WO2013161940A1 - 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 - Google Patents
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 Download PDFInfo
- Publication number
- WO2013161940A1 WO2013161940A1 PCT/JP2013/062202 JP2013062202W WO2013161940A1 WO 2013161940 A1 WO2013161940 A1 WO 2013161940A1 JP 2013062202 W JP2013062202 W JP 2013062202W WO 2013161940 A1 WO2013161940 A1 WO 2013161940A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- component
- image
- noise
- initial
- image processing
- Prior art date
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 65
- 238000004364 calculation method Methods 0.000 claims abstract description 37
- 238000000034 method Methods 0.000 claims description 161
- 238000000926 separation method Methods 0.000 claims description 60
- 230000003252 repetitive effect Effects 0.000 claims description 10
- 230000008569 process Effects 0.000 description 31
- 230000006870 function Effects 0.000 description 28
- 230000001629 suppression Effects 0.000 description 18
- 238000010586 diagram Methods 0.000 description 13
- 230000002238 attenuated effect Effects 0.000 description 8
- 230000009466 transformation Effects 0.000 description 8
- 238000007796 conventional method Methods 0.000 description 7
- 238000004321 preservation Methods 0.000 description 4
- 230000002194 synthesizing effect Effects 0.000 description 4
- 230000015572 biosynthetic process Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000003786 synthesis reaction Methods 0.000 description 3
- 238000004590 computer program Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 230000001131 transforming effect Effects 0.000 description 2
- 241000207961 Sesamum Species 0.000 description 1
- 235000003434 Sesamum indicum Nutrition 0.000 description 1
- 238000000610 breath-figure templating Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
- 230000000007 visual effect Effects 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- 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
- G06T2207/20172—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 processing method, an image processing system, an image processing apparatus, and an image processing program.
- a technology for reducing noise contained in an image is an indispensable technology for reproducing a captured image more clearly.
- a technique for reducing noise for example, there is a noise removal method based on total variation (Total-Variation, TV) norm regularization disclosed in Non-Patent Document 1.
- TV method total variation
- FIG. 19 shows an image processing method using the conventional TV method.
- FIG. 20 is a flowchart for explaining an image processing method using the conventional TV method.
- the input image F is supplied to the skeleton component / residual component separation unit 101.
- the skeleton component / residual component separation unit 101 separates the supplied image F into a skeleton component U TV composed of a strong edge and a flat region and a residual component V TV composed of a texture and noise. That is, the input image F is represented by the sum of the skeleton component U TV and the residual component V TV as in the following formula (1).
- Framework component U TV is obtained by minimizing the total variational norm J (U TV) of U TV represented by the following formula (2).
- x in the formula is the horizontal pixel position of the skeleton component U TV
- y denotes the vertical pixel position of the skeleton component U TV.
- This minimization problem can be solved by iterating the Chambolle Projection method.
- a subgradient method using a TVV norm subgradient can be used as an alternative to the Chambolle Projection method.
- Equation (3) the skeletal component at the time of the (k + 1) th iteration using the subgradient method Is calculated (step S2).
- the stop condition for repetition is described.
- Input image height M, width N, residual component at the kth iteration M-by-n element Then, using the standard deviation ⁇ noise of the noise of the target image estimated in advance, It is conceivable that the iteration is stopped when it becomes (step S3).
- Equation (4) is the residual component Based on the assumption that is a noise.
- using the threshold ⁇ It is conceivable that the iteration is stopped when it becomes (step S3).
- Equation (5) is the result of the kth iteration And k + 1 iterations Are considered to have converged when it is determined that the variation is sufficiently small.
- the iteration is also stopped when the iteration number k reaches the maximum iteration number k max (step S4).
- the number of iterations is less than k,
- FIG. 21 shows an image processing method using the TV method of FIG. 19 that has a process of attenuating the noise component of the residual component separated by the skeleton component / residual component separation unit 101 and combining it with the skeleton component. This is an extended image processing method.
- FIG. 22 is a flowchart for explaining the image processing method shown in FIG.
- the residual component V TV separated by the skeleton component / residual component separation unit 101 is supplied to the noise suppression unit 102, and a function f for attenuating the noise component in the residual component V TV is expressed by the following equation (7).
- Act step S7.
- the skeletal component U TV and the residual component (ie, texture component) f (V TV ) in which the noise component is attenuated are synthesized and output as an image Z with reduced noise (step S6).
- Steps S1 to S5 are the same as the image processing method using the TV method in FIG. Examples of the function f include a soft decision threshold process shown in FIG. 23A, a hard decision threshold process shown in FIG. 23B, and a factor process shown in FIG. It is not something.
- the soft decision threshold process shown in FIG. 23A is expressed by the following equation (8).
- the sign function in the expression is a function that outputs a positive / negative sign.
- the hard decision threshold processing shown in FIG. 23B is expressed by the following equation (9). If ⁇ is an appropriate coefficient, the factor process shown in FIG. 23C is expressed by the following equation (10). Further, it is possible to combine soft decision threshold processing or hard decision threshold processing and facotor processing as shown in Expression (11). In Expression (11), the function is configured using the soft decision threshold processing, but it is naturally possible to configure the same function using the hard decision threshold processing.
- Non-Patent Document 1 is an iterative process in which the above-mentioned Chambolle Projection method is repeatedly applied until the solution converges, there is a problem that the calculation cost increases.
- the TV method can remove high-frequency noise, it has a problem that it cannot be applied to low-frequency noise that occurs due to uneven distribution of noise particles and spreads over a relatively wide range compared to high-frequency noise.
- Low frequency noise is due to uneven distribution density of noise particles in a wide range of several pixels to several tens of pixels. Since the TV norm for such low frequency noise is small, it is regarded as a flat region (no noise). It is. In other words, the TV method cannot handle low frequency noise in principle.
- the present invention was invented in view of the above problems, and an object thereof is to provide an image processing method and an image processing apparatus capable of effectively and rapidly removing noises in various frequency bands.
- the present invention generates an initial noise-removed image by reducing noise while preserving edges in the input image, and is defined in advance based on an initial residual component calculated from the input image and the initial noise-removed image.
- This is an image processing method for controlling a repetitive calculation based on the energy, separating the initial noise-removed image into a skeleton component and a residual component by the controlled repetitive calculation, and generating the skeleton component as an output image.
- the present invention provides an initial noise-removed image generating unit that generates an initial noise-removed image by a noise-removing method that preserves an edge component in an input image, and an initial noise-removed image by a repetitive calculation based on a predetermined energy.
- Image processing comprising: a skeleton component / residual component separation unit that separates a component into a residual component and generates the skeleton component as an output image; and a control unit that controls the iterative calculation based on the initial residual component Device.
- noises in various frequency bands can be removed effectively and at high speed.
- FIG. 4 is a flowchart illustrating an example of a procedure of the image processing method according to the first embodiment.
- FIG. 3A shows an original image including noise.
- FIG. 3B is an image obtained as a result of removing noise from the wavelet shrinkage method of FIG.
- FIG. 3A shows an original image including noise.
- FIG. 3B is an image obtained as a result of removing noise from the wavelet shrinkage method of FIG.
- led-out from 1st Embodiment It is explanatory drawing of the image processing method derived
- 1 is a block diagram illustrating a configuration example of an image processing apparatus according to a first embodiment.
- FIG. 16A shows an original image.
- FIG. 16B shows the result of applying wavelet transform to the original image of FIG.
- FIG. 16C shows the result of recursively applying the wavelet transform three times to the original image of FIG. 10
- FIG. 10 is a flowchart illustrating an example of a procedure of an image processing method according to a fourth embodiment. It is a block diagram which shows the structural example of the image processing apparatus of 4th Embodiment. It is explanatory drawing of the existing image processing method. It is a flowchart explaining the example of a procedure of the existing image processing method.
- FIG. 23A shows input / output responses of the soft decision threshold processing.
- FIG. 23B shows the input / output response of the hard decision threshold process.
