD unit 311 and a second STD unit 312 as shown in FIG.
- the image signal processing unit 012 further includes a first TC degeneration unit 411, a second TC degeneration unit 412, a first synthesis unit 511, a second synthesis unit 512, and a filter coefficient calculation unit 710.
- the original image signal f in is input from the image signal input unit 011, one level of applying the wavelet transform, the low-frequency components of the wavelet transform LL 1 and the high-frequency component LH 1, HL 1, HH 1 is calculated. Then, a low-frequency component LL 1 to the first STD unit 311, and outputs the high frequency component LH 1, HL 1, HH 1 to IWT portion 211.
- the low frequency component LL 1 is an image signal having a resolution one layer lower than the original image signal fin in terms of the wavelet transform.
- the first STD unit 311 applies, for example, the DTVF described in Non-Patent Document 2 to the low-frequency component LL 1 of the wavelet transform input from the WT unit 111, to the structure component u 1 and the texture component v 1 . To separate. Then, the structure component u 1 is output to the first synthesis unit 511, and the texture component v 1 is output to the first TC degeneration unit 411.
- the first TC degeneration unit 411 applies a process of suppressing the noise component to the texture component v 1 input from the first STD unit 311 to generate a texture component v 1 ′ in which the noise is suppressed.
- simple processing such as Equation (1) or Equation (2) may be used, or the nonlinear low-pass filter used in the first STD unit 311 may be applied by changing parameters. May be. Then, the generated texture component v 1 ′ is output to the first synthesis unit 511.
- the first synthesizing unit 511 synthesizes the structure component u1 input from the first STD unit 311 and the texture component v 1 ′ with suppressed noise input from the first TC degeneration unit 411, and corrects the low frequency
- the component LL 1 ′ is generated.
- the corrected low frequency component LL 1 ′ is output to the IWT unit 211.
- the IWT unit 211 includes a corrected low frequency component LL 1 ′ input from the first combining unit 511 and a corrected high frequency component LH 1 ′, HL 1 ′, HH 1 ′ input from the WT unit 111, Using the inverse wavelet transform, a reconstructed image signal LL 0 ′ from which noise has been removed is generated.
- the reconstructed image signal LL 0 ' is the same resolution image between the input image f in. Then, the reconstructed image signal LL 0 ′ from which noise has been removed is output to the filter coefficient calculation unit 710.
- h ⁇ represents the filter coefficient calculated at the pixel position ⁇
- the filter coefficient at the center of h ⁇ is represented by h ⁇ in Expression (6)
- ⁇ LL1 represents the position of each coefficient of the low frequency component LL 1 of the wavelet transform.
- the processing performed in the first STD unit 311 and the second STD unit 312 is not limited to the structure / texture decomposition described in Non-Patent Document 2, and is a process using a nonlinear low-pass filter such as a bilateral filter or an epsilon filter. There may be.
- the reconstructed image signal LL 0 ′ is output to the second STD unit 312.
- the second STD unit 312 receives the DTVF of Non-Patent Document 5 from the set h 0 of filter coefficients input from the filter coefficient calculation unit 710 and the corrected reconstructed image signal LL 0 ′ input from the IWT unit 211. Is applied to separate the structure component u 0 and the texture component v 0 . Then, the structure component u 0 is output to the second synthesis unit 512, and the texture component v 0 is output to the second TC degeneration unit 412.
- Second TC degeneracy unit 412 Similar to the first TC degeneracy unit 411, with respect to the texture component v 0 which is input from the second STD unit 312 applies the processing for suppressing a noise component, the noise is suppressed texture The component v 0 ′ is generated. Then, the generated texture component v 0 ′ is output to the second synthesis unit 512.
- the second synthesis unit 512 Similar to the first synthesis unit 511, the second synthesis unit 512 has a texture component v 0 in which the structure component u 0 input from the second STD unit 312 and the noise input from the second TC degeneration unit 412 are suppressed. 'Is synthesized to generate an output image f out . Then, the generated f out is output to the image signal output unit 013. [Image signal processing method of this embodiment] Next, the image signal processing method of this embodiment will be described.
- FIG. 4 is a flowchart showing the operation of the image signal processing method of the second embodiment.
- the WT unit 111 applies wavelet transform to an input image to obtain a low frequency component and a high frequency component of the input image (S101).
- the first STD unit 311 applies the structure / texture decomposition to the low-frequency component obtained by the wavelet transform by the above-described procedure with respect to the same part to obtain the structure component and the texture component of the low-frequency component (S102).
- the first TC degeneration unit 411 obtains a texture component in which noise is suppressed by the above-described procedure described for the same part (S103).
- the first synthesizing unit 511 synthesizes the low-frequency component structure component and the noise-suppressed texture component by the above-described procedure with respect to the same component to obtain a corrected low-frequency component (S104).
- the IWT unit 211 applies inverse wavelet transform using the corrected low frequency component and high frequency component to obtain a reconstructed image signal (S105).
- the filter calculation unit 710 calculates a set of filter coefficients using the reconstructed image signal (S106).
- the second STD unit 312 applies the above-described procedure with respect to the reconstructed image signal, performs structure / texture decomposition using the set of filter coefficients calculated in S106, and obtains the reconstructed image signal.
- a structure component and a texture component are obtained (S107). Further, the second TC degeneration unit 412 applies the above-described procedure with respect to the texture component of the reconstructed image signal to obtain a texture component in which noise is suppressed (S108). Then, the second synthesizing unit 512 generates an output image signal by synthesizing the structure component of the reconstructed image signal and the texture component in which noise is suppressed (S109).
- each unit can be configured by hardware, but can also be realized by a computer program.
- functions and operations similar to those of the above-described embodiment are realized by a processor that operates according to a program stored in the program memory.
- a filter coefficient calculated from an image signal having a resolution lower than that of the original image is generated with reference to a wider range of pixel values than a filter coefficient calculated from an image signal having the same resolution as that of the original image signal.
- the influence of noise when calculating the filter coefficient is less than that of an image signal having the same resolution as that of the original image signal. Therefore, when the filter coefficient used for structure / texture decomposition is obtained based on this low frequency component, it is less likely to set a filter coefficient that blurs the edges constituting the image due to the influence of noise.
- the image signal processing apparatus does not need to correct the filter coefficient using the calculation results of a wide range of image feature amounts, and thus can suppress an increase in calculation cost.
- the third embodiment is an embodiment of an image signal processing apparatus that calculates a filter coefficient when performing structure / texture separation for an image signal of a certain resolution by using a filter coefficient calculated from a low-frequency component having a low resolution.
- the hierarchy of wavelet transform is arbitrary, but here, the case of one hierarchy will be described.
- FIG. 5 is a conceptual diagram showing an image signal processing device according to the third embodiment.
- the image signal processing apparatus includes an image signal input unit 021, an image signal processing unit 022, and an image signal output unit 023 as shown in FIG.
- the image signal input unit 021 outputs the input image signal of the original image to the image signal processing unit 022.
- the image signal input unit 021 is connected to, for example, an imaging device such as a camera or a scanner, an image database in which image data is captured and stored, or a network to which the image data is connected. An original image signal is input.
- the image signal processing unit 022 performs denoising processing on the input image signal of the original image, and outputs the corrected image signal subjected to denoising processing to the image signal output unit 023.
- the image signal output unit 023 outputs the image signal subjected to the denoising process to the outside.
- the outside is, for example, a display, a printer, a storage medium such as a hard disk or a memory card that holds image data, or a network to which they are connected, and these display, store, or transmit images.
- the image signal processing unit 022 includes a WT unit 121, an IWT unit 221, a first STD unit 321, a second STD unit 322, a first TC degeneration unit 421, a second unit, as shown in FIG. And a TC degeneration unit 422.
- the image signal processing unit 022 further includes a first synthesis unit 521, a second synthesis unit 522, a first filter coefficient calculation unit 721, a second filter coefficient calculation unit 722, and a filter coefficient synthesis unit 820. .
- the first filter coefficient calculation unit 721 calculates a set of filter coefficients h 1 using the low frequency component LL 1 separated by the wavelet transform input from the WT unit 121, and outputs the set to the filter coefficient synthesis unit 820.
- h ⁇ represents the filter coefficient calculated at the pixel position ⁇
- the filter coefficient at the center of h ⁇ is represented by h ⁇ in Expression (6)
- the filter coefficient other than the center is represented by h ⁇ in Expression (7).
- ⁇ LL1 represents the position of each coefficient of the low frequency component LL 1 of the wavelet transform.
- the STD unit is not limited to the structure / texture decomposition of Non-Patent Document 2, and a nonlinear low-pass filter such as a bilateral filter or an epsilon filter may be applied.
- the first filter coefficient calculation unit outputs a set of filters calculated for each pixel position to the filter coefficient synthesizing unit 820 even when a bilateral filter or an epsilon filter is used.
- the low frequency component LL 1 is output to the first STD unit 321.
- the IWT unit 221 uses the inverse wavelet transform from the corrected low frequency component LL 1 ′ input from the first synthesis unit 521 and the high frequency components LH 1 , HL 1 , HH 1 input from the WT unit 121.
- the reconstructed image signal LL 0 ′ from which noise is removed is generated.
- the reconstructed image signal LL 0 ' is the same resolution image between the input image f in. Then, the reconstructed image signal LL 0 ′ from which noise has been removed is output to the second filter coefficient calculation unit 722.
- the second filter coefficient calculation unit 722 calculates a set of filter coefficients h 0 using the input reconstructed image signal LL 0 ′.
- the calculation method is the same as the filter coefficient calculation method in the first filter coefficient calculation unit 721.
- the filter coefficient h 0 is output to the filter coefficient synthesis unit 820.
- the reconstructed image signal LL 0 ′ is output to the second STD unit 322.
- the filter coefficient synthesis unit 820 synthesizes the filter coefficient sets h 0 and h 1 to generate a filter coefficient set h 0 ′.
- the set of filter coefficients h 0 ′ is output to the second STD unit 322.
- the method of synthesizing the filter coefficient sets h 0 and h 1 in the filter coefficient synthesizing unit 820 is as follows.
- the filter coefficients used in the DTVF in this hierarchy are synthesized as in the following formulas (11) and (12).
- c 0 is a synthesis ratio of filter coefficients, and 0 ⁇ c 0 ⁇ 1. The meaning of this synthesis will be described.
- the edge is estimated by analyzing not only local pixel value fluctuations but also wide-range pixel value fluctuations, and filtering by the TV method is performed. It is effective to limit the direction to the direction along the edge.
- Non-Patent Document 6 requires an additional wide-area filter process for analyzing fluctuations in a wide-range pixel value, and the calculation cost is a problem.
- the filter coefficients constituting a set h 1 of the filter coefficient of DTVF calculated from the low frequency components lower than the hierarchy hierarchy filter coefficients obtained from the results of analysis of the variation of a wide range of pixel values from the hierarchy It has become. That is, if the filter coefficient is generated in a flat region, strong smoothing is likely to be applied, and in a region where an edge is present, the coefficient along the edge direction is likely to be strongly weighted. In addition, since the wavelet transform generates a low-frequency component by a low-pass filter, the low-frequency component in a layer lower than the layer has less influence of noise than the layer. Therefore, the filter coefficient h 1 is less affected by noise.
- composition ratio c 0 used in the equations (11) and (12) may be given as a parameter set by the user or may be a value determined adaptively. For example, the variance of the pixel values of the target pixel and the surrounding pixels may be calculated, and the weight of the filter coefficient having a resolution lower than that resolution may be increased in a region where the variance is smaller.
