WO2015067186A1 - 一种用于图像降噪的方法及终端 - Google Patents

一种用于图像降噪的方法及终端 Download PDF

Info

Publication number
WO2015067186A1
WO2015067186A1 PCT/CN2014/090421 CN2014090421W WO2015067186A1 WO 2015067186 A1 WO2015067186 A1 WO 2015067186A1 CN 2014090421 W CN2014090421 W CN 2014090421W WO 2015067186 A1 WO2015067186 A1 WO 2015067186A1
Authority
WO
WIPO (PCT)
Prior art keywords
noise reduction
frequency wavelet
component
wavelet coefficient
low
Prior art date
Application number
PCT/CN2014/090421
Other languages
English (en)
French (fr)
Inventor
朱聪超
罗巍
杨小伟
邓斌
Original Assignee
华为终端有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为终端有限公司 filed Critical 华为终端有限公司
Priority to US15/033,550 priority Critical patent/US9904986B2/en
Priority to JP2016524563A priority patent/JP6216987B6/ja
Priority to EP14860150.3A priority patent/EP3051485B1/en
Publication of WO2015067186A1 publication Critical patent/WO2015067186A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/70Denoising; Smoothing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/10Image enhancement or restoration using non-spatial domain filtering
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/20Image enhancement or restoration using local operators
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10024Color image
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20016Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20048Transform domain processing
    • G06T2207/20064Wavelet transform [DWT]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20172Image enhancement details
    • G06T2207/20192Edge enhancement; Edge preservation

Definitions

  • the present invention relates to the field of image processing, and more particularly to a method and terminal for image noise reduction.
  • the window filtering method which has nothing to do with the image content, has a small amount of calculation, but the detail loss of the image is severe.
  • the Non-local Means algorithm based on image structure similarity analysis performs well in terms of detail retention and color protection, but the algorithm complexity is high. low efficiency. The contradiction between noise reduction effect and efficiency is more prominent.
  • Embodiments of the present invention provide a method and a terminal for image noise reduction, which can improve the noise reduction effect and efficiency of noise reduction on an image.
  • an embodiment of the present invention provides a method for image noise reduction, where the method includes:
  • the three components after the noise reduction Combine to obtain image data after noise reduction
  • the at least one component after the noise reduction is one or two components, combining the at least one component after the noise reduction with other components of the three components to obtain image data after noise reduction.
  • the method further includes:
  • the performing attenuation function based on the edge information based on the edge information of the high frequency wavelet coefficients of each component includes:
  • y ⁇ x+(1 ⁇ )h(x), where ⁇ is a parameter related to edge strength, and h(x) is an attenuation function with respect to x;
  • Wavelet reconstruction is performed according to the high frequency wavelet coefficients after each component noise reduction and the low frequency wavelet coefficients after each component noise reduction, to obtain the at least one component after noise reduction.
  • the at least one component of the luminance component y, the chrominance components u and v of the image data is wavelet-decomposed to obtain each The high frequency wavelet coefficients and low frequency wavelet coefficients of the components, including:
  • n-layer wavelet decomposition on at least one of the luminance component y and the chrominance components u and v of the image data to obtain n-layer high-frequency wavelet coefficients and n-layer low-frequency wavelet coefficients of each component, where n ⁇ 2, n is an integer;
  • At least one component obtained after the noise reduction includes:
  • A recursively denoising the low-frequency wavelet coefficients of the nth layer to obtain the low-frequency wavelet coefficients of the n-th layer after noise reduction, according to the low-frequency wavelet coefficients of the n-th layer noise reduction and the high-frequency wavelet coefficients of the n-th layer Wavelet reconstruction, obtaining the low frequency wavelet coefficients after the n-1th layer of noise reduction;
  • wavelet reconstruction is performed according to the low frequency wavelet coefficient and the i-th high frequency wavelet coefficient after the second noise reduction of the i-th layer, and a component after noise reduction is obtained.
  • the method further includes: performing noise reduction on the high frequency wavelet coefficients of each component by using an edge function based attenuation function;
  • the performing attenuation function based on the edge information based on the edge information of the high frequency wavelet coefficients of each component includes:
  • y j ⁇ j x j +(1- ⁇ j )h(x j ), j ⁇ 1, j is an integer;
  • y j is the value of the jth high frequency wavelet coefficient after noise reduction
  • x j is the value of the jth high frequency wavelet coefficient
  • h(x j ) is the attenuation function for x j
  • ⁇ j is the jth The edge intensity coefficient corresponding to the edge intensity of the pixel corresponding to the high-frequency wavelet coefficient, 0 ⁇ ⁇ j ⁇ 1.