- FIG. 23C shows the input / output response of the factor process.
- FIG. 1 is a functional block diagram for explaining an image processing method according to the first embodiment of the present invention.
- the image processing method includes an initial noise-removed image generation unit 103 that generates an initial noise-removed image by a lightweight noise-removing method that preserves edge components in an input image, and initial noise removal.
- the iterative control unit 104 that controls the subsequent skeletal component / residual component separation unit based on the initial residual component separated from the input image in the image generation unit 103, and the skeletal component and residual component shown in FIG.
- a skeleton component / residual component separation unit 101 for separation is provided.
- the skeleton component / residual component separation method for separating the skeleton component and the residual component will be described by the TV method shown in the conventional method, but is not limited thereto.
- FIG. 2 is a flowchart for explaining the image processing method according to the first embodiment.
- the input image F is supplied to the initial noise removal image generation unit 103.
- the initial noise removal image generation unit 103 applies a lightweight noise removal method for storing edge components to the input image F, and generates an initial noise removal image U init (step S8).
- the initial residual component composed of noise and texture is V init .
- the following equation (12) is established.
- the generated initial noise-removed image U init is supplied to the skeleton component / residual component separation unit 101. Further, the separated initial residual component V init is supplied to the iterative control unit 104.
- a lightweight noise removal method for storing edge components in step S8 will be described.
- a light noise removal method a Wavelet Shrinkage (WS) method, a Bitaltal Filter (BF), or the like can be applied. These methods are image noise removal methods that have high edge preservability and low calculation costs compared to the TV method.
- the WS method obtains an image with reduced noise by two-dimensional wavelet transforming an image, attenuating high-frequency components of wavelet transform coefficients in the 0 direction, and then inversely transforming the two-dimensional wavelet to reconstruct the image.
- the wavelet transform coefficient is w
- the wavelet transform coefficient after shrinkage processing is w ′
- the attenuation is ⁇
- the attenuation processing is expressed by the following formulas (13) and (14), for example.
- the sign function in Formula (13) is a function which outputs a positive / negative sign.
- a wavelet transform coefficient having a small absolute value is almost a noise component. Therefore, by setting an appropriate attenuation amount ⁇ to attenuate the wavelet transform coefficient, the noise component included in the image can be reduced. In addition, since a wavelet transform coefficient having a large absolute value, which represents an edge component, remains, the edge in the input image is saved as a result.
- BF is a derivative type of the weighted smoothing filter, and implements a noise removal method that preserves edges by the following two indicators that determine filter coefficients.
- the first index is the spatial distance between the noise removal target pixel and its surrounding pixels.
- a pixel spatially close to the target pixel is regarded as having a high correlation with the target pixel, and the weight of the filter coefficient is increased.
- the spatially separated pixels are regarded as having a low correlation with the target pixel, and the weight of the filter coefficient is reduced.
- the second index is the difference between the pixel values of the noise removal target pixel and its surrounding pixels.
- the second index works like a general weighted smoothing filter in an area where the pixel values are almost uniform, that is, in a flat area, so that noise can be removed, and an area where the difference in pixel values is large, In other words, in the region including the edge, the weight of the filter coefficient other than the target pixel becomes small, so the edge is stored.
- the calculation cost of WS method and BF is small compared with TV method.
- visual artifacts may occur in the processed image.
- An example is shown in FIG. 3A is an image including noise
- FIG. 3B is an image from which noise is removed by the WS method.
- FIG. 3B unnatural degradation that does not exist in the original image is observed.
- BF cannot remove isolated noise called sesame salt noise in principle.
- the WS method and BF in this embodiment are used only to generate an initial noise-removed image, and artifacts generated by the WS method and isolated noise that cannot be removed by the BF method are removed by the subsequent processing of the present invention. Therefore, it does not become a problem.
- the iteration control unit 104 controls the number of iterations of the TV method based on the initial residual component V init supplied from the initial noise removal image generation unit 103.
- the standard deviation ⁇ init is calculated from the initial residual component V init supplied from the initial noise-removed image generation unit 103 as follows (step S9).
- step S9 the m by n elements of the initial residual component V init are Then, the standard deviation ⁇ init is calculated as expressed by the following formula (15).
- the parameter ⁇ shown in Equation (3) and the maximum number of iterations k max of a fixed value are used, but in this embodiment, the noise standard deviation ⁇ noise estimated in advance and the step S9
- Each parameter is corrected from the calculated standard deviation ⁇ init (step S10).
- Equation (16) the parameter ⁇ ′ represented by the following equation (16) calculated from the standard deviation ⁇ noise of noise estimated in advance and the standard deviation ⁇ init calculated in step S9 is used. That is, in Equation (16), when ⁇ init is close to ⁇ noise , it means that it is in the vicinity of the convergence solution of the TV method, and the convergence speed can be improved. Specifically, the convergence speed is improved by narrowing the search area in one iteration.
- the maximum number of iterations k 'max, the conventional maximum number of iterations k max, and a minimum number of iterations k min be predetermined, and the standard deviation sigma noise of noise estimated beforehand, the initial residual component V init standard deviation sigma init And is defined as in the following expression (17) (step S10).
- the max function is a function that outputs the larger one of the two input values
- the min function is a function that outputs the smaller one of the two input values
- the round function is a rounding function to an integer value.
- the skeleton component / residual component separation unit 101 applies the TV method using the initial noise-removed image U init supplied from the initial noise-removed image generation unit 103 as an initial solution, and satisfies the iteration stop condition determined by the iteration control unit 104.
- a skeleton component U TV composed of a strong edge and a flat region and a residual component V TV composed of texture and noise are separated.
- the following formula (19) is used as the repeated stop condition instead of the formula (4) of the conventional method.
- the skeleton component U TV is regarded as an image from which noise is removed, and a noise-removed image Z is obtained as shown in the following equation (20) (step S6).
- a configuration in which a residual component V TV in which a noise component is attenuated and a skeleton component U TV are combined and a noise-removed image Z is output is also possible. is there.
- the image processing method of the invention has a configuration as shown in FIG. 4 and is represented by a flowchart as shown in FIG.
- FIG. 6 is a diagram illustrating a configuration example of the image processing apparatus according to the first embodiment.
- the image processing apparatus 1000 is an apparatus that removes noise from the input image 1 and outputs an output image 2.
- the image processing apparatus 1000 includes an initial noise removal image generation unit 3, an initial residual component storage memory 4, an initial residual component standard deviation calculation unit 5, an initial residual component standard deviation storage memory 6, and a subgradient A parameter calculation unit 7 for the method, a parameter storage memory 8 for the subgradient method, an iterative control unit 9 for the subgradient method, and a skeleton component / residual component separation unit 10 are provided.
- the initial noise-removed image generation unit 3 applies a lightweight noise removal method for storing the edge component described in step S8, and generates an initial noise-removed image.
- the initial residual component storage memory 4 stores the initial residual component separated from the input image 1 by the initial noise removal image generation unit 3.
- the initial residual component standard deviation calculation unit 5 refers to the initial residual component storage memory 4 and calculates the standard deviation of the initial residual component according to the procedure described in step S9.
- the initial residual component standard deviation storage memory 6 stores the standard deviation of the initial residual component calculated by the standard residual calculation unit 5 of the initial residual component.
- the parameter calculation unit 7 of the subgradient method refers to the standard deviation storage memory 6 of the initial residual component, and performs the sub-gradient method parameter ⁇ ′ and the maximum iteration number parameter of the subgradient method in the procedure described in step S10. k ' max is calculated.
- the subgradient method parameter storage memory 8 stores the subgradient method parameters calculated by the subgradient method parameter calculation unit 7.
- the iterative control unit 9 of the subgradient method refers to the initial residual component storage memory 4 and the parameter storage memory 8 of the subgradient method, and performs the skeleton component / residual component separation unit according to the procedure described in steps S3 and S4. Control the iteration of the subgradient method at 10.