- the second STD unit 322 applies the method of improving the DTVF of Non-Patent Document 5 to perform separation into the structure component u 0 and the texture component v 0 .
- the separation is performed from the filter coefficient set h 0 ′ input from the filter coefficient synthesizer 820 and the corrected reconstructed image signal LL 0 ′ input from the IWT unit 221.
- the structure component u 0 is output to the second synthesis unit 522, and the texture component v 0 is output to the TC degeneration unit 422.
- the second filter coefficient calculation unit 722 and the filter coefficient synthesis unit 820 may be common.
- the filter coefficient synthesis unit 810 and the second STD unit 322 may be common.
- the second filter coefficient calculation unit 722, the filter coefficient synthesis unit 820, and the second STD unit may be common.
- FIG. 6 is a flowchart showing the operation of the image signal processing method of the third embodiment.
- the WT unit 121 applies wavelet transform to the input image to obtain a low frequency component and a high frequency component of the input image (S201).
- the first STD unit 321 applies structure / texture decomposition to the low-frequency component obtained by the wavelet transform in accordance with the procedure described for the same part to obtain a low-frequency component structure component and a texture component (S202).
- the first TC degeneration unit 421 obtains a texture component in which noise is suppressed by the procedure described for the same part (S203).
- the first synthesizing unit 521 synthesizes the structure component of the low frequency component and the texture component in which the noise is suppressed by the procedure described with respect to the same component, and obtains the corrected low frequency component (S204).
- the IWT unit 221 applies the inverse wavelet transform using the thus corrected low frequency component and high frequency component to obtain a reconstructed image signal (S205).
- the first filter coefficient calculation unit 721 calculates a set of filter coefficients using the low frequency component (S206).
- the second filter coefficient calculation unit 722 calculates a set of filter coefficients using the reconstructed image signal (S207).
- the filter coefficient synthesis unit 820 synthesizes the filter coefficient obtained using the low frequency component and the filter coefficient obtained using the reconstructed image signal (S208).
- the second STD unit 322 performs structure / texture decomposition to which the above-described procedure described for the same part is applied, and obtains a structure component and a texture component of the reconstructed image signal (S209). Further, the second TC degeneration unit 422 applies the procedure described with respect to the texture component of the reconstructed image signal to obtain a texture component in which noise is suppressed (S210). Then, the second synthesis unit 522 synthesizes the structure component of the reconstructed image signal and the texture component in which noise is suppressed to generate an output image signal (S211).
- each unit can be configured by hardware, but can also be realized by a computer program.
- functions and operations similar to those of the above-described embodiment are realized by a processor that operates according to a program stored in the program memory.
- a filter coefficient calculated from an image signal having a resolution lower than that of the original image is generated with reference to a wider range of pixel values than a filter coefficient calculated from an image signal having the same resolution as that of the original image signal.
- the influence of noise when calculating the filter coefficient is less than that of an image signal having the same resolution as that of the original image signal. Therefore, it is less likely to set a filter coefficient that blurs the edges constituting the image due to the influence of noise.
- the image signal processing apparatus does not need to calculate a wide range of image feature amounts necessary for correcting the filter coefficient, and thus can suppress an increase in calculation cost.
- FIG. 7 is a conceptual diagram showing an image signal processing device according to the fourth embodiment.
- the image signal processing apparatus includes an image signal input unit 031, an image signal processing unit 032, and an image signal output unit 033.
- the image signal input unit 031 and the image signal output unit 033 are the same as those of the image signal processing apparatus of the third embodiment.
- the image signal processing unit 032 includes a WT unit 131, an IWT unit 231, a first STD unit 331, a second STD unit 332, a first TC degeneration unit 431, A second TC degeneration unit 432.
- the image signal processing unit 032 further includes a first combining unit 531, a second combining unit 532, a WC degeneration unit 631, and a filter coefficient calculation unit 730.
- the image signal processing unit 032 further includes a first gradient strength calculation unit 931, a second gradient strength calculation unit 932, a gradient strength synthesis unit 1030, and a filter correction unit 1130.
- the hierarchy of WT conversion is arbitrary, the case of 1 hierarchy is shown here.
- the filter coefficient calculation unit 730 calculates a set of filter coefficients h 0 using the reconstructed image signal LL 0 ′, and outputs the filter coefficient set h 0 to the filter correction unit 1130.
- the first gradient strength calculation unit 931 calculates a set of gradient strengths at each point ⁇ in the image using the low-frequency component LL 1 of the wavelet transform input from the WT unit 131.
- a set of gradient intensities at each point ⁇ in the image is a set of gradient intensities at each point ⁇ in the image, for example, calculated by the method of Non-Patent Document 6. It is.
- the gradient strength set g 1 is output to the gradient strength synthesis unit 1030.
- the low frequency component LL 1 is output to the first STD unit 331.
- the first gradient strength calculation unit 931 may be the same as the first STD unit 331.
- the second gradient strength calculation unit 932 uses the reconstructed image signal LL 0 ′ to collect a set of gradient strengths at each point ⁇ in the image using the same technique. Is output to the gradient strength combining unit 1030.
- the gradient strength combining unit 1030 combines the gradient strength sets g 0 and g 1 , creates a combined gradient strength set g 0 ′, and outputs it to the filter correction unit 1130.
- the filter correction unit 1130 corrects the filter coefficient set h 0 using the combined gradient strength set g 0 ′, and creates a corrected filter coefficient set h 0 ′.
- the corrected set of filter coefficients h 0 ′ is output to the second STD unit 332.
- Non-Patent Document 6 A method for obtaining a set of gradient intensities described in Non-Patent Document 6 will be described.
- the method of Non-Patent Document 6 introduces a new evaluation axis into the filter coefficient derivation process in the DTVF of Non-Patent Document 5. Specifically, the filter coefficient w ⁇ (u) before normalization of the DTVF calculated by the local variation amount defined by the equation (9) is corrected as the following equation (13).
- g ⁇ is the strength of the gradient of the pixel value from the pixel ⁇ in the direction of the pixel ⁇ adjacent to the pixel ⁇ .
- g ⁇ is calculated as in the following equations (14) and (15).
- the eigenvalues lambda alpha of Harris matrix of Non-Patent Document 7, which is calculated in the pixel alpha of the low-frequency component LL 3 (structure tensor), 0, lambda alpha, 1, and eigenvectors e ⁇ , 0, e ⁇ 1 is used.
- the eigenvalue and eigenvector of the Harris matrix indicate the gradient strength and gradient direction of a range of several pixels centered on the pixel of interest, respectively.
- the function F (g ⁇ ) in the equation (13) can take various forms. Basically, F (g ⁇ ) takes a large value when g ⁇ is small, and F (g ⁇ ) when g ⁇ is large. g ⁇ ) may be designed to take a small value. That is, by reducing the filter coefficient in the direction where the gradient is large, it is possible to suppress the occurrence of edge blurring due to processing that crosses edges.
- the gradient strength combining unit 1030 combines the gradient strength sets g 0 and g 1 obtained by the above method by the following method.
- the pixel position ⁇ ′ (i / 2, j / 2) and the neighboring pixel ⁇ ′ ⁇ N ( ⁇ ′).
- the synthesized g ⁇ used for correction of the filter coefficient is It is represented by
- c 0 is a synthesis ratio
- c 0 may be given as a parameter set by the user or may be a value determined adaptively. For example, the variance of the pixel values of the target pixel and the surrounding pixels may be calculated, and the weight of the filter coefficient having a resolution lower than the resolution may be increased in a region where the variance is smaller.
- the second STD unit 332 performs structure / texture separation on the reconstructed image signal LL 0 ′ using the corrected set of filter coefficients h 0 ′.
- FIG. 8 is a flowchart showing the operation of the image signal processing method of the fourth embodiment.
- the first gradient strength calculation unit 931 calculates a set of gradient strengths from the low frequency components (S302).
- the filter coefficient calculation unit 730 calculates a set of filter coefficients from the image signal after the inverse wavelet transform (S308).
- the second gradient strength calculator 932 calculates a set of gradient strengths from the image signal after the inverse wavelet transform (S309).
- the gradient strength combining unit 1030 combines the set of gradient strengths calculated from the low frequency components and the set of gradient strengths calculated from the image signal after the inverse wavelet transform (S310).
- the filter correction unit 1130 corrects the set of filter coefficients obtained in S308 using the set of gradient intensities synthesized in S310 (S311).
- the second STD unit 332 applies structure / texture decomposition to the image signal after the inverse wavelet transform using the set of filter coefficients corrected in S311 (S312).
- the image signal processing apparatus is calculated using a set of gradient strengths calculated for low frequency components that are image signals having a resolution lower than that of the original image, and a reconstructed image signal that is an image signal having the same resolution as that of the original image. And a set of gradient intensities. Then, the set of filter coefficients is corrected using the set of combined gradient strengths.
- the gradient strength calculated using the reconstructed image signal that is an image signal having the same resolution as the original image is an index for evaluating the local variation amount between adjacent pixels.
- the gradient strength calculated for the low-frequency component, which is an image signal having a resolution lower than that of the original image is an index for evaluating a wide range of pixel value fluctuations.
- the corrected filter coefficient takes into account a wider range of pixel value fluctuations. For this reason, the edge maintenance performance can be improved so that the edges in the image are not blurred by noise removal. In addition, since the calculation is performed using an image signal having a resolution lower than that of the original image, the calculation amount is small and the calculation cost can be suppressed.
- the fifth embodiment is a case where denoising processing is performed on a high-frequency component obtained by wavelet transformation.
- the wavelet transform layer is arbitrary, but here, the case of one layer will be described.
- FIG. 9 is a conceptual diagram showing an image signal processing device of the fifth embodiment.
- the image signal processing apparatus of this embodiment includes an image signal input unit 041, an image signal processing unit 042, and an image signal output unit 043 as shown in FIG.
- the image signal input unit 041 and the image signal output unit 043 are the same as the corresponding parts in the second embodiment.
- the image signal processing unit 042 includes a WT unit 141, an IWT unit 241, a first STD unit 341, a second STD unit 342, a first TC degeneration unit 441, as shown in FIG. A second TC degeneration unit 442.
- the image signal processing unit 042 further includes a first combining unit 541, a second combining unit 542, and a WC (Wavelet).
- the image signal processing unit 042 further includes a first filter coefficient calculation unit 741, a second filter coefficient calculation unit 742, and a filter coefficient synthesis unit 840.
- the difference from the image signal processing apparatus of the third embodiment is that the portion surrounded by a dotted line in FIG. 5B and indicated as “characteristic location”, that is, the high-frequency component by wavelet transform in the WT unit 141 is the WC degeneration unit 641. This is the part that performs degeneration processing. Others are the same as in the case of the third embodiment, so the description thereof will be omitted here, and the parts related to the above-described characteristic portions will be described.
- the original image signal f in is input from the image signal input unit 041, by applying the wavelet transform of one layer, the low-frequency components of the wavelet transform LL 1 and the high-frequency component LH 1, HL 1, HH 1 is calculated. Then, a low-frequency component LL 1 to the first STD 341, and outputs the high frequency component LH 1, HL 1, HH 1 to WC degeneracy unit 641.
- the WC degeneration unit 641 applies the degeneration process of the equation (1) or the equation (2) to the high frequency components LH 1 , HL 1 , and HH 1 of the wavelet transform input from the WT unit 141 to correct the corrected high frequency Components LH 1 ′, HL 1 ′, and HH 1 ′ are calculated. Then, the corrected high frequency components LH 1 ′, HL 1 ′, and HH 1 ′ are output to the IWT unit 241.
- the IWT unit 241 uses the inverse wavelet transform to generate a reconstructed image signal LL 0 ′ from which noise has been removed.