  • the performing the wavelet according to the low frequency wavelet coefficient after the nth layer noise reduction and the high frequency wavelet coefficient of the nth layer Reconstruction, the low-frequency wavelet coefficients after the n-1th layer of noise reduction are obtained, specifically:
  • the wavelet reconstruction according to the low frequency wavelet coefficient and the ith layer high frequency wavelet coefficient after the second noise reduction of the ith layer is specifically:
  • Wavelet reconstruction is performed according to the low frequency wavelet coefficient after the second noise reduction of the i-th layer and the high frequency wavelet coefficient after the i-th layer noise reduction.
  • the recursive noise reduction comprises:
  • the noise reduction result of the kth low frequency wavelet coefficient the value of the kth low frequency wavelet coefficient + f (the noise reduction result of the k-1th low frequency wavelet coefficient - the value of the kth low frequency wavelet coefficient)
  • k>1 k is an integer
  • x represents the difference between the noise reduction result of the k-1th low frequency wavelet coefficient and the kth low frequency wavelet coefficient
  • y represents the drop Noise intensity.
  • the method further includes:
  • the edge strength of the pixel corresponding to the jth high frequency wavelet coefficient comprises: a jth height The edge intensity corresponding to at least one of the three components y, u, and v corresponding to the pixel corresponding to the frequency wavelet coefficient.
  • the edge strength corresponding to the at least one component is based on the high frequency wavelet coefficient and the low frequency wavelet coefficient of the at least one component Edge strength.
  • the attenuation function is a wavelet threshold function, including at least one of the following: a hard threshold function, a soft threshold function .
  • the at least one direction comprises at least one of: from left to right, from right to left, from top to bottom, From the bottom up.
  • an embodiment of the present invention provides a terminal for image noise reduction, where the terminal includes:
  • An image obtaining unit configured to acquire image data of an image
  • An image decomposition unit configured to perform wavelet decomposition on at least one of the luminance component y and the chrominance components u and v of the image data to obtain a high frequency wavelet coefficient and a low frequency wavelet coefficient of each component;
  • An image noise reduction processing unit configured to perform recursive noise reduction on the low frequency wavelet coefficients of each component, to obtain low frequency wavelet coefficients of each component after noise reduction, according to high frequency wavelet coefficients and Decoding the low-frequency wavelet coefficients after each component noise reduction to obtain at least one component after noise reduction;
  • a noise reduction image acquisition unit configured to: when the at least one component after the noise reduction is three components, combine the three components after the noise reduction to obtain image data after noise reduction;
  • the at least one component after the noise reduction is one or two components, combining the at least one component after the noise reduction with other components of the three components to obtain image data after noise reduction.
  • the terminal further includes an image high-frequency processing unit, configured to perform noise reduction on the high-frequency wavelet coefficients of each component by using an edge function-based attenuation function , including performing noise reduction on the high frequency wavelet coefficients of each component according to the following formula,
  • y ⁇ x+(1 ⁇ )h(x), where ⁇ is a parameter related to edge strength, and h(x) is an attenuation function with respect to x;
  • the image noise reduction processing unit is configured to perform wavelet reconstruction according to the high frequency wavelet coefficients of each component and the low frequency wavelet coefficients after each component noise reduction, to obtain at least one component after noise reduction, specifically Wavelet reconstruction is performed according to the high frequency wavelet coefficients after each component noise reduction and the low frequency wavelet coefficients after each component noise reduction, to obtain the at least one component after noise reduction.
  • the image decomposition unit is specifically configured to perform at least one of a luminance component y, a chrominance component u, and a v component of the image data.
  • the n-layer wavelet decomposition obtains n-layer high-frequency wavelet coefficients and n-layer low-frequency wavelet coefficients of each component, where n ⁇ 2, n is an integer;
  • the image noise reduction processing unit is specifically configured to perform the following processing on each component:
  • A recursively denoising the low-frequency wavelet coefficients of the nth layer to obtain the low-frequency wavelet coefficients of the n-th layer after noise reduction, according to the low-frequency wavelet coefficients of the n-th layer noise reduction and the high-frequency wavelet coefficients of the n-th layer Wavelet reconstruction, obtaining the low frequency wavelet coefficients after the n-1th layer of noise reduction;
  • wavelet reconstruction is performed according to the low frequency wavelet coefficient and the i-th high frequency wavelet coefficient after the second noise reduction of the i-th layer, and a component after noise reduction is obtained.
  • the terminal further includes an image high-frequency processing unit, configured to perform edge-based on the high-frequency wavelet coefficients of each component Degrading the attenuation function of the information, comprising performing noise reduction on each layer of the high frequency wavelet coefficients of the n layers of each component according to the following formula,
  • y j ⁇ j x j +(1- ⁇ j )h(x j ), j ⁇ 1, j is an integer;
  • y j is the value of the jth high frequency wavelet coefficient after noise reduction
  • x j is the value of the jth high frequency wavelet coefficient
  • h(x j ) is the attenuation function for x j
  • ⁇ j is the jth The edge intensity coefficient corresponding to the edge intensity of the pixel corresponding to the high-frequency wavelet coefficient, 0 ⁇ ⁇ j ⁇ 1.