- the skeleton component / residual component separation unit 10 is supplied from the initial noise removal image generation unit 3 in accordance with the procedure described in steps S1, S2, S3, S4, and S5 under the control of the iterative control unit 9 of the subgradient method.
- the subgradient method is executed using the initial noise-removed image as an initial solution, the initial noise-removed image is separated into a skeleton component and a residual component, and the skeleton component is set as an output image 2.
- FIG. 6 Note that the configuration of the image processing apparatus illustrated in FIG. 6 is merely an example, and other configurations may be employed as long as the apparatus implements similar functions.
- the image processing apparatus 1000 can be realized by a computer, and each component constituting the image processing apparatus, that is, the initial noise removal image generation unit 3, the initial residual component storage memory 4, and the initial residual component Standard deviation calculation unit 5, standard deviation storage memory 6 of initial residual component, subgradient method parameter calculation unit 7, subgradient method parameter storage memory 8, subgradient method iteration control unit 9, skeleton
- the component / residual component separation unit 10 can be realized as a program for causing a central processing unit (CPU) of a computer to realize the above-described functions.
- each component constituting the image processing apparatus can be realized by a computer and can be realized as a program is not limited to the first embodiment, but is the same in other embodiments.
- the image processing apparatus 1000 can be configured to generate the output image 1 by synthesizing the residual component with the noise component attenuated and the skeleton component, like the image processing apparatus 1001 shown in FIG.
- the image processing apparatus 1001 includes an initial noise-removed image generation unit 3, an initial residual component storage memory 4, an initial residual component standard deviation calculation unit 5, and an initial residual component standard.
- a deviation storage memory 6, a subgradient method parameter calculation unit 7, a subgradient method parameter storage memory 8, a subgradient method iteration control unit 9, and a skeleton component / residual component separation unit 10 are provided. Since these processes are the same as those of the image processing apparatus 1000, description thereof is omitted.
- the image processing apparatus 1001 includes a noise suppression unit 11 and a synthesis unit 12.
- the noise suppression unit 11 suppresses the noise component from the residual component supplied from the skeleton component / residual component separation unit 10 in the procedure of step S7.
- the synthesis unit 12 synthesizes the skeleton component supplied from the skeleton component / residual component separation unit 10 and the residual component (that is, the texture component) in which the noise component supplied from the noise suppression unit 11 is attenuated, and outputs an image. 2 is generated.
- the configuration of the image processing apparatus shown in FIG. 7 is an example, and other configurations may be adopted as long as the apparatus realizes the same function.
- an initial denoising image is generated using a lightweight denoising method that preserves edge components to obtain an initial solution of the TV method, and the residual component separated from the input image
- FIG. 8 is a functional block diagram for explaining the image processing method according to the second embodiment of the present invention.
- the image processing method includes an initial noise-removed image generation unit 103 that generates an initial noise-removed image by a lightweight noise-removing method that preserves edge components in an input image, and initial noise removal.
- the iterative control unit 104 that controls the subsequent skeletal component / residual component separating unit 101 based on the initial residual component separated from the input image in the image generating unit 103, and the skeletal component and residual component shown in FIG.
- Skeleton component / residual component separating unit 101 the initial residual component separated from the input image by the initial noise-removed image generating unit 103, and the skeletal component and residual component based on the TV method
- the residual noise separated from the initial noise-removed image by the skeleton component / residual component separating unit is separated from the input image by the initial noise-removed image generating unit 103.
- the skeleton component / residual component separation method for separating the skeleton component and the residual component will be described by the TV method shown in the conventional method, but is not limited thereto.
- FIG. 9 is a flowchart for explaining an image processing method according to the second embodiment of the present invention.
- the input image F is supplied to the initial noise removal image generation unit 103.
- the initial noise removal image generation unit 103 applies a lightweight noise removal method for storing edge components to the input image F, and generates an initial noise removal image U init (step S8). Further, the initial residual component V init composed of texture and noise is separated from the input image F.
- the lightweight noise removal method for storing the edge component is the same as that in the first embodiment, and thus description thereof is omitted.
- the generated initial noise-removed image U init is supplied to the skeleton component / residual component separation unit 101.
- the separated initial residual component V init is supplied not only to the iterative control unit 104 but also to the noise suppression unit 201.
- the iteration control unit 104 controls the iteration of the TV method based on the initial residual component V init supplied from the initial noise removal image generation unit 103.
- Steps S9 and S10 indicating the iterative control method are the same as those in the first embodiment, and thus description thereof is omitted.
- the standard deviation ⁇ init of the initial residual component V init calculated in step S9 is supplied to the noise suppression unit 201.
- the skeleton component / residual component separation unit 101 applies the TV method using the initial noise-removed image U init supplied from the initial noise-removed image generation unit 103 as an initial solution, and satisfies the iteration stop condition determined by the iteration control unit 104. Based on this, a skeleton component U TV composed of strong edges and flat regions and a residual component V TV composed of texture and noise are separated. Steps S1, S2, S3, S4, and S5 indicating the separation method are the same as those in the first embodiment, and thus description thereof is omitted.
- the residual component V TV is supplied to the noise suppression unit 201.
- the noise suppression unit 201 uses the residual component V init and the skeleton component / residual component supplied from the initial noise removal image generation unit 103.
- a function f for attenuating the noise component is applied to the residual component V TV supplied from the separation unit 101 to generate a residual component f (V init , V TV ) in which the noise component is attenuated (step S11).
- step S ⁇ b> 11 the function f is configured by expanding the expression (11) as a basic expression. For example, using threshold values ⁇ 1 and ⁇ 2 and coefficients ⁇ 1 and ⁇ 2 , a combination of soft decision threshold processing and factor processing is applied to residual components V init and V TV as shown in the following equation (22).
- the function f can be configured by applying.
- ⁇ 1 and ⁇ 2 are obtained by using the standard deviation ⁇ noise of the target image noise estimated in advance and the standard deviation ⁇ init of the initial residual component supplied from the iterative control unit 104, Set as a 1 and a 2 are coefficients.
- the noise component is attenuated only from the residual component V TV separated by the skeleton component / residual component separation unit 101 as shown in FIG. 2 of the first embodiment.
- the difference from the first embodiment is that the attenuation amount of the noise component is set based on the standard deviation ⁇ init of the initial residual component V init supplied from the iterative control unit 104.
- the order of the soft decision threshold process and the factor process may be reversed, or the hard decision threshold process may be combined.
- the function f is not limited to the expression (22) as long as the process has the same effect.
- a noise-removed image Z is obtained by synthesizing (that is, texture components) f (V init , V TV ) (step S6). Based on the above-described operation, the image processing method of the invention generates a noise-removed image.
- FIG. 10 is a diagram illustrating a configuration example of the image processing apparatus according to the second embodiment.
- the image processing device 1002 is a device that removes noise from the input image 1 and outputs an output image 2.
- the image processing apparatus 1002 includes an initial noise-removed image generation unit 3, an initial residual component storage memory 4, an initial residual component standard deviation calculation unit 5, and an initial residual unit. Difference component standard deviation storage memory 6, subgradient method parameter calculation unit 7, subgradient method parameter storage memory 8, subgradient method iteration control unit 9, skeleton component / residual component separation unit 10, And a synthesizing unit 12. Since these processes are the same as those of the image processing apparatus 1001, description thereof will be omitted.
- the image processing apparatus 1002 includes a noise suppression unit 13 that executes processing different from that of the image processing apparatus 1001.
- the noise suppression unit 13 the initial residual component obtained by referring to the initial residual component storage memory 4 based on the standard deviation of the initial residual component obtained by referring to the standard residual storage memory 6 of the initial residual component, From the residual component supplied from the skeleton component / residual component separation unit 10, the noise component is suppressed by the procedure of step S11.
- the configuration of the image processing apparatus shown in FIG. 10 is an example, and other configurations may be adopted as long as the apparatus realizes the same function.
- the noise can be removed at high speed while maintaining the edge and texture more.