- the generation is based on the corrected low frequency component LL 1 ′ input from the first synthesis unit 541 and the corrected high frequency components LH 1 ′, HL 1 ′, HH 1 ′ input from the WC degeneration unit 641. Do.
- the reconstructed image signal LL 0 ' is the same resolution image between the input image f in. Then, the reconstructed image signal LL 0 ′ from which noise has been removed is output to the second STD unit 342.
- FIG. 10 is a flowchart showing the operation of the image signal processing method of the fifth embodiment.
- FIG. 11 is a conceptual diagram showing an image signal processing device of the sixth embodiment. Here, a case where the wavelet transformation to be performed is in three stages will be described.
- the image signal processing apparatus of the sixth embodiment includes an image signal input unit 051, an image signal processing unit 052, and an image signal output unit 053.
- the image signal input unit 051 and the image signal output unit 053 are the same as those of the image signal processing apparatus of the second embodiment.
- the image signal processing unit 052 includes a first WT unit 151, a second WT unit 152, a third WT unit 153, a first WC degeneration unit 651, and a second WC degeneration, as shown in FIG. Unit 652, third WC degeneration unit 653, and first IWT unit 251.
- the image signal processing unit 052 further includes a second IWT unit 252, a third IWT unit 253, a first STD unit 351, a second STD unit 352, a third STD unit 353, and a fourth STD unit 354.
- the image signal processing unit 052 further includes a second TC degeneration unit 452, a third TC degeneration unit 453, and a fourth TC degeneration unit 454.
- first synthesis unit 551, the second synthesis unit 552, the third synthesis unit 553, the fourth synthesis unit 554, the first filter coefficient calculation unit 751, the second filter coefficient calculation unit 752, and the third A filter coefficient calculation unit 753 and a fourth filter coefficient calculation unit 754 are provided.
- a first filter coefficient synthesis unit 851, a second filter coefficient synthesis unit 852, and a third filter coefficient synthesis unit 853 are provided.
- wavelet transform with three layers is shown, but the number of layers of wavelet transform can be arbitrarily set.
- First WT unit 151 the original image signal f in is input from the image signal input unit 051, by applying the wavelet transform of one layer, the low-frequency component LL 1 and the high-frequency component LH 1 of the wavelet transform, HL 1 , HH 1 is calculated. Then, a low-frequency component LL 1 to the second WT unit 152, and outputs the high frequency component LH 1, HL 1, HH 1 to WC degeneracy unit 653.
- the low frequency component LL 1 is an image signal having a resolution one layer lower than the original image signal fin in terms of the wavelet transform.
- the second WT unit 152 applies a one-layer wavelet transform to the low-frequency component LL 1 of the first-layer wavelet transform input from the first WT unit 151, so that the low-frequency component LL 2 and the high-frequency component LH are applied. 2 , HL 2 and HH 2 are calculated. Then, a low frequency component LL 2 Third WT 153 outputs the high-frequency component LH 2, HL 2, HH 2 Second WC degeneracy unit 452.
- the low frequency component LL 2 is an image signal having a resolution one layer lower than the low frequency component LL 1 in terms of wavelet transform. That is, the low-frequency component LL 2, from the original image signal f in, an image signal having a wavelet transform on two levels lower resolution.
- the third WT unit 153 applies one-layer wavelet transform to the low-frequency component LL 2 of the second-layer wavelet transform input from the second WT unit 152, and thereby applies the low-frequency component LL 3 and the high-frequency component LH. 3 , HL 3 and HH 3 are calculated. Then, a low frequency component LL 3 First filter coefficient calculating section 751, and outputs the high frequency component LH 3, HL 3, HH 3 to WC degeneracy unit 651.
- the low frequency component LL 3 is an image signal having a resolution one layer lower than the low frequency component LL 2 in the wavelet transform. That is, the low frequency component LL 3 is an image signal having a resolution three layers lower than the original image signal fin in terms of the wavelet transform.
- the first filter coefficient calculation unit 751 calculates a filter coefficient using the low-frequency component LL 3 of the wavelet transform input from the third WT unit 153. That is, in DTVF, a set of filter coefficients calculated by Expression (6) and Expression (7) Is output to the first filter coefficient synthesis unit 851.
- the low frequency component LL 3 is output to the first STD unit 351.
- the first STD unit 351 processes the low-frequency component LL 3 of the third-layer wavelet transform input from the third WT unit 153 by the same processing as the STD unit 321 in the fifth embodiment, and the structure component u 3 and the texture. separating the components v 3. Then, the structure component u 3 is output to the first synthesis unit 551, and the texture component v 3 is output to the first TC degeneration unit 451.
- First TC degeneracy unit 451 with respect to the texture component v 3 which is input from the STD 351, noise is generated a texture component v 3 'suppression. Then, the generated texture component v 3 ′ is output to the first synthesis unit 551.
- the first synthesizing unit 551 corrects the third texture component u 3 input from the first STD unit 351 and the texture component v 3 ′ in which noise input from the first TC degeneration unit 451 is suppressed.
- the low-frequency component LL 3 ′ of the hierarchical wavelet transform is generated. The generation is performed by the same processing as that of the synthesis unit 521 in the fifth embodiment. Then, the corrected low-frequency component LL 3 ′ of the third-layer wavelet transform is output to the first IWT unit 251.
- the first WC degeneration unit 651 performs the corrected third-layer wavelet transform high-frequency component LH on the third-layer wavelet transform high-frequency component LH 3 , HL 3 , HH 3 input from the third WT unit 153. 3 ′, HL 3 ′, and HH 3 ′ are calculated. Then, the corrected high-frequency components LH 3 ′, HL 3 ′, and HH 3 ′ of the third-layer wavelet transform are output to the first IWT unit 251.
- the first IWT unit 251 performs low-level wavelet transformation of the second layer in which noise is removed from the low-frequency component LL 3 ′ and the high-frequency components LH 3 ′, HL 3 ′, and HH 3 ′ using inverse wavelet transformation.
- a frequency component LL 2 ′ is generated.
- the low-frequency component LL 3 ′ is a corrected low-frequency component of the third-layer wavelet transform input from the first synthesis unit 551.
- the high frequency components LH 3 ′, HL 3 ′, and HH 3 ′ are the high frequency components of the wavelet transform of the third layer after correction input from the first WC degeneration unit 651. Then, the low-frequency component LL 2 ′ of the second-layer wavelet transform from which noise has been removed is output to the second filter coefficient calculation unit 752.
- the second filter coefficient calculation unit 752 outputs a set of filter coefficients h 2 from the low frequency component LL 2 ′ to the first filter coefficient synthesis unit 851.
- the low frequency component LL 2 ′ is output to the second STD unit 352.
- the first filter coefficient synthesis unit 851 synthesizes the filter coefficient sets h 3 and h 2 to create a filter coefficient set h 2 ′, and outputs the filter coefficient set h 2 ′ to the second STD unit 352.
- the second STD unit 352 separates the low-frequency component LL 2 ′ of the corrected second-layer wavelet transform input from the first IWT unit 251 into the structure component u 2 and the texture component v 2 .
- the separation is performed using a set of filter coefficients h 2 ′ input from the first filter coefficient synthesis unit 851.
- the structure component u 2 is output to the second synthesis unit 552 and the texture component v 2 is output to the second TC degeneration unit 452.
- Second TC degeneracy unit 452 with respect to the texture component v 2 input from the second STD 352, noise is generated a texture component v 2 'suppression. Then, the generated texture component v 2 ′ is output to the second synthesis unit 552.
- the second synthesizing unit 552 corrects the second corrected structure component u 2 input from the second STD unit 352 and the texture component v 2 ′ from which the noise input from the second TC degeneration unit 452 is suppressed.
- a low-frequency component LL 2 ′′ of the hierarchical wavelet transform is generated. Then, the corrected low frequency component LL 2 ′′ is output to the second IWT unit 252.
- the first filter coefficient calculation unit 751 and the first STD unit 351 can be shared.
- Two or three of the second filter coefficient calculation unit 752, the first filter coefficient synthesis unit 851, and the second STD unit 352 can be made common.
- Two or three of the third filter coefficient calculation unit 753, the second filter coefficient synthesis unit 852, and the third STD unit 353 may be shared.
- Two or three of the fourth filter coefficient calculation unit 754, the third filter coefficient synthesis unit 853, and the fourth STD unit 354 may be made common.
- FIG. 12 is a flowchart showing the operation of the image signal processing method of the sixth embodiment.
- the case where the wavelet transformation to be performed is in the L stage will be described.
- L 3.
- L-layer multi-resolution wavelet transform is applied to the input image (S501).
- a set of filter coefficients is calculated using the low-frequency component of the l-th layer (S503).
- the structure / texture decomposition is applied to the low-frequency component of the l-th layer in the procedure described with respect to the first STD unit 351 to obtain the structure component and texture component of the low-frequency component of the l-th layer (S504).
- a texture component in which noise is suppressed is obtained from the texture component of the low frequency component of the l-th layer in the procedure described for the first TC degeneration unit 451 (S505).
- the structure component of the l-th layer low frequency component and the texture component of the l-th layer in which noise is suppressed are synthesized by the procedure described for the first synthesis unit 551, and the corrected low-frequency component of the l-th layer is synthesized.
- a reduction process is applied to the high frequency component obtained by the wavelet transform in the procedure described with respect to the WC reduction unit 651 to obtain a corrected high frequency component (S507).
- the inverse wavelet transform is applied using the low-frequency component and the high-frequency component of the l-th layer corrected in this way to obtain the low-frequency component of the l-1-th layer (S508).
- the value of l is decreased by one in order to raise the target hierarchy (S509).
- a set of filter coefficients from the low-frequency component of the l-th layer (because the value of l is reduced by 1 in S509, it becomes a low-frequency component of the layer one level higher than the low-frequency component used in S503). Is calculated (S510).
- the set of filter coefficients calculated in S503 and the set of filter coefficients calculated in S511 are combined (S511).
- the synthesized filter coefficients for the l-th layer low frequency component applying the procedure described with respect to the second STD unit 352 to perform structure texture decomposition, the l-th layer low frequency component structure component And a texture component are obtained (S512).
- the procedure described for the second TC degeneration unit 452 is applied to the texture component of the low-frequency component of the l-th layer to obtain the texture component of the low-frequency component of the l-th layer in which noise is suppressed (S513).
- the low-frequency component of the l-th layer corrected by combining the structure component of the low-frequency component of the l-th layer and the texture component of the low-frequency component of the l-th layer in which noise is suppressed is generated (S514). ). If l is larger than 0, the process returns to immediately before S507 to continue the processing (S515-Yes). If l is 0, the process ends (S515-No).
- each unit can be configured by hardware, but can also be realized by a computer program.
- functions and operations similar to those of the above-described embodiment are realized by a processor that operates according to a program stored in the program memory.
- only some functions of the above-described embodiments can be realized by a computer program.
- the seventh embodiment is an embodiment relating to an image signal processing apparatus in which the technique of the fourth embodiment is combined with the sixth embodiment.
- FIG. 13 is a conceptual diagram showing an image signal processing apparatus according to the seventh embodiment. Here, the case where the hierarchy of the wavelet transform to be performed is three stages will be described.
- the image signal processing apparatus includes an image signal input unit 061, an image signal processing unit 062, and an image signal output unit 063.
- the image signal input unit 061 and the image signal output unit 063 are the same as those of the image signal processing apparatus of the second embodiment.