  • the image noise reduction processing unit is configured to perform, according to the n-th layer, the low-frequency wavelet coefficient and the n-th layer
  • the high-frequency wavelet coefficients are wavelet reconstructed, and the low-frequency wavelet coefficients obtained after the n-1th layer of noise reduction are specifically used for the low-frequency wavelet coefficients after the n-th layer noise reduction and the high-frequency noise after the n-th layer noise reduction.
  • the wavelet coefficients are reconstructed by wavelet, and the low-frequency wavelet coefficients of the n-1th layer after noise reduction are obtained;
  • the image noise reduction processing unit is configured to perform wavelet reconstruction according to the low frequency wavelet coefficient and the ith layer high frequency wavelet coefficient after the second noise reduction of the ith layer, and is specifically configured to perform secondary reduction according to the ith layer
  • the low-frequency wavelet coefficients after noise and the high-frequency wavelet coefficients after the i-th layer noise reduction are wavelet reconstructed.
  • the recursive noise reduction in the fifth embodiment of the second aspect, includes:
  • the noise reduction result of the kth low frequency wavelet coefficient the value of the kth low frequency wavelet coefficient + f (the noise reduction result of the k-1th low frequency wavelet coefficient - the value of the kth low frequency wavelet coefficient)
  • k>1 k is an integer
  • x represents the difference between the noise reduction result of the k-1th low frequency wavelet coefficient and the kth low frequency wavelet coefficient
  • y represents the drop Noise intensity.
  • the terminal further includes a detail restoring unit, configured to, in the image noise reduction processing unit, the kth low frequency wavelet After the coefficient is subjected to recursive noise reduction, the k-th low-frequency wavelet coefficient is restored in detail according to the following formula;
  • the edge strength of the pixel corresponding to the jth high frequency wavelet coefficient comprises: a jth height The edge intensity corresponding to at least one of the three components y, u, and v corresponding to the pixel corresponding to the frequency wavelet coefficient.
  • the edge strength corresponding to the at least one component is based on the high frequency wavelet coefficient and the low frequency wavelet coefficient of the at least one component Edge strength.
  • the attenuation function is a wavelet threshold function, including at least one of the following: a hard threshold function, a soft threshold function .
  • the at least one direction comprises at least one of: from left to right, from right to left, from top to bottom, From the bottom up.
  • the method and terminal for image noise reduction provided by the embodiment of the present invention, by performing wavelet decomposition on at least one of three components y, u, and v of the image data, for each component of the at least one component
  • the low-frequency wavelet coefficients are subjected to recursive noise reduction to obtain low-frequency wavelet coefficients after each component is denoised; wavelet reconstruction is performed according to the high-frequency wavelet coefficients of each component and the low-frequency wavelet coefficients of each component after noise reduction Obtaining at least one component after noise reduction; when the at least one component after the noise reduction is three components, combining the three components after the noise reduction to obtain image data after noise reduction; When at least one component of the noise is one or two components, the at least one component after the noise reduction is combined with other components of the three components to obtain image data after noise reduction.
  • the low-frequency wavelet coefficients are recursively denoised, which reduces the amount of data calculated, the computational complexity is low, and the efficiency of image denoising is improved.
  • the pixels in the image are utilized by recursive denoising. The relationship guarantees the effect of image noise reduction.
  • FIG. 1 is a flow chart of a method for image noise reduction according to an embodiment of the present invention
  • FIG. 2 is a flow chart of another method for image noise reduction according to an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of performing 1-layer wavelet decomposition on an image in a method for image noise reduction according to an embodiment of the present invention
  • FIG. 4 is a schematic diagram of performing 2-layer wavelet decomposition on an image in a method for image noise reduction according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram of performing 2-layer wavelet decomposition on an image in a method for image noise reduction according to an embodiment of the present invention
  • FIG. 6 is a schematic diagram of a noise reduction intensity function curve and a detail retention intensity function curve in a method for image noise reduction according to an embodiment of the present invention
  • FIG. 7 is a schematic diagram of a noise reduction effect of another method for image noise reduction according to an embodiment of the present invention.
  • FIG. 8 is a schematic structural diagram of a terminal for image noise reduction according to an embodiment of the present invention.
  • FIG. 9 is another schematic structural diagram of a terminal for image noise reduction according to an embodiment of the present invention.