- FIG. 11 is a functional block diagram for explaining an image processing method according to the third embodiment of the present invention.
- the image processing method includes an initial noise-removed image generation unit 103 that generates an initial noise-removed image based on the WS method, and an initial noise-removed image generation unit 103 that is separated from an input image.
- the iterative control unit 104 that controls the subsequent skeletal component / residual component separating unit based on the initial residual component, and the skeletal component / residual component separating unit that separates the skeletal component and residual component shown in FIG.
- the constraint space generation unit 304 that generates a space for constraining the skeleton component and the wavelet transform coefficient of the skeleton component generated by the skeleton component / residual component separation unit 101 are controlled. Characterized by comprising a projection unit 306 for projecting the constraint space generated in the space generator 304.
- the skeleton component / residual component separation method for separating the skeleton component and the residual component will be described by the TV method shown in the conventional method, but is not limited thereto.
- FIG. 12 is a flowchart for explaining the image processing method according to the first embodiment.
- the input image is supplied to the initial noise removal image generation unit 103.
- the initial noise-removed image generation unit 103 generates an initial noise-removed image using the WS method.
- the wavelet transform unit 301 performs wavelet transform on the supplied input image (step S12), and supplies the wavelet transform coefficients to the shrinkage unit 302 and the constraint space generation unit 304.
- the shrinkage unit 302 performs shrinkage on the wavelet transform coefficients supplied from the wavelet transform unit 301 (step S13). For example, the processing of Expression (13) and Expression (14) can be applied as one form of Shrinkage. Further, the wavelet transform coefficient to which the shrinkage is applied is supplied to the wavelet inverse transform unit 303, and the shrinkage attenuation amount ⁇ is supplied to the constraint space generation unit 304.
- the wavelet inverse transform unit 303 performs wavelet inverse transform on the wavelet transform coefficient supplied from the shrinkage unit 302 (step S14), and obtains an initial noise-removed image U init and a residual component V init .
- the obtained initial noise-removed image U init is supplied to the skeleton component / residual component separating unit 101, and the residual component V init is supplied to the iterative control unit 104.
- the constraint space generation unit 304 generates a constraint space S using the wavelet transform unit coefficients supplied from the wavelet transform unit 301 and the attenuation amount ⁇ supplied from the shrinkage unit 302 (step S15).
- the constraint space S is expressed by the following equation (25). Is done.
- WT ( ⁇ ) represents wavelet transform
- c represents a coefficient. Expression (25) indicates that when the constraint space S is wavelet transformed, the wavelet transformation coefficient is within the range of ⁇ c ⁇ ⁇ of the wavelet transformation coefficient of the input image.
- the generated constraint space S is supplied to the projection unit 306.
- the iteration control unit 104 controls the iteration of the TV method based on the noise component V init supplied from the wavelet inverse transformation unit 303. Steps S9 and S10 indicating the iterative control method are the same as those in the first embodiment, and thus description thereof is omitted.
- the skeleton component / residual component separation unit 101 applies the TV method using the initial noise-removed image U init supplied from the initial noise-removed image generation unit 103 as an initial solution, and satisfies the iteration stop condition determined by the iteration control unit 104. Based on this, a skeleton component U TV and a residual component V TV composed of texture and noise are separated. Steps S1, S2, S3, S4, and S5 indicating the separation method are the same as those in the first embodiment, and thus description thereof is omitted.
- the obtained skeleton component U TV is supplied to the wavelet transform unit 305.
- the wavelet transform unit 305 performs wavelet transform on the skeleton component U TV supplied from the skeleton component / residual component separation unit 101 (step S ⁇ b> 16), and supplies the wavelet transform coefficient to the projection unit 306.
- the projection unit 306 performs a projection process represented by the following equation (26) using the constraint space S supplied from the constraint space generation unit 304 and the wavelet transform coefficient supplied from the wavelet transform unit 305 (step S17).
- t is a wavelet transform coefficient output from the wavelet transform unit 305 and obtained by applying wavelet transform to the skeleton component U TV
- t ′ is a wavelet transform coefficient output by the projection processing to the constraint space S. It is.
- the wavelet transform coefficient of the output image Z is constrained to vary within a range of ⁇ c ⁇ ⁇ of the wavelet transform coefficient w m, n of the input image F.
- the projecting unit 306 supplies the wavelet transform coefficient after applying the projection processing of Expression (26) to the wavelet inverse transform unit 307.
- the wavelet inverse transform unit 307 performs wavelet inverse transform on the wavelet transform coefficients supplied from the projection unit 306 (step S18), and obtains an output image from which noise has been removed.
- the image processing method of the present invention generates a noise-removed image.
- FIG. 13 is a diagram illustrating a configuration example of an image processing apparatus according to the third embodiment.
- the image processing apparatus 1003 is an apparatus that performs image processing on the input image 1 and outputs an output image 2.
- the image processing apparatus 1003 includes an initial residual component storage memory 4, an initial residual component standard deviation calculation unit 5, and an initial residual component standard deviation storage memory 6.
- the subgradient method parameter calculation unit 7, the subgradient method parameter storage memory 8, the subgradient method iterative control unit 9, the skeleton component / residual component separation unit 10, the constraint space generation unit 14, the constraint A storage memory 15 and a constraint processing unit 16 are provided. Since these processes are the same as those of the image processing apparatus 1000, description thereof is omitted.
- the initial noise-removed image generation unit 3 executes initial noise-removed image generation using the WS method in steps S12, S13, and S14, and calculates the wavelet transform coefficient and attenuation obtained in the process of initial noise-removed image generation. To the constraint space generation unit 14.
- the constrained space generation unit 14 generates a constrained space from the wavelet transform coefficient and attenuation supplied from the initial noise-removed image generating unit 3 in the procedure of step S15.
- the skeleton component is processed in steps S16, S17, and S18. Limit the range of fluctuations.
- the configuration of the image processing apparatus shown in FIG. 13 is an example, and other configurations may be adopted as long as the apparatus realizes the same function.
- the third embodiment it is possible to solve the problem that the sense of resolution is deteriorated due to excessive suppression of noise by the TV method, and it is possible to remove noise at high speed while maintaining more edges and textures.
- FIG. 15 is a functional block diagram for explaining an image processing method according to the fourth embodiment of the present invention.
- FIG. 15 shows the flow of processing when multi-resolution wavelet transform of three layers is performed, but it can be easily extended to other than three layers.
- FIG. 16 shows an example of multi-resolution wavelet transform.
- FIG. 16A shows an original image
- FIG. 16B shows a one-stage wavelet transform
- FIG. 16C shows a three-layer wavelet transform.
- L indicates a low frequency side
- H indicates a high frequency side
- a combination of L and H represents a subband.
- LH indicates that the horizontal direction is the low frequency side and the vertical direction is the high frequency side.
- the LL component of the multi-resolution wavelet transform is a wavelet transform coefficient, but also has an aspect as a reduced image of the original image.
- the LL component in the third layer is a reduced image having a resolution of 1/64 of the original image, which is 1/8 in both the horizontal and vertical directions.
- the TV method is a noise removal process in the image region, it can be performed on the wavelet transform coefficient of the LL component. Since the LL component in the deep hierarchy is considered to be composed of the low frequency component of the original image, the low frequency noise can be removed by executing the denoising process in the deep hierarchy.
- FIG. 17 is a flowchart for explaining an image processing method according to the fourth embodiment.
- the input image F is supplied to the wavelet transform unit 401.
- the wavelet transform unit 401 performs wavelet transform on the supplied input image F and supplies the LL 1 component of the wavelet transform coefficient to the wavelet transform 402.
- the LL 1 component at this time is a reduced image having a resolution of 1/4 with respect to the original image. Further, the LH 1 , HL 1 , and HH 1 components are supplied to the shrinkage unit 410.
- the wavelet transform unit 402 performs wavelet transform on the LL 1 component of the first layer supplied from the wavelet transform unit 401, and supplies the LL 2 component of the obtained wavelet transform coefficient to the wavelet transform 403.