- the image signal processing unit 062 includes a first WT unit 161, a second WT unit 162, a third WT unit 163, a first WC degeneration unit 661, and a second WC degeneration, as shown in FIG. Unit 662, third WC degeneration unit 663, and first IWT unit 261. Furthermore, a second IWT unit 262, a third IWT unit 263, a first STD unit 361, a second STD unit 362, a third STD unit 363, and a fourth STD unit 364 are provided.
- the first TC degeneration unit 461, the second TC degeneration unit 462, the third TC degeneration unit 463, the fourth TC degeneration unit 464, the first combination unit 561, the second combination unit 562, and the third A synthesis unit 563 and a fourth synthesis unit 564 are provided.
- the first filter coefficient calculation unit 761, the second filter coefficient calculation unit 762, the third filter coefficient calculation unit 763, the first gradient strength calculation unit 961, the second gradient strength calculation unit 962, and the third gradient An intensity calculation unit 963 and a fourth gradient intensity calculation unit 964 are provided.
- a first gradient intensity combining unit 1061, a second gradient intensity combining unit 1062, and a third gradient intensity combining unit 1063 are provided.
- an example of wavelet transform with three layers is shown, but the number of layers of wavelet transform can be arbitrarily set.
- strength calculation part 961 and the 1st STD part 361 can also be made common.
- Two or more of the first filter coefficient calculation unit 761, the second gradient strength calculation unit 962, the first gradient strength synthesis unit 1061, the first filter correction unit 1161, and the second STD unit 362 may be shared.
- Two or more of the second filter coefficient calculation unit 762, the third gradient strength calculation unit 963, the second gradient strength synthesis unit 1062, the second filter correction unit 1162, and the third STD unit 363 may be shared.
- FIG. 14 is a flowchart showing the image signal processing method of the present embodiment.
- the case where the layer of the wavelet transform to be performed is in the L stage will be described.
- L 3.
- L-layer multi-resolution wavelet transform is applied to the input image (S601).
- a set of gradient intensities is calculated from the low frequency components of the l-th layer (S603).
- the structure / texture decomposition is applied to the low-frequency component of the l-th layer to obtain the structure component and texture component of the low-frequency component of the l-th layer (S604). Further, a texture component in which noise is suppressed is obtained from the texture component of the low frequency component of the l-th layer (S605). Then, the structure component of the low frequency component of the l-th layer and the texture component of the l-th layer in which noise is suppressed are combined to obtain a corrected low-frequency component of the l-th layer (S606). On the other hand, a corrected high frequency component is obtained for the high frequency component obtained by the wavelet transform (S607).
- the inverse wavelet transform is applied using the low-frequency component and the high-frequency component of the l-th layer corrected in this way to obtain the low-frequency component of the l-1-th layer (S608). Then, the value of l is decreased by one in order to raise the target hierarchy (S609).
- a set of gradient intensities is calculated from the low frequency components of the l-th layer (S610).
- a set of filter coefficients is calculated from the low frequency components of the l-th layer (S611). Further, the set of gradient strengths obtained in S603 and the set of gradient strengths obtained in S610 are combined (S612).
- the set of filter coefficients obtained in S611 is corrected using the set of gradient strengths obtained in S612 (S613).
- the structure / texture decomposition is applied to the low frequency component of the l-th layer to obtain the structure component and texture component of the low-frequency component of the l-th layer (S614). Further, the texture component of the low frequency component of the l-th layer in which noise is suppressed is obtained for the texture component of the low-frequency component of the l-th layer (S615). Then, the low-frequency component of the l-th layer corrected by synthesizing the structure component of the low-frequency component of the l-th layer and the texture component of the low-frequency component of the l-th layer in which noise is suppressed is generated (S616). .
- the image signal processing device of this embodiment is obtained by adding the WC degeneration unit of the fifth embodiment to the image signal processing device of the fourth embodiment and then expanding the image signal processing to multi-resolution. Therefore, in addition to the effects of the image signal processing apparatuses of the fourth embodiment and the fifth embodiment, an effect of enabling more effective noise removal with respect to noise of various frequencies is obtained.
- An image signal input unit for inputting an image signal of the original image;
- a wavelet transform unit that generates a low-frequency component and a high-frequency component by wavelet transforming the image signal;
- a first structure / texture separation unit for separating the low-frequency component into a first structure component and a first texture component;
- a texture component degeneration unit that removes noise with respect to the first texture component and generates the first texture component after the processing;
- a first synthesis unit that synthesizes the first structure component and the corrected first texture component, and generates a synthesized low-frequency component;
- An inverse wavelet transform unit that generates an image signal after inverse wavelet transform by performing inverse wavelet transform on the high frequency component and the synthesized low frequency component;
- a second structure / texture separation unit that separates the image signal after the inverse wavelet transform into a second structure component and a second texture component; Removing a noise with respect to
- the image signal processing device further comprising: (Appendix 3)
- the second structure / texture separation unit further includes second filter coefficient calculation means for calculating a filter coefficient for separating the second structure component and the second texture component from the low frequency component. 3.
- the image signal processing apparatus according to 2. (Appendix 4) In the second structure / texture separation unit, the filter coefficient when the second structure component and the second texture component are separated into the filter coefficient calculated by the first filter coefficient calculation unit, and the second The image signal processing apparatus according to appendix 3, further comprising filter coefficient synthesis means that is obtained by synthesizing the filter coefficients calculated by the filter coefficient calculation means.
- a second wavelet transform unit that generates a second low frequency component and a second high frequency component by wavelet transforming the low frequency component;
- a third structure / texture separation unit for separating the second low-frequency component into a third structure component and a third texture component;
- a third texture component degeneration unit that corrects the value of the third texture component and generates a corrected third texture component;
- a third combining unit that combines the third structure component and the corrected third texture component to generate a combined second low-frequency component;
- An inverse wavelet transform unit that generates an image signal after the second inverse wavelet transform by performing an inverse wavelet transform on the second high frequency component and the second low frequency component after the synthesis;
- a fourth structure / texture separation unit for separating the image signal after the second inverse wavelet transform into a fourth structure component and a fourth texture component;
- a fourth texture component degeneration unit that corrects the value of the fourth texture component and generates a corrected fourth texture component;
- the image signal processing apparatus according to appendix 12, further comprising second gradient strength combining means that is obtained by combining the gradient strength of the pixel value at each point in the image.
- the image signal processing device according to any one of appendices 9 to 13, further comprising a second wavelet coefficient degeneracy unit that performs a process of removing noise of the second high-frequency component.
- An image signal input unit for inputting an image signal of the original image;
- a first wavelet transform unit that generates a first low-frequency component and a first high-frequency component by performing wavelet transform on the image signal;
- a second wavelet transform unit that generates a second low-frequency component and a second high-frequency component by wavelet transforming the low-frequency component;
- Noise removing means for removing noise from the first low frequency component using the second low frequency component and the second high frequency component;
- a first inverse wavelet transform unit for generating an image signal after wavelet transform;
- a first structure / texture separation unit that separates the image signal after the inverse wavelet transform into a first structure component and a first texture component;
- a first texture component degeneration unit that removes noise with respect to the first texture component and generates the first texture component after the processing;
- a first combining unit that combines the first structure component and the corrected first texture component to generate an image signal processing signal;
- An image signal output unit for outputting the corrected image signal;
- An image signal processing apparatus comprising: (Appendix 16)
- the noise removing means is A second structure / texture separation unit for separating the second low-frequency component into a second structure component and a second texture component; Removing a noise with respect to the second texture component and generating a second texture component after the processing;
- a second synthesis unit for synthesizing the second structure component and the corrected second texture component to generate an image signal processing signal;
- the image signal processing device further comprising: (Appendix 17
- the filter coefficient synthesizing unit further synthesizes the filter coefficient calculated by the first filter coefficient calculating unit and the filter coefficient calculated by the second filter coefficient calculating unit, and the filter coefficient synthesized by the filter coefficient synthesizing unit is 21.
- the image signal processing device further comprising: a second gradient intensity calculating unit that is obtained from two low-frequency components or a processed second low-frequency component obtained by processing the second low-frequency component.
- the gradient intensity of the pixel value at each point in the image for correction is calculated by the first gradient intensity number calculating means, and the gradient intensity of the pixel value at each point in the image and the second gradient intensity number are calculated.
- the image signal processing apparatus according to appendix 23, further comprising gradient intensity combining means that is obtained by combining the gradient intensity of the pixel value at each point in the image calculated by the calculating means.
- (Appendix 25) The image signal processing device according to any one of appendices 2 to 8, 10 to 14, and 18 to 24, wherein a filter coefficient of a nonlinear low-pass filter is used as the filter coefficient.
- Appendix 26 The image signal processing apparatus according to appendix 8, wherein the correction of the high frequency component in the wavelet coefficient reduction unit corrects the high frequency component to an absolute value or less without changing a sign.
- Appendix 27 The image signal processing apparatus according to supplementary note 14, wherein the correction of the second high frequency component in the second wavelet coefficient reduction unit corrects the high frequency component to an absolute value or less without changing a sign.
- the composition ratio is determined by the distribution of local pixel values. 25.
- a filter coefficient calculating unit configured to calculate a filter coefficient used for separating the second structure component and the second texture component; the filter coefficient between the target pixel and a pixel adjacent to the target pixel; An indicator based on the amount of change,
- the image signal processing device according to appendix 1, wherein the image signal processing device is calculated by combining indices based on gradient strength and gradient direction of pixel values in a range of several pixels centered on a pixel of interest.
- Filter coefficient calculation means for calculating a filter coefficient used for separating the second structure component and the second texture component, and the filter coefficient between the target pixel and a pixel adjacent to the target pixel.
- An indicator based on the amount of change The image signal processing device according to appendix 1, wherein the image signal processing device is calculated by combining indices based on gradient strength and gradient direction of pixel values in a range of several pixels centered on a pixel of interest.
- the apparatus further comprises filter coefficient calculation means for calculating a filter coefficient used when separating the first structure component and the first texture component, and the filter coefficient is calculated between the target pixel and a pixel adjacent to the target pixel.
- An indicator based on the amount of change The image signal processing device according to appendix 15, wherein the image signal processing device is calculated by combining indices based on gradient strength and gradient direction of pixel values in a range of several pixels centered on the target pixel.
- (Appendix 35) Filter coefficient calculation means for calculating a filter coefficient used for separating the second structure component and the second texture component, and the filter coefficient is centered on a target pixel calculated by the low frequency component
- the image signal processing device according to appendix 1, wherein the image signal processing unit calculates the gradient based on the gradient intensity and gradient direction of the pixel values in the several pixel range.
- (Appendix 36) The supplementary note 4, wherein a synthesis ratio for synthesizing the filter coefficient calculated by the first filter coefficient calculating means and the filter coefficient calculated by the second filter coefficient calculating means is determined by local pixel value dispersion. Image signal processing apparatus.
- Appendix 37 The image signal processing apparatus according to appendix 1, wherein the structure / texture separation unit further includes a nonlinear low-pass filter.
- Appendix 38 Means for multi-resolution decomposition of the original image signal using wavelet transform; Using the edge-preserving nonlinear low-pass filter in order from the lowest resolution, the low-frequency component of the resolution is divided into a structure component composed of edges and flat components in the image, and a texture component composed of noise and fine patterns.
- Means for separating Means for correcting each value of the separated texture component below its absolute value without changing the sign; Means for combining the structure component and the corrected texture component to generate a corrected low frequency component of the resolution; When the resolution is lower than the original image resolution, a low frequency component having a resolution one higher than the resolution is obtained by inverse wavelet transform from the corrected low frequency component at the resolution and the high frequency component at the resolution.
- An image signal processing apparatus comprising means for repeating until the resolution is the same as the original image,
- the filter coefficient of the nonlinear low-pass filter calculated at a resolution lower than the resolution is used as the filter coefficient of the nonlinear low-pass filter of the resolution.