  • the method may include the following steps:
  • Step 102 performing wavelet decomposition on at least one of the luminance component y and the chrominance components u and v of the image data to obtain a high frequency wavelet coefficient and a low frequency wavelet coefficient of each component;
  • the at least one component after the noise reduction is one or two components, combining the at least one component after the noise reduction with other components of the three components to obtain image data after noise reduction.
  • the three components y, u, and v after the noise reduction are combined.
  • the image data after noise reduction if the at least one component after the noise reduction is a component, such as the y component, the y component after the noise reduction and the other two of the three components
  • the components, that is, the u and v components are combined to obtain the image data after the noise reduction; if the at least one component after the noise reduction is two components, such as y and u components, the y and u components after the noise reduction are obtained.
  • the image data of the noise reduction is obtained by combining with other components of the three components, that is, the v component.
  • the method may further include:
  • the performing attenuation function based on the edge information based on the edge information of the high frequency wavelet coefficients of each component includes:
  • y ⁇ x + (1 - ⁇ ) h (x), where ⁇ is a parameter related to edge strength and h (x) is an attenuation function with respect to x.
  • the step 104 is specifically 104′: performing wavelet reconstruction according to the high-frequency wavelet coefficients after each component noise reduction and the low-frequency wavelet coefficients after each component noise reduction, and obtaining noise reduction.
  • the at least one component is specifically 104′: performing wavelet reconstruction according to the high-frequency wavelet coefficients after each component noise reduction and the low-frequency wavelet coefficients after each component noise reduction, and obtaining noise reduction.
  • the at least one component is specifically 104′: performing wavelet reconstruction according to the high-frequency wavelet coefficients after each component noise reduction and the low-frequency wavelet coefficients after each component noise reduction, and obtaining noise reduction.
  • the step 1031 can be performed before or after the step 103 or at the same time, which is not limited by the present invention.
  • the edge intensity of the edge information of the image is referenced, and the degree of noise reduction is also different according to the edge intensity, so that the noise of the high-frequency wavelet coefficients can be reduced.
  • more details such as edge information of the image are retained.
  • Different methods are used to reduce the noise of high-frequency wavelet coefficients and low-frequency wavelet coefficients.
  • the characteristics of the information contained in the high-frequency wavelet coefficients and the low-frequency wavelet coefficients are considered.
  • the high-frequency part generally contains edge information, and the high-frequency wavelet coefficients are guaranteed.
  • the noise reduction effect of the low frequency wavelet coefficients the noise reduction effect of the whole image is improved, and the quality of the whole image is guaranteed.
  • the step 102 may include: a luminance component y, a chrominance component u and a v of the image data. At least one of the components is subjected to n-layer wavelet decomposition to obtain n-layer high-frequency wavelet coefficients and n-layer low-frequency wavelet coefficients of each component, where n ⁇ 2, and n is an integer.
  • Each image or photo may have three components of y, u, and v, and the noise component y component is denoised, and the luminance noise existing in the image may be mainly removed; and the chrominance components u and v components are denoised.
  • the color noise present in the image can be mainly removed. Which components are specifically denoised can be selected according to the type of noise present in the image.
  • One or more wavelet decompositions may be performed on at least one of the three components y, u, and v of the image data to obtain n-layer high-frequency wavelet coefficients of each component (wherein each layer may have three high-frequency wavelets)
  • the coefficients are HL, LH, and HH) and the n-layer low-frequency wavelet coefficients (where each layer can have a low-frequency wavelet coefficient, which is LL).
  • the low frequency wavelet coefficients of the first layer can be denoted as LL1
  • the high frequency wavelet coefficients of the first layer can be respectively recorded as: HL1, LH1 and HH1.
  • Other layers can be deduced by analogy.
  • the steps 103 and 104 may include: processing each component as follows:
  • A recursively denoising the low-frequency wavelet coefficients of the nth layer to obtain the low-frequency wavelet coefficients of the n-th layer after noise reduction, according to the low-frequency wavelet coefficients of the n-th layer noise reduction and the high-frequency wavelet coefficients of the n-th layer Wavelet reconstruction, obtaining the low frequency wavelet coefficients after the n-1th layer of noise reduction;
  • wavelet reconstruction is performed according to the low frequency wavelet coefficient and the i-th high frequency wavelet coefficient after the second noise reduction of the i-th layer, and a component after noise reduction is obtained.
  • the image data of the image with noise is decomposed by wavelet, and the low-frequency wavelet coefficients of the highest layer of the wavelet decomposition to the layers of the first layer are recursively denominated layer by layer, which can eliminate large-area flakes. Noise, and the computational complexity is small.
  • the data volume of the low-frequency wavelet coefficients of each layer of wavelet decomposition is 1/4 of the total data volume of the layer, which reduces the amount of data to be processed and improves the efficiency.