- the LL 2 component at this time is a reduced image having a resolution of 1/16 with respect to the original image. Further, the LH 2 , HL 2 , and HH 2 components are supplied to the shrinkage unit 407.
- the wavelet transform unit 403 performs wavelet transform on the LL 2 component of the second layer supplied from the wavelet transform unit 402 and supplies the obtained wavelet transform coefficients LL 3, LH 3 , HL 3 , and HH 3 to the shrinkage unit 404. .
- step S19 a three-layer multi-resolution wavelet transform is realized.
- noise removal processing is executed from the lowest resolution layer (step S20).
- the shrinkage unit 404 applies the shrinkage process to the wavelet transform coefficients supplied from the wavelet transform 403, and supplies the wavelet transform coefficients after applying the shrinkage to the inverse wavelet transform unit 405.
- the wavelet inverse transform unit 405 performs inverse wavelet transform on the wavelet transform coefficient supplied from the shrinkage unit 404 to generate an initial noise-removed image U 3, init having a resolution of 1/16 with respect to the original image (step S21).
- the skeleton component / residual component separation unit 407 is supplied. Further, the initial residual component V 3, init separated from the input image F is supplied to the iterative control unit 406.
- the iteration control unit 406 controls the iteration of the TV method in the skeleton component / residual component separation unit 407 based on the noise components V 3 and init supplied from the wavelet inverse transformation unit 405. Steps S9 and S10 indicating the iterative control method are the same as those in the first embodiment, and thus description thereof is omitted.
- the skeleton component / residual component separation unit 407 applies the TV method using the initial noise-removed image U 3, init supplied from the inverse wavelet transform unit 405 as an initial solution , and satisfies the iterative stop condition determined by the iterative control unit 104. Based on this, the skeleton component U 3, TV and the residual component V 3, TV composed of texture and noise are separated. Steps S1, S2, S3, S4, and S5 indicating the separation method are the same as those in the first embodiment, and thus description thereof is omitted. Since there is an unprocessed upper layer (step S22), the obtained skeleton component U 3, TV is regarded as the LL ′ 2 component of the wavelet transform coefficient and supplied to the inverse wavelet transform 408.
- step S23 Since the processing at the third layer is completed, the processing at the second layer is started (step S23).
- the shrinkage process is applied to the LH 2 , HL 2 , and HH 2 components of the wavelet transform coefficients supplied from the wavelet transform unit 402, and the wavelet transform coefficients LH ′ 2 and HL ′ 2 after the shrinkage application.
- HH ′ 2 components are supplied to the inverse wavelet transform unit 409.
- the wavelet inverse transform unit 409 In the wavelet inverse transform unit 409, the LH ′ 2 , HL ′ 2 , and HH ′ 2 components of the wavelet transform coefficient supplied from the shrinkage unit 408 and the LL ′ 2 component of the wavelet transform coefficient supplied from the TV method 406 are obtained. Then, the wavelet inverse transform is performed to generate an initial noise-removed image U 2, init having a resolution of 1/4 with respect to the original image (step S21), and is supplied to the skeleton component / residual component separation unit 411. Further, the initial residual component V 2, init separated from the input image F is supplied to the iterative control unit 410.
- the iteration control unit 410 controls the iteration of the TV method in the skeleton component / residual component separation unit 411 based on the noise component V 2, init supplied from the wavelet inverse transformation unit 409. Steps S9 and S10 indicating the iterative control method are the same as those in the first embodiment, and thus description thereof is omitted.
- the skeleton component / residual component separation unit 411 applies the TV method using the initial noise-removed image U 2, init supplied from the wavelet inverse transform unit 409 as an initial solution , and satisfies the iteration stop condition determined by the iteration control unit 104. Based on this, the skeleton component U 2, TV and the residual component V 2, TV composed of texture and noise are separated. Steps S1, S2, S3, S4, and S5 indicating the separation method are the same as those in the first embodiment, and thus description thereof is omitted. Since there is an unprocessed upper layer (step S22), the obtained skeleton component U 2, TV is regarded as the LL ′ 1 component of the wavelet transform coefficient and supplied to the wavelet inverse transform 413.
- step S23 Since the processing at the second hierarchy has been completed, the processing at the first hierarchy is started (step S23).
- the shrinkage unit 412 applies the shrinkage process to the LH 1 , HL 1 , and HH 1 components of the wavelet transform coefficients supplied from the wavelet transform unit 401, and the wavelet transform coefficients LH ′ 1 and HL ′ 1 after the shrinkage application.
- HH ′ 1 component is supplied to the wavelet inverse transform unit 413.
- the wavelet inverse transform unit 413 In the wavelet inverse transform unit 413, the LH ′ 1 , HL ′ 1 , and HH ′ 1 components of the wavelet transform coefficients supplied from the shrinkage unit 412 and the LL of the wavelet transform coefficients supplied from the skeleton component / residual component separation unit 411 'Perform inverse wavelet transform using one component to generate an initial noise-removed image U 1, init having the same resolution as the original image (step S21), and supply it to the skeleton component / residual component separation unit 415. Further, the initial residual component V 1, init separated from the input image F is supplied to the iterative control unit 414.
- the iteration control unit 414 controls the iteration of the TV method in the skeleton component / residual component separation unit 415 based on the noise component V 1, init supplied from the wavelet inverse transformation unit 413. Steps S9 and S10 indicating the iterative control method are the same as those in the first embodiment, and thus description thereof is omitted.
- the skeleton component / residual component separation unit 415 applies the TV method using the initial noise-removed image U 1, init having the same resolution as the original image supplied from the wavelet inverse transformation unit 413 as an initial solution, and the skeleton component U 1, TV and texture component V 1, TV are obtained.
- the obtained skeleton component U 1, TV is output as a noise-removed image Z (step S6).
- the image processing method of the invention Based on the above-described operation, the image processing method of the invention generates a noise-removed image.
- FIG. 18 is a diagram illustrating a configuration example of an image processing apparatus according to the fourth embodiment.
- the image processing apparatus 1004 is an apparatus that performs image processing on the input image 1 and outputs an output image 2.
- the image processing apparatus 1004 includes an initial residual component storage memory 4, an initial residual component standard deviation calculator 5, and an initial residual component standard deviation storage memory 6.
- the multi-resolution wavelet transform unit 17 applies multi-resolution wavelet transform to the input image 1 in the procedure of step S19.
- the high frequency component storage memory 18 stores the high frequency component of the wavelet transform coefficient supplied from the multi-resolution wavelet transform unit 17.
- the initial noise-removed image generation unit 3 refers to the low-frequency component of the wavelet transform coefficient of the processing target layer supplied from the multi-resolution wavelet transform unit 17 or the output image control unit 19 and the high-frequency component storage memory 18. From the acquired high-frequency component of the wavelet transform coefficient of the processing target layer, an initial noise-removed image is generated by the procedure of step S21.
- the output image control unit 19 uses the skeleton component supplied from the skeleton component / residual component separation unit 10 as the output image 2 when the resolution of the processing target layer matches the resolution of the input image 1.
- the skeletal component supplied from the skeleton component / residual component separation unit 10 is converted to a wavelet transform coefficient having a high one-layer resolution by the procedures of steps S22 and S23. It is regarded as a low frequency component and supplied to the initial noise removal image generation unit 3.
- FIG. 18 Note that the configuration of the image processing apparatus illustrated in FIG. 18 is merely an example, and other configurations may be employed as long as the apparatus implements similar functions.
- the solution converges with the TV method for each layer. Reduces the number of required Chambolle Projection iterations and removes not only high-frequency noise but also low-frequency noise at high speed.
- 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 embodiments are realized by a processor that operates according to a program stored in the program memory.
- Appendix 2 Extract edge components and texture components from the residual components, The image processing method according to appendix 1, wherein the skeleton component, the edge component, and the texture component are synthesized to generate an output image.