- the structure / texture separation unit includes a non-linear low-pass filter, and the filter coefficient of the non-linear low-pass filter is an index based on a pixel of interest and a variation amount between pixels adjacent to the pixel of interest;
- the image signal processing apparatus according to appendix 1, wherein the image signal processing apparatus is calculated by combining indices based on gradient strength and gradient direction of pixel values in a range of several pixels centered on a target pixel.
- (Appendix 40) Inputting an image signal of the original image; Generating a low frequency component and a high frequency component by wavelet transforming the image signal; Separating the low frequency component into a first structure component and a first texture component; Removing noise from the first texture component and generating a first texture component after removing the noise; Synthesizing the first structure component and the first texture component after removing the noise, and generating a low-frequency component after synthesis; Generating an image signal after inverse wavelet transform by performing inverse wavelet transform on the high frequency component and the synthesized low frequency component; Separating the image signal after the inverse wavelet transform into a second structure component and a second texture component; Removing noise from the second texture component and generating a second texture component after removing the noise; Combining the second structure component and the corrected second texture component to generate an image signal processing signal; Outputting the corrected image signal;
- An image signal processing method comprising: (Appendix 41) Processing to input the image signal of the original image; Processing to generate a low
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Image Processing (AREA)
- Facsimile Image Signal Circuits (AREA)
Abstract
Description
式(3)を解く方法としては、非特許文献5のDigital TV Filter(DTVF)がある。今、画像uの画素位置α=(i、j)における画素値をuαと表記する。また、αの近傍画素位置の集合をN(α)と表記する。近傍を8近傍とするとき、
N(α)={(i、j±1)、(i±1、j)、(i+1、j±1)、(i-1、j±1)}
である。DTVFでは、式(4)を解くために、局所変動量に基づくフィルタ処理を用いる。入力画像信号をu (0)、N回フィルタ後の出力画像信号をu (N)とすると、画素位置αにおけるフィルタ出力uα (N)は下記式(5)で表される。
を用いる。
そして、IWT部2005において、最低解像度の1つ上の階層におけるノイズの抑圧された低周波成分LL2’の生成を、逆ウェーブレット変換によって行う。ここで、IWTは、Inverse WT、すなわち、逆ウェーブレット変換である。その生成は、最低解像度の低周波成分LL3と、縮退処理が適用された最低解像度の高周波成分LH3’、HL3’、HH3’とから行う。
第一実施形態は、本発明の最小構成の画像信号処理装置に関する実施形態である。
[本実施形態の画像信号処理装置]
図1は第一実施形態の画像信号処理装置を表わす概念図である。同図(b)に示す構成が、本発明の最小構成である。
[本実施形態の画像信号処理方法]
続いて、本実施形態の画像信号処理方法について説明する。
[本実施形態の効果]
本実施形態の画像信号処理装置では、デノイズ処理の際のストラクチャ・テクスチャ分解に用いるフィルタ係数を、ウェーブレット変換を用いて生成する原画像よりノイズの少ない画像から算出することができる。ウェーブレット変換における低周波成分は、原画像の画像信号より解像度の低い画像信号である。原画像より低い解像度の画像信号から算出されたフィルタ係数は、原画像信号と同じ解像度の画像信号から算出されるフィルタ係数より、より広範囲の画素値を参照して生成されている。また、原画像より低い解像度の画像信号は、当該解像度の画像よりノイズが抑圧されているため、フィルタ係数算出時におけるノイズの影響が原画像信号と同じ解像度の画像信号より少ない。よって、この低周波成分に基づいてストラクチャ・テクスチャ分解に用いるフィルタ係数を求めると、ノイズの影響で画像を構成するエッジを不鮮明にするようなフィルタ係数を設定することが少なくなる。そのため、低周波成分に含まれるエッジ成分を維持しつつ、低周波成分のノイズを効果的に抑圧できる。さらに、ノイズを抑圧した低周波成分からウェーブレット逆変換を用いて原画像と同じ解像度の画像を生成すると、この画像は原画像よりノイズが抑圧されているため、フィルタ係数算出時におけるノイズの影響が原画像信号より少ない。そのため、ノイズの影響で画像を構成するエッジを不鮮明にするようなフィルタ係数を設定することが少なくなりエッジ成分を維持しつつ、ノイズを効果的に抑圧できる。加えて、本実施形態の画像信号処理装置は、広範囲の画像特徴量の計算結果を用いてフィルタ係数の補正を行う必要がないため、計算コストの増大を抑えることができる。
<第二実施形態>
第二実施形態は、ある解像度の画像信号に対するストラクチャ・テクスチャ分解に用いるフィルタ係数を算出する画像信号処理装置の実施形態である。本実施形態においてウェーブレット変換の階層は任意であるが、ここでは1階層の場合について説明する。
[本実施形態の画像信号処理装置]
図3は、第二実施形態の画像信号処理装置を表わす概念図である。
[本実施形態の画像信号処理方法]
続いて、本実施形態の画像信号処理方法について説明する。
[本実施形態の効果]
本実施形態の画像信号処理装置では、デノイズ処理の際のストラクチャ・テクスチャ分解に用いるフィルタ係数を、ウェーブレット変換を用いて生成する原画像よりノイズの少ない画像から算出することができる。ウェーブレット変換における低周波成分は、原画像の画像信号より解像度の低い画像信号である。原画像より低い解像度の画像信号から算出されたフィルタ係数は、原画像信号と同じ解像度の画像信号から算出されるフィルタ係数より、より広範囲の画素値を参照して生成されている。また、原画像より低い解像度の画像信号は、当該解像度の画像よりノイズが抑圧されているため、フィルタ係数算出時におけるノイズの影響が原画像信号と同じ解像度の画像信号より少ない。よって、この低周波成分に基づいてストラクチャ・テクスチャ分解に用いるフィルタ係数を求めると、ノイズの影響で画像を構成するエッジを不鮮明にするようなフィルタ係数を設定することが少なくなる。そのため、低周波成分に含まれるエッジ成分を維持しつつ、低周波成分のノイズを効果的に抑圧できる。さらに、ノイズを抑圧した低周波成分からウェーブレット逆変換を用いて原画像と同じ解像度の画像を生成すると、この画像は原画像よりノイズが抑圧されているため、フィルタ係数算出時におけるノイズの影響が原画像信号より少ない。そのため、ノイズの影響で画像を構成するエッジを不鮮明にするようなフィルタ係数を設定することが少なくなりエッジ成分を維持しつつ、ノイズを効果的に抑圧できる。加えて、本実施形態の画像信号処理装置は、広範囲の画像特徴量の計算結果を用いてフィルタ係数の補正を行う必要がないため、計算コストの増大を抑えることができる。
<第三実施形態>
第三実施形態は、ある解像度の画像信号に対するストラクチャ・テクスチャ分離をする際のフィルタ係数を、低解像度な低周波成分から算出したフィルタ係数を用いて算出する画像信号処理装置の実施形態である。ここで、ウェーブレット変換の階層は任意であるが、ここでは1階層の場合について説明する。
[本実施形態の画像信号処理装置]
図5は、第三実施形態の画像信号処理装置を表わす概念図である。
続いて、本実施形態の画像信号処理方法について説明する。
[本実施形態の効果]
本実施形態の画像信号処理装置では、ストラクチャ・テクスチャ分解に用いるフィルタ係数を、画像信号にウェーブレット変換を適用して得られた低周波成分から算出されるフィルタ係数を合成することで生成する。低周波成分は、原画像の画像信号より解像度の低い画像信号である。原画像より低い解像度の画像信号から算出されたフィルタ係数は、原画像信号と同じ解像度の画像信号から算出されるフィルタ係数より、より広範囲の画素値を参照して生成されている。また、原画像より低い解像度の画像信号は、当該解像度の画像よりノイズが抑圧されているため、フィルタ係数算出時におけるノイズの影響が原画像信号と同じ解像度の画像信号より少ない。よって、ノイズの影響で画像を構成するエッジを不鮮明にするようなフィルタ係数を設定することが少なくなる。加えて、本実施形態の画像信号処理装置は、フィルタ係数の補正に必要な広範囲の画像特徴量の計算を行う必要がないため、計算コストの増大を抑えることができる。
<第四実施形態>
第四実施形態は、ストラクチャ・テクスチャ分離処理時に用いるフィルタの集合を、画像を構成する各点の画素値の勾配強度の集合(以下、「勾配強度の集合」という。)により補正する画像信号処理装置に関する実施形態である。
[本実施形態の画像信号処理装置]
図7は、第四実施形態の画像信号処理装置を表わす概念図である。
である。勾配強度の集合g1は勾配強度合成部1030に出力される。低周波成分LL1は第一STD部331に出力される。
で表される。ここで、c0は合成比率であり、0≦c0≦1である。c0はユーザが設定するパラメータとして与えてもよいし、適応的に定まる値としてもよい。たとえば、注目画素とその周辺の画素の画素値の分散を計算し、分散が小さい領域ほど、当該解像度より低い解像度のフィルタ係数の重みを大きくするようにしてもよい。
[本実施形態の画像信号処理方法]
図8は、第四実施形態の画像信号処理方法の動作を表わすフローチャートである。同図では、第三実施形態の画像信号処理装置と異なる点を、点線で囲み、「特徴箇所」と表示してある。すなわち、以下の点が異なる。
第一勾配強度算出部931は、低周波成分から勾配強度の集合を算出する(S302)。
フィルタ係数算出部730は、逆ウェーブレット変換後の画像信号からフィルタ係数の集合を算出する(S308)。
第二勾配強度算出部932は、逆ウェーブレット変換後の画像信号から勾配強度の集合を算出する(S309)。
勾配強度合成部1030は、低周波成分から算出した勾配強度の集合と、逆ウェーブレット変換後の画像信号から算出した勾配強度の集合とを合成する(S310)。
フィルタ補正部1130は、S310で合成した勾配強度の集合を用いて、S308で求めたフィルタ係数の集合を補正する(S311)。
第二STD部332は、S311で補正したフィルタ係数の集合を用いて、逆ウェーブレット変換後の画像信号にストラクチャ・テクスチャ分解を適用する(S312)。
本実施形態の画像信号処理装置は、原画像より低い解像度の画像信号である低周波成分について算出した勾配強度の集合と、原画像と同じ解像度の画像信号である再構成画像信号を用いて算出した勾配強度の集合とを合成する。そして、合成した勾配強度の集合を用いてフィルタ係数の集合を補正する。ここで、原画像と同じ解像度の画像信号である再構成画像信号を用いて算出した勾配強度は、隣接する画素間との局所的な変動量を評価する指標であり。また、原画像より低い解像度の画像信号である低周波成分について算出した勾配強度は、広範囲の画素値の変動を評価する指標である。従い、補正後のフィルタ係数は、より広範囲の画素値変動を考慮したものとなっている。このため、画像中のエッジがノイズ除去により不鮮明にならないようにエッジの維持性能を向上させることができる。加えて、原画像より低い解像度の画像信号を用いての計算なので、計算量が少なく、計算コストを抑えることができる。