  • the wavelet decomposition speed is faster, which further improves the efficiency of image noise reduction.
  • the window size must be expanded to more than twice the noise size, and as the window size increases, the computational complexity increases rapidly.
  • This scheme adopts recursive noise reduction, which has low computational complexity and is not affected by the size of the window. Like the “bulldozer”, the noise is gradually eroded step by step.
  • the recursive noise reduction utilizes the relationship between the low frequency wavelet coefficients (or between pixel points of the image) to perform recursive noise reduction, and the effect is equivalent to using the information of the entire image for noise reduction and improving noise reduction. The effect guarantees the quality of the image.
  • the method may further include:
  • the performing attenuation function based on the edge information based on the edge information of the high frequency wavelet coefficients of each component includes:
  • y j ⁇ j x j +(1- ⁇ j )h(x j ), j ⁇ 1, j is an integer;
  • y j is the value of the jth high frequency wavelet coefficient after noise reduction
  • x j is the value of the jth high frequency wavelet coefficient
  • h(x j ) is the attenuation function for x j
  • ⁇ j is the jth
  • ⁇ j may be a value obtained by normalizing the edge intensity of the pixel point corresponding to the j-th high-frequency wavelet coefficient.
  • the value of j is related to the number of high frequency wavelet coefficients of the corresponding layer.
  • the step 1031' may be performed before or after the step 103 or at the same time, which is not limited by the present invention.
  • the edge intensity of the edge information of the image is referenced, and the degree of noise reduction is also different according to the edge intensity, so that the noise of the high-frequency wavelet coefficients can be reduced.
  • more details such as edge information of the image are retained.
  • Different methods are used to reduce the noise of high-frequency wavelet coefficients and low-frequency wavelet coefficients.
  • the characteristics of the information contained in the high-frequency wavelet coefficients and the low-frequency wavelet coefficients are considered.
  • the high-frequency part generally contains edge information, and the high-frequency wavelet coefficients are guaranteed.
  • the noise reduction effect of the low frequency wavelet coefficients the noise reduction effect of the whole image is improved, and the quality of the whole image is guaranteed.
  • the low frequency wavelet coefficient and the nth layer are lower according to the nth layer after noise reduction in step A.
  • the wavelet coefficients are wavelet reconstructed to obtain the low-frequency wavelet coefficients after the n-1th layer is denoised, specifically:
  • Wavelet reconstruction is performed according to the low-frequency wavelet coefficients after the n-th noise reduction and the high-frequency wavelet coefficients after the n-th noise reduction, and the low-frequency wavelet coefficients of the n-1th layer after noise reduction are obtained.
  • the wavelet reconstruction according to the low-frequency wavelet coefficient and the ith-layer high-frequency wavelet coefficient after the second-stage noise reduction in the ith layer is as follows:
  • Wavelet reconstruction is performed according to the low frequency wavelet coefficient after the second noise reduction of the i-th layer and the high frequency wavelet coefficient after the i-th layer noise reduction.
  • the recursive noise reduction may include:
  • the noise reduction result of the kth low frequency wavelet coefficient the value of the kth low frequency wavelet coefficient + f (the noise reduction result of the k-1th low frequency wavelet coefficient - the value of the kth low frequency wavelet coefficient)
  • k>1 k is an integer
  • x represents the difference between the noise reduction result of the k-1th low frequency wavelet coefficient and the kth low frequency wavelet coefficient
  • y represents the drop Noise intensity.
  • the noise intensity that is, the different noise reduction levels, may also be different noise reduction levels preset by the system or set by the user, the noise reduction intensity function being based on the noise estimation level of the image and/or the user's function.
  • the settings to determine the function may also be different noise reduction levels preset by the system or set by the user, the noise reduction intensity function being based on the noise estimation level of the image and/or the user's function.
  • the recursive noise reduction can be understood as that, for each low-frequency wavelet coefficient, if there is a low-frequency wavelet coefficient in at least one direction, the low-frequency wavelet coefficient is calculated according to the value of the previous low-frequency wavelet coefficient after noise reduction. The value after noise.
  • the noise reduction result of the first low-frequency wavelet coefficient may be the value of the low-frequency wavelet coefficient, or a noise reduction algorithm may be used.
  • the value of the low-frequency wavelet coefficient is obtained by performing noise reduction.
  • the noise reduction algorithm may adopt an existing noise reduction algorithm, and the comparison of the present invention is not limited.
  • the method may further include:
  • the detail is used to recover the detail, and the low-frequency wavelet coefficients after each layer of detail recovery are obtained.
  • the low-frequency wavelet coefficients after noise reduction can be adjusted. At the same time retain more image details.