- Appendix 3 An edge component and a texture component are extracted from the initial residual component and the residual component, The image processing method according to appendix 1, wherein the skeleton component, the edge component, and the texture component are synthesized to generate an output image.
- Appendix 4 The image processing method according to appendix 2 or appendix 3, wherein the edge component and the texture component are extracted based on a standard deviation of the initial residual component and a standard deviation of the residual component.
- Appendix 5 Generate multiple images with different resolutions from the input image, When an output image is generated by applying the image processing method described in any one of appendix 1 to appendix 4 for each different image, an output image generated based on a low-resolution image is one resolution high An image processing method used for generating an initial noise-removed image in an image.
- An initial noise-removed image generating means for generating an initial noise-removed image by a noise-removing method for storing edge components in the input image
- a skeleton component / residual component separation unit that separates the initial noise-removed image into a skeleton component and a residual component by repetitive calculation based on a predetermined energy, and generates the skeleton component as an output image
- An image processing apparatus comprising: control means for controlling the repetitive calculation based on the initial residual component.
- the skeleton component / residual component separating means is: Extract edge components and texture components from the residual components, The image processing apparatus according to appendix 6, wherein the skeleton component, the edge component, and the texture component are combined to generate an output image.
- the skeleton component / residual component separating means is: An edge component and a texture component are extracted from the initial residual component and the residual component, The image processing apparatus according to appendix 6, wherein the skeleton component, the edge component, and the texture component are combined to generate an output image.
- the skeleton component / residual component separating means is: The image processing apparatus according to appendix 7 or appendix 8, wherein the edge component and the texture component are extracted based on a standard deviation of the initial residual component and a standard deviation of the residual component.
- (Appendix 10) Means for generating a plurality of images having different resolutions from an input image;
- the initial noise-removed image generating means, the skeleton component / residual component separating means, and the control means are provided corresponding to each resolution image,
- Appendix 12 Processing to extract an edge component and a texture component from the residual component;
- Appendix 13 Processing to extract an edge component and a texture component from the initial residual component and the residual component;
- Appendix 14 The program according to appendix 12 or appendix 13, wherein the edge component and the texture component are extracted based on a standard deviation of the initial residual component and a standard deviation of the residual component.
- Appendix 15 A process of generating a plurality of images having different resolutions from an input image; When an output image is generated by applying the processing according to any one of appendix 11 to appendix 14 for each different image, an output image generated based on an image with a low resolution is converted into an image with a high resolution. A program used to generate an initial noise-removed image.
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Facsimile Image Signal Circuits (AREA)
Abstract
Description
骨格成分UT-Vは、以下の式(2)で表されるUT-Vの全変分ノルムJ(UT-V)を最小化することで得られる。
なお、式中のxは骨格成分UT-Vの水平方向画素位置、yは骨格成分UT-Vの垂直方向画素位置を示す。
を計算する(ステップS2)。
反復の停止条件について述べる。入力画像の高さをM、幅をN、k回目の反復の時の残差成分
のm行n列要素を
とすると、予め推定した対象画像のノイズの標準偏差σnoiseを用いて、
となったときに反復を停止する(ステップS3)ことが考えられる。式(4)はつまり、残差成分
がノイズであるという仮定に基づく。あるいは、閾値εを用いて、
となったときに反復を停止する(ステップS3)ことが考えられる。式(5)はつまり、k回目の反復の結果
とk+1回目の反復
を比較し、変動量が十分小さいと判断したときに解が収束したとみなしている。
図21は、骨格成分・残差成分分離部101で分離された残差成分のノイズ成分を減衰させ、骨格成分に合成する処理を有する、図19のT-V法を用いた画像処理方法を拡張した画像処理方法である。
関数fとしては、図23(a)で示される軟判定閾値処理、図23(b)で示される硬判定閾値処理、図23(c)で示されるfactor処理などがあるが、これらに限定されるものではない。関数fの入力値をx、出力値をf(x)、閾値をτとすると、図23(a)で示される軟判定閾値処理は、以下の式(8)で表される。
なお、式中のsign関数は正負符号を出力する関数である。
本発明の第1の実施の形態を説明する。
生成された初期ノイズ除去画像Uinitは骨格成分・残差成分分離部101に供給される。また、分離された初期残差成分Vinitは反復制御部104に供給される。
なお、式(13)中のsign関数は正負符号を出力する関数である。
とすると、標準偏差σinitは以下の式(15)で表されるものとして計算する。
従来のT-V法では式(3)で示されるパラメータβや固定値の最大反復回数kmaxが用いられるが、本実施の形態では、予め推定したノイズの標準偏差σnoiseと、ステップS9で計算した標準偏差σinitから、各パラメータを補正する(ステップS10)。
すなわち、式(16)において、σinitがσnoiseに近い場合には、T-V法の収束解の近傍にあることを意味しており、収束速度を向上させることが可能となる。具体的には、1回の反復において探索する領域を狭めることで収束速度を向上させる。
なお、max関数は2つの入力値から大きい方を出力する関数、min関数は2つの入力値から小さい方を出力する関数、round関数は整数値への丸め関数を示す。式(17)より、予め推定したノイズの標準偏差σnoiseとVinitの標準偏差σinitとが近い値であるときは、良い初期解が得られたとみなしてT-V法の反復回数を削減する。逆に、予め推定したノイズの標準偏差σnoiseとVinitの標準偏差σinitとが近い値でないときは、良い初期解が得られなかったとみなしてT-V法の反復回数をkmaxに近い値に保つ。
本実施の形態の骨格成分・残差成分分離部101での処理では、ステップS1における初期値設定において、従来法と異なり、
と設定する。さらに、ステップS3の反復停止判定において、従来法の式(4)ではなく、以下の式(19)を反復停止条件として用いる。
式(19)より、従来のT-V法の反復停止条件と同様に、σnoiseに基づいてT-V法の反復を制御することができ、Vinitの標準偏差σinitがσnoiseに近い場合には探索する領域を狭めることで少ない反復回数でT-V法の解が収束するため、処理の高速化が実現できる。また、ステップS4の反復停止判定において、従来法のkmaxでなく、ステップS10で設定したk’maxを用いる。なお、ステップS2、S5における処理は従来法と同様であるため、説明を省略する。