<第五実施形態>
第五実施形態は、ウェーブレット変換により得られた高周波成分にデノイズ処理を行う場合である。なお、本実施形態において、ウェーブレット変換の階層は任意であるが、ここでは1階層の場合について説明する。
[本実施形態の画像信号処理装置]
図9は、第五実施形態の画像信号処理装置を表わす概念図である。
Coefficient(ウェーブレット係数))縮退部641と、を備える。画像信号処理部042は、さらに、第一フィルタ係数算出部741と、第二フィルタ係数算出部742と、フィルタ係数合成部840と、を備える。
[本実施形態の画像信号処理方法]
図10は、第五実施形態の画像信号処理方法の動作を表わすフローチャートである。同図では、第三実施形態の画像信号処理装置と異なる点を、点線で囲み、「特徴箇所」と表示してある。すなわち、WC縮退部641は、本実施形態は、ウェーブレット変換で得られた高周波成分に対しては、縮退処理を適用し、補正された高周波成分を得る(S405)という点が異なる。
[本実施形態の効果]
本実施形態は、まず、第三実施形態と同じ効果を得ることができる。加えて、本実施形態では、ウェーブレット変換後の高周波成分に対してもデノイズ処理をしているので、一層ノイズの少ない画像信号を得ることができる。
<第六実施形態>
第六実施形態は、第五実施形態におけるウェーブレット変換を多段階にするものである。
[本実施形態の画像信号処理装置]
図11は、第六実施形態の画像信号処理装置を表わす概念図である。ここでは、行うウェーブレット変換が3段階の場合について説明する。
を第一フィルタ係数合成部851に出力する。低周波成分LL3は第一STD部351に出力される。
は第二フィルタ係数合成部852に出力される。
図12は、第六実施形態の画像信号処理方法の動作を表わすフローチャートである。ここでは、行うウェーブレット変換がL段階の場合について説明する。図11に表した画像信号処理部052の場合はL=3である。
[本実施形態の効果]
本実施形態の画像信号処理装置は、第四実施形態の画像信号処理装置の画像信号処理を多重解像度に拡張したものである。従い、第四実施形態の画像信号処理装置の効果に加えて、様々な周波数のノイズに対してより有効なノイズ除去が可能という効果が得られる。
[第七実施形態]
第七実施形態は、第六実施形態に、第四実施形態の技術を組み合わせた画像信号処理装置に関する実施形態である。
[本実施形態の画像信号処理装置]
図13は、第七実施形態の画像信号処理装置を表わす概念図である。ここでは、行うウェーブレット変換の階層が3段階の場合について説明する。
[本実施形態の画像信号処理方法]
続いて、本実施形態の画像信号処理方法について説明する。図14は、本実施形態の画像信号処理方法を示すフローチャートである。ここでは、行うウェーブレット変換の階層がL段階の場合について説明する。図13に表した画像信号処理部062の場合はL=3である。
[本実施形態の効果]
本実施形態の画像信号処理装置は、第四実施形態の画像信号処理装置に第五実施形態のWC縮退部を追加した上で、その画像信号処理を多重解像度に拡張したものである。従い、第四実施形態及び第五実施形態の画像信号処理装置の効果に加えて、様々な周波数のノイズに対してより有効なノイズ除去が可能になるという効果が得られる。
(付記1)
原画像の画像信号を入力する画像信号入力部と、
前記画像信号をウェーブレット変換することにより低周波成分と高周波成分とを生成するウェーブレット変換部と、
前記低周波成分を、第一ストラクチャ成分と、第一テクスチャ成分とに分離する第一ストラクチャ・テクスチャ分離部と、
前記第一テクスチャ成分についてノイズを除去し、当該処理後の第一テクスチャ成分を生成するテクスチャ成分縮退部と、
前記第一ストラクチャ成分と前記補正後の第一テクスチャ成分とを合成し、合成後の低周波成分を生成する第一合成部と、
前記高周波成分と前記合成後の低周波成分とを逆ウェーブレット変換することにより、逆ウェーブレット変換後の画像信号を生成する逆ウェーブレット変換部と、
前記逆ウェーブレット変換後の画像信号を、第二ストラクチャ成分と、第二テクスチャ成分とに分離する第二ストラクチャ・テクスチャ分離部と、
前記第二テクスチャ成分についてノイズを除去し、当該処理後の第二テクスチャ成分を生成する第二テクスチャ成分縮退部と、
前記第二ストラクチャ成分と、前記補正後の第二テクスチャ成分とを合成し、画像信号処理信号を生成する第二合成部と、
前記補正後の画像信号を出力する画像信号出力部と、
を備える画像信号処理装置。
(付記2)
前記第二ストラクチャ・テクスチャ分離部における、前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際のフィルタ係数を、前記逆ウェーブレット変換後の画像信号から算出する第一フィルタ係数算出手段をさらに備える、付記1記載の画像信号処理装置。
(付記3)
前記第二ストラクチャ・テクスチャ分離部における、前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際のフィルタ係数を、前記低周波成分から算出する第二フィルタ係数算出手段をさらに備える、付記2記載の画像信号処理装置。
(付記4)
前記第二ストラクチャ・テクスチャ分離部における、前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際の前記フィルタ係数を、前記第一フィルタ係数算出手段により算出したフィルタ係数と、前記第二フィルタ係数算出手段により算出したフィルタ係数とを合成することにより求めるフィルタ係数合成手段をさらに備える、付記3記載の画像信号処理装置。
(付記5)
前記第二ストラクチャ・テクスチャ分離部における、前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際の前記フィルタ係数の補正をするための画像中の各点における画素値の勾配強度を、逆ウェーブレット変換後の画像信号から求める第一勾配強度算出手段をさらに備える、付記2記載の画像信号処理装置。
(付記6)
前記補正をするための画像中の各点における画素値の勾配強度を、前記低周波成分から求める第二勾配強度算出手段をさらに備える、付記5記載の画像信号処理装置。
(付記7)
前記補正をするための画像中の各点における画素値の勾配強度を、前記第一勾配強度数算出手段により算出した前記画像中の各点における画素値の勾配強度と、前記第二勾配強度数算出手段により算出した前記画像中の各点における画素値の勾配強度とを合成することにより求める勾配強度合成手段をさらに備える、付記6記載の画像信号処理装置。
(付記8)
前記高周波成分のノイズを除去するウェーブレット係数縮退部をさらに備える付記1乃至7のうちのいずれか1に記載された画像信号処理装置。
(付記9)
前記低周波成分をウェーブレット変換することにより第二低周波成分と第二高周波成分とを生成する第二ウェーブレット変換部と、
前記第二低周波成分を、第三ストラクチャ成分と、第三テクスチャ成分とに分離する第三ストラクチャ・テクスチャ分離部と、
前記第三テクスチャ成分についてその値を補正し、補正後の第三テクスチャ成分を生成する第三テクスチャ成分縮退部と、
前記第三ストラクチャ成分と、前記補正後の第三テクスチャ成分を合成し合成後の第二低周波成分を生成する第三合成部と、
前記第二高周波成分と前記合成後の第二低周波成分とを逆ウェーブレット変換することにより、第二逆ウェーブレット変換後の画像信号を生成する逆ウェーブレット変換部と、
前記第二逆ウェーブレット変換後の画像信号を、第四ストラクチャ成分と、第四テクスチャ成分とに分離する第四ストラクチャ・テクスチャ分離部と、
前記第四テクスチャ成分についてその値を補正し、補正後の第四テクスチャ成分を生成する第四テクスチャ成分縮退部と、
前記第四ストラクチャ成分と、前記補正後の第四テクスチャ成分とを合成し、第二画像信号処理信号を生成する第四合成部をさらに備える付記1乃至8のうちのいずれか一に記載の画像信号処理装置。
(付記10)
前記第三ストラクチャ・テクスチャ分離部における、前記第三ストラクチャ成分と、前記第三テクスチャ成分とに分離する際のフィルタ係数を、前記第二逆ウェーブレット変換後の画像信号から算出する第三フィルタ係数算出手段をさらに備える、付記9記載の画像信号処理装置。
(付記11)
前記第三ストラクチャ・テクスチャ分離部における、前記第三ストラクチャ成分と、前記第三テクスチャ成分とに分離する際の前記フィルタ係数を、前記フィルタ係数合成手段において合成したフィルタ係数と、前記第三フィルタ係数算出手段で算出したフィルタ係数とを合成する、第二フィルタ係数合成手段をさらに備える、付記10記載の画像信号処理装置。
(付記12)
前記第三ストラクチャ・テクスチャ分離部における、前記第三ストラクチャ成分と、前記第三テクスチャ成分とに分離する際の前記フィルタ係数の補正をするための画像中の各点における画素値の勾配強度を、前記第二逆ウェーブレット変換部における逆ウェーブレット変換後の画像信号から求める第三勾配強度算出手段をさらに備える、付記10記載の画像信号処理装置。
(付記13)
前記補正をするための画像中の各点における画素値の勾配強度を、前記勾配強度合成手段において合成した画像中の各点における画素値の勾配強度と、前記第三勾配強度数算出手段により算出した前記画像中の各点における画素値の勾配強度とを合成することにより求める第二勾配強度合成手段をさらに備える、付記12記載の画像信号処理装置。
(付記14)
前記第二高周波成分のノイズを除去する処理をする第二ウェーブレット係数縮退部をさらに備える付記9乃至13のうちのいずれか1に記載された画像信号処理装置。
(付記15)
原画像の画像信号を入力する画像信号入力部と、
前記画像信号をウェーブレット変換することにより第1低周波成分と第1高周波成分とを生成する第1ウェーブレット変換部と、
前記低周波成分をウェーブレット変換することにより第2低周波成分と第2高周波成分とを生成する第二ウェーブレット変換部と、
第2低周波成分と第2高周波成分を用いて第1低周波成分からノイズを除去するノイズ除去手段と、
前記ノイズ除去手段によりノイズを除去した第1低周波成分と前記第二高周波成分または前記第2高周波成分を処理して得られた処理後の第2高周波成分とを逆ウェーブレット変換することにより、逆ウェーブレット変換後の画像信号を生成する第1逆ウェーブレット変換部と、
前記逆ウェーブレット変換後の画像信号を、第1ストラクチャ成分と、第1テクスチャ成分とに分離する第1ストラクチャ・テクスチャ分離部と、
前記第1テクスチャ成分についてノイズを除去し、当該処理後の第1テクスチャ成分を生成する第1テクスチャ成分縮退部と、
前記第1ストラクチャ成分と、前記補正後の第1テクスチャ成分とを合成し、画像信号処理信号を生成する第1合成部と、
前記補正後の画像信号を出力する画像信号出力部と、
を備える画像信号処理装置。
(付記16)
前記ノイズ除去手段が、
前記第2低周波成分を第2ストラクチャ成分と第2テクスチャ成分とに分離する第2ストラクチャ・テクスチャ分離部と、
前記第2テクスチャ成分についてノイズを除去し、当該処理後の第2テクスチャ成分を生成する第2テクスチャ成分縮退部と、
前記第二ストラクチャ成分と、前記補正後の第二テクスチャ成分とを合成し、画像信号処理信号を生成する第2合成部と、
を備える付記15記載の画像信号処理装置。
(付記17)
前記ノイズ除去手段が、前記第2高周波成分のノイズを除去するウェーブレット係数縮退部を備える付記15または16に記載された画像信号処理装置。
(付記18)
前記ストラクチャ・テクスチャ分離と、逆ウェーブレット変換と、縮退処理の組み合わせにより前記低周波成分のノイズを除去するノイズ除去手段を備える、付記17記載の画像信号処理装置。
(付記19)
前記第1逆ウェーブレット変換後の画像信号から前記第1ストラクチャ・テクスチャ分離部におけるストラクチャ・テクスチャ分離に用いるフィルタ係数を算出する第1フィルタ係数算出手段をさらに備える、付記15乃至18のいずれか一に記載の画像信号処理装置。
(付記20)
前記第2低周波成分から前記第1ストラクチャ・テクスチャ分離部におけるストラクチャ・テクスチャ分離に用いるフィルタ係数を算出する第2フィルタ係数算出手段をさらに備える、付記15乃至19のいずれか一に記載の画像信号処理装置。
(付記21)
前記第1フィルタ係数算出手段により算出したフィルタ係数と、前記第2フィルタ係数算出手段により算出したフィルタ係数とを合成するフィルタ係数合成手段をさらに備え、当該フィルタ係数合成手段により合成されたフィルタ係数を用いて前記第1ストラクチャ・テクスチャ分離部におけるストラクチャ・テクスチャ分離を行う、付記20記載の画像信号処理装置。
(付記22)
前記第1ストラクチャ・テクスチャ分離部における、前記第1ストラクチャ成分と、前記第1テクスチャ成分とに分離する際のフィルタ係数を補正をするための画像中の各点における画素値の勾配強度を、前記第1逆ウェーブレット変換部における逆ウェーブレット変換後の画像信号から求める第1勾配強度算出手段をさらに備える、付記15乃至21のいずれか一に記載の画像信号処理装置。
(付記23)
前記第1ストラクチャ・テクスチャ分離部における、前記第1ストラクチャ成分と、前記第1テクスチャ成分とに分離する際のフィルタ係数を補正するための画像中の各点における画素値の勾配強度を、前記第2低周波成分、または前記第2低周波成分を処理して得られた処理後の第2低周波成分から求める第2勾配強度算出手段をさらに備える、付記22に記載の画像信号処理装置。
(付記24)