  • the edge strength of the pixel corresponding to the jth high frequency wavelet coefficient includes: jth The edge intensity corresponding to at least one of the three components y, u, and v corresponding to the pixel corresponding to the high-frequency wavelet coefficient.
  • the edge intensity of the pixel corresponding to the j-th high-frequency wavelet coefficient may include the color corresponding to the chrominance components u and v.
  • the edge intensity (the edge intensity corresponding to the chrominance component), in turn, includes the intensity of the edge of the y component (the edge intensity corresponding to the luminance component), that is, the edge corresponding to the three components y, u, and v The sum of the strengths.
  • the maximum value of the chrominance and the brightness edge intensity at the pixel corresponding to the j-th high-frequency wavelet coefficient may be taken.
  • the edge intensity at the pixel point corresponding to the j-th high-frequency wavelet coefficient in this case, the edge intensity of one of the three components of the corresponding y, u, and v components.
  • an edge strength corresponding to one or more components of the three components of y, u, and v may be selected as the edge intensity of the pixel corresponding to the j-th high-frequency wavelet coefficient, for example, y may be selected.
  • Two of the three components of u and v are larger in intensity, and the sum of the edge intensities of the two is used as the edge intensity of the pixel corresponding to the j-th high-frequency wavelet coefficient.
  • an edge intensity corresponding to the at least one component is an edge intensity based on a high frequency wavelet coefficient and a low frequency wavelet coefficient of the at least one component.
  • the edge intensity can be extracted for both the high-frequency wavelet coefficient and the low-frequency wavelet coefficient (the value of the edge intensity can be adopted as a classic Sobel operator, Laplace operator, etc.), and then take the high-frequency wavelet coefficient and the low-frequency wavelet coefficient edge intensity (including the high-frequency wavelet coefficients HLj, LHj, HHj and the low-frequency wavelet coefficient LLj corresponding edge strength)
  • the maximum value is used as the brightness edge intensity; for the calculation of the chrominance edge intensity, the edge intensity can be extracted from both the high-frequency wavelet coefficient and the low-frequency wavelet coefficient, and then the maximum value of the high-frequency wavelet coefficient and the low-frequency wavelet coefficient edge intensity is taken as the color.
  • Degree edge strength in actual operation, because the high frequency wavelet coefficient of chromaticity is generally weak, in order to save the calculation amount, only the edge intensity of the low frequency
  • the attenuation function in the foregoing embodiment is specifically a wavelet threshold function, and may include at least one of the following: a hard threshold function and a soft threshold function.
  • the attenuation function may be a hard threshold or a soft threshold function, or an attenuation function combined with a hard threshold and a soft threshold, and the like.
  • the at least one direction in the foregoing embodiment includes at least one of the following: from left to right, from right to left, from top to bottom, and from bottom to top.
  • the refraction noise reduction in the above four directions can be performed on the low frequency wavelet coefficients of each layer, and the low frequency wavelet coefficients after the noise reduction of the layer are obtained.
  • the symmetry of the noise reduction effect can be ensured, and the noise reduction effect of the image can be improved while ensuring the image quality.
  • the recursive noise reduction of each low-frequency wavelet coefficient uses only the information of the four low-frequency wavelet coefficients of the upper, lower, left and right in four directions, but the recursive drop of each low-frequency wavelet coefficient Noise is equivalent to indirectly utilizing the information of the entire image, which is equivalent to increasing the size of the window filtering. Therefore, the large-area noise can be well eliminated, and the computational complexity is low, which ensures the noise reduction effect and improves the noise reduction efficiency.
  • n 3
  • color noise reduction can only involve processing of u and v components, and the processing of u and v components can be the same or similar
  • the following processing of u component can refer to the following process.
  • Step 1 Perform 3-layer wavelet decomposition on the chrominance component u component of the image to obtain the first layer low-frequency wavelet coefficient LL1 and the first-layer high-frequency wavelet coefficients HL1, LH1, HH1, see FIG. 3; obtain the second layer low-frequency wavelet The coefficient LL2 and the second layer high frequency wavelet coefficients HL2, LH2, HH2, see Fig. 4; the third layer low frequency wavelet coefficient LL3 and the second layer high frequency wavelet coefficients HL3, LH3, HH3 are obtained, see Fig. 5.
  • Step 2 Perform recursive noise reduction in four directions on the lowest layer (the third layer in this embodiment) of the wavelet decomposition LL3;
  • the first low frequency wavelet coefficient for the line is the first low frequency wavelet coefficient for the line:
  • the noise reduction result of the first low frequency wavelet coefficient the value of the low frequency wavelet coefficient to the other low frequency wavelet coefficients of the line:
  • Reference point value noise reduction result of the left low frequency wavelet coefficient
  • the noise reduction result of the low frequency wavelet coefficient the value of the low frequency wavelet coefficient + f (the value of the reference point - the value of the low frequency wavelet coefficient)
  • the noise reduction result of the first low frequency wavelet coefficient is the value of the low frequency wavelet coefficient
  • the noise reduction result of the kth low frequency wavelet coefficient the value of the kth low frequency wavelet coefficient +f (k-th
  • k > 1 k is an integer
  • x represents the k-1th low frequency
  • y represents the noise reduction strength
  • the noise reduction intensity function is a function determined according to the noise estimation level of the image and/or the setting of the function by the user. .