また、当然ながら、本実施の形態では、図21のように、ノイズ成分を減衰させた残差成分VT-Vと、骨格成分UT-Vを合成して、ノイズ除去画像Zを出力する構成も可能である。発明の画像処理方法では、図4のような構成となり、図5のようなフローチャートで表される。骨格成分・残差成分分離部101で分離された残差成分VT-Vをノイズ抑制部102に供給し、以下の式(21)のように、VT-Vにノイズ成分を減衰させる関数fを作用させ(ステップS7)、さらに、ノイズ成分が減衰された残差成分f(VT-V)と、骨格成分・残差成分分離部101から供給される骨格成分UT-Vを合成し、ノイズ除去画像Zとして出力する(ステップS6)。
次に、第1の実施の形態の画像処理方法を適用した具体的な画像処理装置について説明する。図6は、第1の実施の形態の画像処理装置の構成例を示す図である。
本発明の第2の実施の形態を説明する。
ここで、τ1とτ2は、予め推定した対象画像のノイズの標準偏差σnoiseと反復制御部104から供給された初期残差成分の標準偏差σinitを用いて、
として設定される。a1とa2は係数である。
上述した動作に基づいて、発明の画像処理方法はノイズ除去画像を生成する。
本発明の第3の実施の形態を説明する。
なお、式中WT(・)はウェーブレット変換、cは係数を表す。式(25)は、制約空間Sをウェーブレット変換すると、そのウェーブレット変換係数は、入力画像のウェーブレット変換係数の±c×λの範囲内にあることを示す。
なお、tはウェーブレット変換部305から出力された、骨格成分UT-Vにウェーブレット変換を適用して得られたウェーブレット変換係数であり、t’は制約空間Sへの射影処理で出力されるウェーブレット変換係数である。
本発明の第4の実施の形態を説明する。
入力画像におけるエッジを保存しながらノイズを低減して初期ノイズ除去画像を生成し、
前記入力画像と前記初期ノイズ除去画像とから算出される初期残差成分に基づいて、予め定義されたエネルギーに基づく繰り返し演算を制御し、
前記制御された繰り返し演算によって、前記初期ノイズ除去画像を骨格成分と残差成分に分離し、前記骨格成分を出力画像として生成する
画像処理方法。
前記残差成分からエッジ成分とテクスチャ成分を抽出し、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する付記1に記載の画像処理方法。
前記初期残差成分と前記残差成分とからエッジ成分とテクスチャ成分とを抽出し、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する
付記1に記載の画像処理方法。
前記初期残差成分の標準偏差と前記残差成分の標準偏差とに基づいて、前記エッジ成分と前記テクスチャ成分とを抽出する
付記2又は付記3に記載の画像処理方法。
入力画像から、複数の解像度の異なる画像を生成し、
異なる画像毎に、付記1から付記4のいずれかに記載の画像処理方法を適用して出力画像を生成する際に、低い解像度の画像に基づいて生成された出力画像を、一つ解像度の高い画像における初期ノイズ除去画像の生成に用いる
画像処理方法。
入力画像中のエッジ成分を保存するノイズ除去方法によって初期ノイズ除去画像を生成する初期ノイズ除去画像生成手段と、
予め定義されたエネルギーに基づく繰り返し演算によって、前記初期ノイズ除去画像を骨格成分と残差成分に分離し、前記骨格成分を出力画像として生成する骨格成分・残差成分分離手段と、
前記初期残差成分に基づいて、前記繰り返し演算を制御する制御手段と
を有する画像処理装置。
前記骨格成分・残差成分分離手段は、
前記残差成分からエッジ成分とテクスチャ成分を抽出し、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する付記6に記載の画像処理装置。
前記骨格成分・残差成分分離手段は、
前記初期残差成分と前記残差成分とからエッジ成分とテクスチャ成分とを抽出し、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する
付記6に記載の画像処理装置。
前記骨格成分・残差成分分離手段は、
前記初期残差成分の標準偏差と前記残差成分の標準偏差とに基づいて、前記エッジ成分と前記テクスチャ成分とを抽出する
付記7又は付記8に記載の画像処理装置。
入力画像から、複数の解像度の異なる画像を生成する手段を有し、
前記初期ノイズ除去画像生成手段、前記骨格成分・残差成分分離手段及び前記制御手段を、各解像度の画像に対応して設け、
前記初期ノイズ除去画像生成手段は、対応する解像度よりも一つ解像度の低いに基づいて生成された出力画像を、初期ノイズ除去画像の生成に用いる
付記6から付記9のいずれかに記載の画像処理装置。
入力画像におけるエッジを保存しながらノイズを低減して初期ノイズ除去画像を生成する処理と、
前記入力画像と前記初期ノイズ除去画像とから算出される初期残差成分に基づいて、予め定義されたエネルギーに基づく繰り返し演算を制御する処理と、
前記制御された繰り返し演算によって、前記初期ノイズ除去画像を骨格成分と残差成分に分離し、前記骨格成分を出力画像として生成する処理と
をコンピュータに実行させるプログラム。
前記残差成分からエッジ成分とテクスチャ成分を抽出する処理と、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する処理と
をコンピュータに実行させる付記11に記載のプログラム。
前記初期残差成分と前記残差成分とからエッジ成分とテクスチャ成分とを抽出する処理と、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する処理と
をコンピュータに実行させる付記11に記載のプログラム。
前記初期残差成分の標準偏差と前記残差成分の標準偏差とに基づいて、前記エッジ成分と前記テクスチャ成分とを抽出する
付記12又は付記13に記載のプログラム。
入力画像から、複数の解像度の異なる画像を生成する処理を有し、
異なる画像毎に、付記11から付記14のいずれかに記載の処理を適用して出力画像を生成する際に、低い解像度の画像に基づいて生成された出力画像を、一つ解像度の高い画像における初期ノイズ除去画像の生成に用いる
プログラム。
2 出力画像
3 初期ノイズ除去画像生成部
4 初期残差成分保存メモリ
5 標準偏差計算部
6 標準偏差保存メモリ
7 パラメータ計算部
8 パラメータ保存メモリ
9 反復制御部
10 骨格成分・残差成分分離部
11 ノイズ抑制部
12 合成部
13 ノイズ抑制部
14 制約空間生成部
15 制約保存メモリ
16 制約処理部
19 出力画像制御部
101 骨格成分・残差成分分離部
102 ノイズ抑制部
103 初期ノイズ除去画像生成部
104 反復制御部
201 ノイズ抑制部
301 ウェーブレット変換部
302 Shrinkage部
303 ウェーブレット逆変換部
304 制約空間生成部
305 ウェーブレット変換部
306 射影部
307 ウェーブレット逆変換部
401 ウェーブレット変換部
402 ウェーブレット変換部
403 ウェーブレット変換部
412 Shrinkage部
404 Shrinkage部
405 ウェーブレット逆変換部
406 反復制御部
407 骨格成分・残差成分分離部
408 Shrinkage部
409 ウェーブレット逆変換部
410 反復制御部
411 骨格成分・残差成分分離部
413 ウェーブレット逆変換部
414 反復制御部
415 骨格成分・残差成分分離部
1000 画像処理装置
Claims (10)
- 入力画像におけるエッジを保存しながらノイズを低減して初期ノイズ除去画像を生成し、
前記入力画像と前記初期ノイズ除去画像とから算出される初期残差成分に基づいて、予め定義されたエネルギーに基づく繰り返し演算を制御し、
前記制御された繰り返し演算によって、前記初期ノイズ除去画像を骨格成分と残差成分に分離し、前記骨格成分を出力画像として生成する
画像処理方法。 - 前記残差成分からエッジ成分とテクスチャ成分を抽出し、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する請求項1に記載の画像処理方法。 - 前記初期残差成分と前記残差成分とからエッジ成分とテクスチャ成分とを抽出し、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する
請求項1に記載の画像処理方法。 - 前記初期残差成分の標準偏差と前記残差成分の標準偏差とに基づいて、前記エッジ成分と前記テクスチャ成分とを抽出する
請求項2又は請求項3に記載の画像処理方法。 - 入力画像から、複数の解像度の異なる画像を生成し、
異なる画像毎に、請求項1から請求項4のいずれかに記載の画像処理方法を適用して出力画像を生成する際に、低い解像度の画像に基づいて生成された出力画像を、一つ解像度の高い画像における初期ノイズ除去画像の生成に用いる
画像処理方法。 - 入力画像中のエッジ成分を保存するノイズ除去方法によって初期ノイズ除去画像を生成する初期ノイズ除去画像生成手段と、
予め定義されたエネルギーに基づく繰り返し演算によって、前記初期ノイズ除去画像を骨格成分と残差成分に分離し、前記骨格成分を出力画像として生成する骨格成分・残差成分分離手段と、
前記初期残差成分に基づいて、前記繰り返し演算を制御する制御手段と
を有する画像処理装置。 - 前記骨格成分・残差成分分離手段は、
前記残差成分からエッジ成分とテクスチャ成分を抽出し、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する請求項6に記載の画像処理装置。 - 前記骨格成分・残差成分分離手段は、
前記初期残差成分と前記残差成分とからエッジ成分とテクスチャ成分とを抽出し、
前記骨格成分と前記エッジ成分と前記テクスチャ成分とを合成し、出力画像を生成する
請求項6に記載の画像処理装置。 - 前記骨格成分・残差成分分離手段は、
前記初期残差成分の標準偏差と前記残差成分の標準偏差とに基づいて、前記エッジ成分と前記テクスチャ成分とを抽出する
請求項7又は請求項8に記載の画像処理装置。 - 入力画像から、複数の解像度の異なる画像を生成する手段を有し、
前記初期ノイズ除去画像生成手段、前記骨格成分・残差成分分離手段及び前記制御手段を、各解像度の画像に対応して設け、
前記初期ノイズ除去画像生成手段は、対応する解像度よりも一つ解像度の低いに基づいて生成された出力画像を、初期ノイズ除去画像の生成に用いる
請求項6から請求項9のいずれかに記載の画像処理装置。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2014512683A JP6222481B2 (ja) | 2012-04-27 | 2013-04-25 | 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 |
US14/397,202 US9430816B2 (en) | 2012-04-27 | 2013-04-25 | Image processing method, image processing system, image processing device, and image processing program capable of removing noises over different frequency bands, respectively |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2012-102019 | 2012-04-27 | ||
JP2012102019 | 2012-04-27 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2013161940A1 true WO2013161940A1 (ja) | 2013-10-31 |
Family
ID=49483244
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2013/062202 WO2013161940A1 (ja) | 2012-04-27 | 2013-04-25 | 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 |
Country Status (3)
Country | Link |