前記補正をするための画像中の各点における画素値の勾配強度を、前記第1勾配強度数算出手段により算出した前記画像中の各点における画素値の勾配強度と、前記第2勾配強度数算出手段により算出した前記画像中の各点における画素値の勾配強度とを合成することにより求める勾配強度合成手段をさらに備える、付記23記載の画像信号処理装置。
(付記25)
前記フィルタ係数として、非線形ローパスフィルタのフィルタ係数を用いる付記2乃至8、10乃至14及び18乃至24のうちのいずれか1に記載の画像信号処理装置。
(付記26)
前記ウェーブレット係数縮退部における高周波成分の補正が、その高周波成分を、符号を変えずに絶対値以下に補正する付記8記載の画像信号処理装置。
(付記27)
前記第二ウェーブレット係数縮退部における第二高周波成分の補正が、その高周波成分を、符号を変えずに絶対値以下に補正する付記14記載の画像信号処理装置。
(付記28)
前記ウェーブレット係数縮退部における高周波成分の補正が、その高周波成分を、符号を変えずに絶対値以下に補正する付記17記載の画像信号処理装置。
(付記29)
前記第一フィルタ係数算出手段により求めたフィルタ係数と、前記第二フィルタ係数算出手段により求めたフィルタ係数との合成比率を、局所的な画素値の分散に基づいて決定する付記7記載の画像信号処理装置。
(付記30)
前記フィルタ係数合成手段により合成したフィルタ係数と、前記第三フィルタ係数算出手段により求めたフィルタ係数との合成比率を、局所的な画素値の分散に基づいて決定する付記7記載の画像信号処理装置。
(付記31)
前記第三フィルタ係数算出手段により求めたフィルタ係数と、前記第四フィルタ係数算出手段により求めたフィルタ係数を合成することにより求めたフィルタ係数を用い、その合成比率を、局所的な画素値の分散に基づいて決定する付記24記載の画像信号処理装置。
(付記32)
前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際に用いられるフィルタ係数を算出するフィルタ係数算出手段をさらに備え、前記フィルタ係数を、注目画素と、注目画素に隣接する画素間の変動量に基づく指標と、
注目画素を中心とする数画素範囲の画素値の勾配強度と勾配方向に基づく指標の合成により算出する、付記1記載の画像信号処理装置。
(付記33)
前記第2ストラクチャ成分と、前記第2テクスチャ成分とに分離する際に用いられるフィルタ係数を算出するフィルタ係数算出手段をさらに備え、前記フィルタ係数を、注目画素と、注目画素に隣接する画素間の変動量に基づく指標と、
注目画素を中心とする数画素範囲の画素値の勾配強度と勾配方向に基づく指標の合成により算出する、付記1記載の画像信号処理装置。
(付記34)
前記第1ストラクチャ成分と、前記第1テクスチャ成分とに分離する際に用いられるフィルタ係数を算出するフィルタ係数算出手段をさらに備え、前記フィルタ係数を、注目画素と、注目画素に隣接する画素間の変動量に基づく指標と、
注目画素を中心とする数画素範囲の画素値の勾配強度と勾配方向に基づく指標の合成により算出する、付記15記載の画像信号処理装置。
(付記35)
前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際に用いられるフィルタ係数を算出するフィルタ係数算出手段をさらに備え、前記フィルタ係数を、前記低周波成分により算出された注目画素を中心とする数画素範囲の画素値の勾配強度と勾配方向に基づく指標を用いて算出する、付記1記載の画像信号処理装置。
(付記36)
前記第一フィルタ係数算出手段により算出したフィルタ係数と、前記第二フィルタ係数算出手段により算出したフィルタ係数とを合成する際の合成比率を、局所的な画素値の分散により決定する、付記4記載の画像信号処理装置。
(付記37)
前記ストラクチャ・テクスチャ分離部が非線形ローパスフィルタをさらに備える付記1記載の画像信号処理装置。
(付記38)
原画像信号をウェーブレット変換を用いて多重解像度分解する手段と、
最低解像度から順次、エッジ保存型の非線形ローパスフィルタを用いて、当該解像度の低周波成分を、画像中のエッジや平坦成分から構成されるストラクチャ成分と、ノイズや細かい模様からなるテクスチャ成分と、に分離する手段と、
前記分離されたテクスチャ成分の各値を符号を変えずにその絶対値以下に補正する手段と、
前記ストラクチャ成分と、前記補正されたテクスチャ成分と、を合成して、補正された前記当該解像度の低周波成分を生成する手段と、
当該解像度が原画像解像度よりも低い場合には、前記補正された当該解像度における低周波成分と、当該解像度における高周波成分と、から逆ウェーブレット変換によって当該解像度より一つ上の解像度の低周波成分を生成することを、
当該解像度が原画像と同じ解像度になるまで繰り返す手段と
を備える画像信号処理装置であって、
前記エッジ保存型の非線形ローパスフィルタのフィルタ係数算出において、当該解像度が最低解像度でないときは、当該解像度より低い解像度で算出された非線形ローパスフィルタのフィルタ係数を、当該解像度の非線形ローパスフィルタのフィルタ係数に合成することを特徴とする
画像信号処理装置。
(付記39)
前記ストラクチャ・テクスチャ分離部が非線形ローパスフィルタを備え、前記非線形ローパスフィルタのフィルタ係数は、注目画素と、注目画素に隣接する画素間の変動量に基づく指標と、
注目画素を中心とする数画素範囲の画素値の勾配強度と勾配方向に基づく指標の合成によって算出されることを特徴とする付記1記載の画像信号処理装置。
(付記40)
原画像の画像信号を入力するステップと、
前記画像信号をウェーブレット変換することにより低周波成分と高周波成分とを生成するステップと、
前記低周波成分を、第一ストラクチャ成分と、第一テクスチャ成分とに分離するステップと、
前記第一テクスチャ成分からノイズを除去し、ノイズを除去後の第一テクスチャ成分を生成するステップと、
前記第一ストラクチャ成分と前記ノイズを除去後の第一テクスチャ成分とを合成し、合成後の低周波成分を生成するステップと、
前記高周波成分と前記合成後の低周波成分とを逆ウェーブレット変換することにより、逆ウェーブレット変換後の画像信号を生成するステップと、
前記逆ウェーブレット変換後の画像信号を、第二ストラクチャ成分と、第二テクスチャ成分とに分離するステップと、
前記第二テクスチャ成分からノイズを除去し、ノイズを除去後の第二テクスチャ成分を生成するステップと、
前記第二ストラクチャ成分と、前記補正後の第二テクスチャ成分とを合成し、画像信号処理信号を生成するステップと、
前記補正後の画像信号を出力するステップと、
を備える画像信号処理方法。
(付記41)
原画像の画像信号を入力する処理と、
前記画像信号をウェーブレット変換することにより低周波成分と高周波成分とを生成する処理と、
前記低周波成分を、第一ストラクチャ成分と、第一テクスチャ成分とに分離する処理と、
前記第一テクスチャ成分からノイズを除去し、当該ノイズを除去後の第一テクスチャ成分を生成する処理と、
前記第一ストラクチャ成分と前記補正後の第一テクスチャ成分とを合成し、合成後の低周波成分を生成する処理と、
前記高周波成分と前記合成後の低周波成分とを逆ウェーブレット変換することにより、逆ウェーブレット変換後の画像信号を生成する処理と、
前記逆ウェーブレット変換後の画像信号を、第二ストラクチャ成分と、第二テクスチャ成分とに分離する処理と、
前記第二テクスチャ成分からノイズを除去し、当該ノイズを除去後後の第二テクスチャ成分を生成する処理と、
前記第二ストラクチャ成分と、前記補正後の第二テクスチャ成分とを合成し、画像信号処理信号を生成する処理と、
前記補正後の画像信号を出力する処理と、
をコンピュータに実行させる画像信号処理プログラム。
002、012、022、032、042、052、062 画像信号処理部
003、013、023、033、043、053、063 画像信号出力部
101、111、121、131、141 WT部
151、161 第一WT部
152、162 第二WT部
153、163 第三WT部
201、211、221、231、241 IWT部
251、261 第一IWT部
252、262 第二IWT部
253、263 第三IWT部
301、311、321、331、341、351、361 第一STD部
302、312、322、332、342、352、362 第二STD部
343、353、363 第三STD部
344、354、364 第四STD部
401、411、421、431、441、451、461 第一TC縮退部
402、412、422、432、442、452、462 第二TC縮退部
443、453、463 第三TC縮退部
444、454、464 第四TC縮退部
501、511、521、531、541、551、561 第一合成部
502、512、522、532、542、552、562 第二合成部
543、553、563 第三合成部
544、554、564 第四合成部
641 WC縮退部
651、661 第一WC縮退部
652、662 第二WC縮退部
653、663 第二WC縮退部
710 フィルタ係数算出部
721、741、751、761 第一フィルタ係数算出部
722、742、752、762 第二フィルタ係数算出部
753、763 第三フィルタ係数算出部
754 第四フィルタ係数算出部
820、830、840 フィルタ係数合成部
851 第一フィルタ係数合成部
852 第二フィルタ係数合成部
853 第三フィルタ係数合成部
930 勾配強度算出部
961 第一勾配強度算出部
962 第二勾配強度算出部
963 第三勾配強度算出部
1030 勾配強度合成部
1061 第一勾配強度合成部
1062 第二勾配強度合成部
1063 第三勾配強度合成部
1130 フィルタ補正部
1161 第一フィルタ補正部
1162 第二フィルタ補正部
1163 第三フィルタ補正部
Claims (10)
- 原画像の画像信号を入力する画像信号入力手段と、
前記画像信号をウェーブレット変換することにより低周波成分と高周波成分とを生成するウェーブレット変換手段と、
前記低周波成分を、第一ストラクチャ成分と、第一テクスチャ成分とに分離する第一ストラクチャ・テクスチャ分離手段と、
前記第一テクスチャ成分についてノイズを除去し、当該処理後の第一テクスチャ成分を生成するテクスチャ成分縮退手段と、
前記第一ストラクチャ成分と前記補正後の第一テクスチャ成分とを合成し、合成後の低周波成分を生成する第一合成手段と、
前記高周波成分と前記合成後の低周波成分とを逆ウェーブレット変換することにより、逆ウェーブレット変換後の画像信号を生成する逆ウェーブレット変換手段と、
前記逆ウェーブレット変換後の画像信号を、第二ストラクチャ成分と、第二テクスチャ成分とに分離する第二ストラクチャ・テクスチャ分離手段と、
前記第二テクスチャ成分についてノイズを除去し、当該処理後の第二テクスチャ成分を生成する第二テクスチャ成分縮退手段と、
前記第二ストラクチャ成分と、前記補正後の第二テクスチャ成分とを合成し、画像信号処理信号を生成する第二合成手段と、
前記補正後の画像信号を出力する画像信号出力手段と、
を備える画像信号処理装置。 - 前記第二ストラクチャ・テクスチャ分離手段における、前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際のフィルタ係数を、前記逆ウェーブレット変換後の画像信号から算出する第一フィルタ係数算出手段をさらに備える、請求項1記載の画像信号処理装置。
- 前記第二ストラクチャ・テクスチャ分離手段における、前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際のフィルタ係数を、前記低周波成分から算出する第二フィルタ係数算出手段をさらに備える、請求項2記載の画像信号処理装置。
- 前記第二ストラクチャ・テクスチャ分離手段における、前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際の前記フィルタ係数を、前記第一フィルタ係数算出手段により算出したフィルタ係数と、前記第二フィルタ係数算出手段により算出したフィルタ係数とを合成することにより求めるフィルタ係数合成手段をさらに備える、請求項3記載の画像信号処理装置。
- 前記第二ストラクチャ・テクスチャ分離手段における、前記第二ストラクチャ成分と、前記第二テクスチャ成分とに分離する際の前記フィルタ係数の補正をするための画像中の各点における画素値の勾配強度を、逆ウェーブレット変換後の画像信号から求める第一勾配強度算出手段をさらに備える、請求項2記載の画像信号処理装置。
- 前記補正をするための画像中の各点における画素値の勾配強度を、前記低周波成分から求める第二勾配強度算出手段をさらに備える、請求項5記載の画像信号処理装置。
- 前記補正をするための画像中の各点における画素値の勾配強度を、前記第一勾配強度数算出手段により算出した前記画像中の各点における画素値の勾配強度と、前記第二勾配強度数算出手段により算出した前記画像中の各点における画素値の勾配強度とを合成することにより求める勾配強度合成手段をさらに備える、請求項6記載の画像信号処理装置。
- 前記高周波成分のノイズを除去するウェーブレット係数縮退手段をさらに備える請求項1乃至7のうちのいずれか1に記載された画像信号処理装置。
- 原画像の画像信号を入力し、
前記画像信号をウェーブレット変換することにより低周波成分と高周波成分とを生成し、
前記低周波成分を、第一ストラクチャ成分と、第一テクスチャ成分とに分離し、
前記第一テクスチャ成分からノイズを除去し、ノイズを除去後の第一テクスチャ成分を生成し、
前記第一ストラクチャ成分と前記ノイズを除去後の第一テクスチャ成分とを合成し、合成後の低周波成分を生成し、
前記高周波成分と前記合成後の低周波成分とを逆ウェーブレット変換することにより、逆ウェーブレット変換後の画像信号を生成し、