  • the noise reduction intensity of the image may be obtained according to a predetermined rule according to the noise estimation level of the image, that is, different noise reduction levels, or may be different noise reduction levels preset by the system or set by the user, and the peak point is
  • the position (the position at which the noise reduction intensity is the largest) may be obtained according to the noise estimation level of the image and/or the user's setting of the function, and the amplitude of the peak point may be freely set by the user or preset by the device.
  • x is closer to both ends (indicating that the edge is stronger), and y is closer to 0, so that the noise reduction result of the low frequency wavelet coefficient is closer to the original value of the low frequency wavelet coefficient.
  • LUT Look Up Table
  • the above 1) to 4) recursively denoise the low-frequency wavelet coefficients LL3 of the third-layer wavelet decomposition of the image from four different directions, and obtain the low-frequency wavelet coefficients of the third layer after noise reduction.
  • the order of the above-mentioned four directions at the time of execution (that is, the order of the first to fourth passes at the time of execution) can be adjusted as needed, and the present invention is not limited thereto, and only one of the embodiments of the present invention is exemplified. Kind of situation.
  • the embodiment of the present invention exemplifies recursive noise reduction in four directions, and the direction may be different according to requirements, for example, recursive noise reduction can be performed only in two directions (such as two directions of symmetry: left to right and From right to left), there are more directions, such as diagonally diagonally forming four directions: left Down to the top right, top right to bottom left, top left to bottom right, bottom right to top left.
  • the noise reduction filtering for each low-frequency wavelet coefficient uses only the information of the four low-frequency wavelet coefficients of the upper, lower, left and right, but actually Inter-grounding utilizes the information of the entire image, which is equivalent to increasing the size of the window filtering, so that the chip-like color noise of the large-face junction existing in the image can be well eliminated, and the computational complexity is low.
  • the low-frequency wavelet coefficient LL3 of the layer 3 wavelet decomposition of the image after performing the above recursive noise reduction on the low-frequency wavelet coefficient LL3 of the layer 3 wavelet decomposition of the image, reference may also be made to the original low-frequency wavelet coefficient and the detail-preserving intensity function of the layer 3 of the image, and the obtained The low-frequency wavelet coefficients of the third layer after noise reduction are restored in detail as follows:
  • the position of the peak point (the position where the detail remains strong) may be obtained according to the noise estimation level of the image and/or the setting of the function by the user, and the amplitude of the peak point may be a user free setting or a device preset. of.
  • the noise estimation level of the image may be obtained according to the noise estimation level of the image and/or the setting of the function by the user
  • the amplitude of the peak point may be a user free setting or a device preset. of.
  • the noise level of the image since the position where the noise is the largest corresponds to the peak of the noise reduction intensity, more details of the image may be lost, so that the peak point of the detail retention intensity function curve may be determined according to the position where the noise is the largest.
  • Position (ie the x value corresponding to the peak point),
  • y g(x)
  • the amplitude of the peak point of the function that is, the maximum detail retention strength, which can be set by the user freely or preset by the device.
  • Step 3 performing noise reduction on the third layer high-frequency wavelet coefficients HL3, LH3, and HH3 based on the attenuation function of the edge information;
  • the edge information based attenuation function noise reduction includes noise reduction of the high frequency wavelet coefficients of the layer according to the following formula:
  • y j ⁇ j x j +(1- ⁇ j )h(x j ), j ⁇ 1, j is an integer;
  • ⁇ j is the value of the jth high frequency wavelet coefficient after noise reduction
  • x j is the value of the jth high frequency wavelet coefficient
  • h(x j ) is the attenuation function for x j
  • ⁇ j is the jth
  • ⁇ j may be a value obtained by normalizing the edge intensity of the pixel point corresponding to the j-th high-frequency wavelet coefficient.
  • the edge strength of the pixel corresponding to the j-th high-frequency wavelet coefficient may include: at least one of the three components y, u, and v corresponding to the pixel corresponding to the j-th high-frequency wavelet coefficient The edge strength corresponding to the amount.
  • the edge information of the chrominance components u, v may be unstable or not obvious.