---|---|
US (1) | US9430816B2 (ja) |
JP (1) | JP6222481B2 (ja) |
WO (1) | WO2013161940A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2023073805A1 (ja) * | 2021-10-26 | 2023-05-04 | 日本電信電話株式会社 | モデル学習装置、方法およびプログラム |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2015143624A1 (en) * | 2014-03-25 | 2015-10-01 | Spreadtrum Communications(Shanghai) Co., Ltd. | Methods and systems for denoising images |
US9792692B2 (en) * | 2015-05-29 | 2017-10-17 | Ncr Corporation | Depth-based image element removal |
WO2018220780A1 (ja) * | 2017-06-01 | 2018-12-06 | 日本電気株式会社 | 画像生成装置、画像生成方法およびプログラムを記憶する記憶媒体 |
US10643313B2 (en) * | 2018-01-19 | 2020-05-05 | Bae Systems Information And Electronic Systems Integration Inc. | Methods for image denoising and deblurring |
CN109785249A (zh) * | 2018-12-22 | 2019-05-21 | 昆明理工大学 | 一种基于持续性记忆密集网络的图像高效去噪方法 |
KR102279867B1 (ko) * | 2020-06-02 | 2021-07-21 | 주식회사 슈프리마아이디 | 조명 노이즈를 제거한 이미지 생성 방법 및 상기 이미지 생성 방법을 이용한 이미지 생성 장치 |
CN114066738A (zh) * | 2020-07-31 | 2022-02-18 | 北京小米移动软件有限公司 | 图像处理方法、装置、电子设备及存储介质 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001167261A (ja) * | 1999-09-27 | 2001-06-22 | Fuji Photo Film Co Ltd | 画像処理方法および装置並びに記録媒体 |
JP2010033453A (ja) * | 2008-07-30 | 2010-02-12 | Olympus Corp | 成分抽出補正装置、成分抽出補正方法、成分抽出補正プログラム、又は電子機器 |
JP2010166598A (ja) * | 2010-03-10 | 2010-07-29 | Kanagawa Univ | 画像処理装置および撮像装置 |
JP2011043901A (ja) * | 2009-08-19 | 2011-03-03 | Olympus Corp | 画像処理装置、画像処理方法、画像処理プログラム、および、電子機器 |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1059811A2 (en) | 1999-06-10 | 2000-12-13 | Fuji Photo Film Co., Ltd. | Method and system for image processing, and recording medium |
JP5044518B2 (ja) * | 2008-09-17 | 2012-10-10 | 株式会社東芝 | 画像処理装置及びコンピュータ端末 |
JP4776705B2 (ja) * | 2009-03-06 | 2011-09-21 | 株式会社東芝 | 画像処理装置および方法 |
US8588461B2 (en) * | 2010-03-22 | 2013-11-19 | Brigham Young University | Robust watermarking for digital media |
JP5159844B2 (ja) * | 2010-09-03 | 2013-03-13 | 株式会社東芝 | 画像処理装置 |
-
2013
- 2013-04-25 WO PCT/JP2013/062202 patent/WO2013161940A1/ja active Application Filing
- 2013-04-25 JP JP2014512683A patent/JP6222481B2/ja active Active
- 2013-04-25 US US14/397,202 patent/US9430816B2/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001167261A (ja) * | 1999-09-27 | 2001-06-22 | Fuji Photo Film Co Ltd | 画像処理方法および装置並びに記録媒体 |
JP2010033453A (ja) * | 2008-07-30 | 2010-02-12 | Olympus Corp | 成分抽出補正装置、成分抽出補正方法、成分抽出補正プログラム、又は電子機器 |
JP2011043901A (ja) * | 2009-08-19 | 2011-03-03 | Olympus Corp | 画像処理装置、画像処理方法、画像処理プログラム、および、電子機器 |
JP2010166598A (ja) * | 2010-03-10 | 2010-07-29 | Kanagawa Univ | 画像処理装置および撮像装置 |
Non-Patent Citations (1)
Title |
---|
HARUYA AIZAWA ET AL.: "Removal of Signal- Dependent Noise through the TV-L1 Nonlinear Image Decomposition", THE TRANSACTIONS OF THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS, vol. J91-D, no. 8, 1 August 2008 (2008-08-01), pages 1998 - 2001 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2023073805A1 (ja) * | 2021-10-26 | 2023-05-04 | 日本電信電話株式会社 | モデル学習装置、方法およびプログラム |
Also Published As
Publication number | Publication date |
---|---|
US20150086126A1 (en) | 2015-03-26 |
JPWO2013161940A1 (ja) | 2015-12-24 |
JP6222481B2 (ja) | 2017-11-01 |
US9430816B2 (en) | 2016-08-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
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 | |
Feng et al. | Speckle reduction via higher order total variation approach | |
US9665792B2 (en) | Method and apparatus for generating sharp image based on blurry image | |
WO2009107197A1 (ja) | 画像処理装置、画像処理方法および画像処理プログラム | |
WO2010086973A1 (ja) | 画像処理装置、画像処理プログラムおよび画像処理方法 | |
JP5027757B2 (ja) | 動画像雑音除去装置、その方法およびそのプログラム | |
KR101362011B1 (ko) | 링잉 아티펙트없는 블러 제거 방법 | |
US10229479B2 (en) | Image signal processing apparatus, image signal processing method and image signal processing program | |
Deng | Guided wavelet shrinkage for edge-aware smoothing | |
Chen et al. | A novel multi-image super-resolution reconstruction method using anisotropic fractional order adaptive norm | |
CN107590781B (zh) | 基于原始对偶算法的自适应加权tgv图像去模糊方法 | |
JP2009271725A (ja) | 画像復元装置、画像復元方法及び画像復元プログラム | |
Ding et al. | Image deblurring using a pyramid-based Richardson-Lucy algorithm | |
Hiraoka et al. | Reduction of iterative calculation and quality improvement for generation of moire-like images using bilateral filter | |
Lakshman et al. | Image interpolation using shearlet based sparsity priors | |
KR20110090068A (ko) | 영상의 움직임 보상 노이즈 제거 장치 및 방법 | |
JP5784472B2 (ja) | 画像処理装置、方法及びプログラム | |
Liu et al. | A novel variational model for retinex in presence of severe noises | |
Wang et al. | A new bandwidth adaptive non-local kernel regression algorithm for image/video restoration and its GPU realization | |
Wang et al. | A new method for motion-blurred image blind restoration based on huber Markov random field | |
Hong et al. | Single image motion deblurring using anisotropic regularization | |
JP6982479B2 (ja) | 超解像化装置及びプログラム | |
JP7437249B2 (ja) | 雑音除去装置及びそのプログラム | |
Young et al. | Optimized decoding of JPEG images based on generalized graph Laplacians | |
Lee et al. | Non-blind image deconvolution with adaptive regularization |
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: 13782270 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2014512683 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 14397202 Country of ref document: US |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 13782270 Country of ref document: EP Kind code of ref document: A1 |