前記逆ウェーブレット変換後の画像信号を、第二ストラクチャ成分と、第二テクスチャ成分とに分離し、
前記第二テクスチャ成分からノイズを除去し、ノイズを除去後の第二テクスチャ成分を生成し、
前記第二ストラクチャ成分と、前記補正後の第二テクスチャ成分とを合成し、画像信号処理信号を生成し、
前記補正後の画像信号を出力する、
画像信号処理方法。 - 原画像の画像信号を入力する処理と、
前記画像信号をウェーブレット変換することにより低周波成分と高周波成分とを生成する処理と、
前記低周波成分を、第一ストラクチャ成分と、第一テクスチャ成分とに分離する処理と、
前記第一テクスチャ成分からノイズを除去し、当該ノイズを除去後の第一テクスチャ成分を生成する処理と、
前記第一ストラクチャ成分と前記補正後の第一テクスチャ成分とを合成し、合成後の低周波成分を生成する処理と、
前記高周波成分と前記合成後の低周波成分とを逆ウェーブレット変換することにより、逆ウェーブレット変換後の画像信号を生成する処理と、
前記逆ウェーブレット変換後の画像信号を、第二ストラクチャ成分と、第二テクスチャ成分とに分離する処理と、
前記第二テクスチャ成分からノイズを除去し、当該ノイズを除去後後の第二テクスチャ成分を生成する処理と、
前記第二ストラクチャ成分と、前記補正後の第二テクスチャ成分とを合成し、画像信号処理信号を生成する処理と、
前記補正後の画像信号を出力する処理と、
をコンピュータに実行させる画像信号処理プログラムを記録した記録媒体。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US15/521,345 US10229479B2 (en) | 2014-11-13 | 2015-11-06 | Image signal processing apparatus, image signal processing method and image signal processing program |
JP2016558877A JPWO2016075914A1 (ja) | 2014-11-13 | 2015-11-06 | 画像信号処理装置、画像信号処理方法及び画像信号処理プログラム |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2014-230381 | 2014-11-13 | ||
JP2014230381 | 2014-11-13 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2016075914A1 true WO2016075914A1 (ja) | 2016-05-19 |
Family
ID=55954017
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2015/005569 WO2016075914A1 (ja) | 2014-11-13 | 2015-11-06 | 画像信号処理装置、画像信号処理方法及び画像信号処理プログラム |
Country Status (3)
Country | Link |
---|---|
US (1) | US10229479B2 (ja) |
JP (1) | JPWO2016075914A1 (ja) |
WO (1) | WO2016075914A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20220131808A (ko) * | 2021-03-22 | 2022-09-29 | 재단법인대구경북과학기술원 | 이미지 분류 모델 생성 방법 및 장치 |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3459043A4 (en) * | 2016-06-21 | 2019-04-24 | Zhejiang Dahua Technology Co., Ltd | SYSTEMS AND METHOD FOR IMAGE PROCESSING |
US11113796B2 (en) * | 2018-02-09 | 2021-09-07 | Delta Electronics, Inc. | Image enhancement circuit and method thereof |
CN109816764B (zh) * | 2019-02-02 | 2021-06-25 | 深圳市商汤科技有限公司 | 图像生成方法及装置、电子设备和存储介质 |
US12086961B2 (en) * | 2020-12-11 | 2024-09-10 | Samsung Electronics Co., Ltd | Method and apparatus for adaptive hybrid fusion |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014103230A1 (ja) * | 2012-12-26 | 2014-07-03 | 日本電気株式会社 | 信号処理装置、信号処理方法、および記憶媒体 |
WO2014126153A1 (ja) * | 2013-02-18 | 2014-08-21 | 日本電気株式会社 | 画像処理方法、画像処理装置、及び画像処理プログラム |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007148945A (ja) | 2005-11-30 | 2007-06-14 | Tokyo Institute Of Technology | 画像修復方法 |
EP2003612A4 (en) * | 2006-03-31 | 2010-10-13 | Nikon Corp | IMAGE PROCESSING |
US8515171B2 (en) * | 2009-01-09 | 2013-08-20 | Rochester Institute Of Technology | Methods for adaptive and progressive gradient-based multi-resolution color image segmentation and systems thereof |
RU2012141557A (ru) | 2010-03-01 | 2014-04-10 | Шарп Кабусики Кайся | Устройство для увеличения изображения, программа для увеличения изображения, запоминающая среда, в которой хранится программа для увеличения изображения, и устройство отображения |
JP4892096B1 (ja) | 2010-11-30 | 2012-03-07 | 株式会社東芝 | 画像処理装置および画像処理方法 |
US9111339B1 (en) * | 2013-07-31 | 2015-08-18 | Marvell International Ltd. | System and method for reducing noise from an image |
-
2015
- 2015-11-06 WO PCT/JP2015/005569 patent/WO2016075914A1/ja active Application Filing
- 2015-11-06 US US15/521,345 patent/US10229479B2/en active Active
- 2015-11-06 JP JP2016558877A patent/JPWO2016075914A1/ja active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014103230A1 (ja) * | 2012-12-26 | 2014-07-03 | 日本電気株式会社 | 信号処理装置、信号処理方法、および記憶媒体 |
WO2014126153A1 (ja) * | 2013-02-18 | 2014-08-21 | 日本電気株式会社 | 画像処理方法、画像処理装置、及び画像処理プログラム |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20220131808A (ko) * | 2021-03-22 | 2022-09-29 | 재단법인대구경북과학기술원 | 이미지 분류 모델 생성 방법 및 장치 |
KR102611121B1 (ko) | 2021-03-22 | 2023-12-07 | 재단법인대구경북과학기술원 | 이미지 분류 모델 생성 방법 및 장치 |
Also Published As
Publication number | Publication date |
---|---|
JPWO2016075914A1 (ja) | 2017-08-24 |
US20170308995A1 (en) | 2017-10-26 |
US10229479B2 (en) | 2019-03-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP7427748B2 (ja) | 画像処理システム及び医用情報処理システム | |
WO2016075914A1 (ja) | 画像信号処理装置、画像信号処理方法及び画像信号処理プログラム | |
CN105469363B (zh) | 基于模糊图像生成清晰图像的方法和装置 | |
JP2011180798A (ja) | 画像処理装置、および画像処理方法、並びにプログラム | |
JP2001057677A (ja) | 画像処理方法および装置並びに記録媒体 | |
JP6222481B2 (ja) | 画像処理方法、画像処理システム、画像処理装置および画像処理プログラム | |
JP2017010092A (ja) | 画像処理装置、撮像装置、画像処理方法、画像処理プログラム、および、記憶媒体 | |
JP2007018379A (ja) | 画像処理方法及び画像処理装置 | |
JP5105286B2 (ja) | 画像復元装置、画像復元方法及び画像復元プログラム | |
Tsutsui et al. | Halo artifacts reduction method for variational based realtime retinex image enhancement | |
Ponomaryov | Real-time 2D–3D filtering using order statistics based algorithms | |
Chen et al. | A novel multi-image super-resolution reconstruction method using anisotropic fractional order adaptive norm | |
Jindal et al. | A novel image zooming technique using wavelet coefficients | |
KR101707337B1 (ko) | 영상 디노이징을 위한 다중해상도 nlm 필터링 방법 | |
CN107590781B (zh) | 基于原始对偶算法的自适应加权tgv图像去模糊方法 | |
Katkovnik et al. | Frequency domain blind deconvolution in multiframe imaging using anisotropic spatially-adaptive denoising | |
KR20100097858A (ko) | 예제 기반 신경회로망을 이용한 고해상도 영상 확대 | |
JP2017010094A (ja) | 画像処理装置、撮像装置、画像処理方法、画像処理プログラム、および、記憶媒体 | |
JP5903305B2 (ja) | 画像処理装置及び画像処理方法 | |
JP4456096B2 (ja) | ノイズ除去方法および装置並びにプログラム | |
KR101675117B1 (ko) | 다계층의 저해상도 영상을 이용한 고해상도 영상 생성 방법 및 그 장치 | |
Khetkeeree et al. | Hybrid filtering for image sharpening and smoothing simultaneously | |
Hiraoka et al. | Reduction of iterative calculation and quality improvement for generation of moire-like images using bilateral filter | |
JPWO2015115168A1 (ja) | 画像処理装置 | |
JP2017130167A (ja) | 画像処理装置、撮像装置および画像処理プログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 15859116 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2016558877 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15521345 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 15859116 Country of ref document: EP Kind code of ref document: A1 |