  • the edge intensity of the pixel corresponding to the j-th high-frequency wavelet coefficient It may include both the chrominance edge intensity corresponding to the chrominance components u, v (the edge intensity corresponding to the chrominance component) and the luminance edge intensity corresponding to the y component (the edge intensity corresponding to the luminance component), ie It can be the sum of the edge intensities corresponding to the three components y, u, and v.
  • the maximum value of the chrominance and the brightness edge intensity at the pixel corresponding to the j-th high-frequency wavelet coefficient (ie, the maximum value of the edge intensity corresponding to the three components y, u, and v) may be taken. ), as the edge intensity at the pixel point corresponding to the j-th high-frequency wavelet coefficient, in this case, the edge intensity of one of the three components of the corresponding y, u, and v components.
  • an edge strength corresponding to one or more components of the three components of y, u, and v may be selected as the edge intensity of the pixel corresponding to the j-th high-frequency wavelet coefficient, for example, y may be selected.
  • Two of the three components of u and v are larger in intensity, and the sum of the edge intensities of the two is used as the edge intensity of the pixel corresponding to the j-th high-frequency wavelet coefficient.
  • the edge information of the luminance and the chrominance is referred to. Therefore, while filtering the high-frequency noise, the detailed information of the image can still be well preserved.
  • the edge strength corresponding to the at least one component may be an edge strength based on the high frequency wavelet coefficient and the low frequency wavelet coefficient of the at least one component.
  • both the high-frequency wavelet coefficient and the low-frequency wavelet coefficient contain edge information, and the edge intensity can be extracted for both the high-frequency wavelet coefficient and the low-frequency wavelet coefficient, and the component is obtained based on the high-frequency wavelet coefficient and the edge strength of the low-frequency wavelet coefficient. Corresponding edge strength.
  • the edge intensity can be extracted for both the high-frequency wavelet coefficient and the low-frequency wavelet coefficient (for example, the value of the edge intensity can be adopted by the classic Sobel. Operator, Laplace operator, etc. for edge detection), and then take the edge intensity of the high frequency wavelet coefficient and the low frequency wavelet coefficient (including the edge intensity corresponding to the high frequency wavelet coefficients HLm, LHm, HHm and low frequency wavelet coefficient LLm, m
  • the maximum value of m for the third layer is 3) as the luminance edge strength.
  • chroma edge strength edge intensity corresponding to u or v
  • the calculation can extract the edge intensity from both the high-frequency wavelet coefficient and the low-frequency wavelet coefficient, and then take the maximum value of the high-frequency wavelet coefficient and the edge intensity of the low-frequency wavelet coefficient as the chroma edge intensity, and the high-frequency wavelet coefficient of the chroma component is generally Very weak, it is also possible to extract only the edge intensity of the low frequency wavelet coefficients to save computation.
  • the edge intensity corresponding to at least one component is referenced, and when the edge intensity of a component is calculated, the high-frequency wavelet coefficient and the low-frequency wavelet coefficient are referenced, thereby filtering In addition to high frequency noise, more details of the image can be preserved.
  • the attenuation function is specifically a wavelet threshold function, and may include at least one of the following: a hard threshold function and a soft threshold function.
  • the attenuation function may be a hard threshold or a soft threshold function, or an attenuation function combined with a hard threshold and a soft threshold, and the like.
  • the soft threshold function is taken as an example for processing.
  • the following formula is one of the soft threshold functions:
  • the T is a threshold value, and the method for determining the threshold value may adopt various existing methods, which is not limited by the embodiment of the present invention.
  • the soft threshold function is used for noise reduction. While filtering the high-frequency noise, the detailed information of the image can still be well preserved.
  • Step 4 performing wavelet reconstruction on the third layer low frequency wavelet coefficient LL3 processed in step two and the third layer high frequency wavelet coefficients HL3, LH3, and HH3 processed in step three, to obtain the second layer after noise reduction (ie, N-1 layer) low frequency wavelet coefficient LL2;
  • Step 5 performing the operation similar to step 2 on the second layer (ie, the i-th layer, i initial value is n-1) of the second layer after noise reduction in step 4, and obtaining the second noise reduction
  • Two-layer low-frequency wavelet coefficient LL2 (Note: Since LL3 and LL2 represent information of different scales (hierarchies) and different frequencies, so that noise can be separately degraded for different scales and different frequencies); for the second layer of high-frequency wavelet coefficients HL2 , LH2, HH2 perform similar operations as step three, and obtain the second layer of high frequency wavelet system after noise reduction. Number HL2, LH2, HH2;
  • the value of i is 2, that is, i>1, then the wavelet reconstruction of LL2, HL2, LH2, and HH2 after noise reduction in step 5 is performed to obtain the low frequency of the first layer (ie, the i-1th layer) after noise reduction.
  • Step 6 performing the operation similar to the second step on the first layer low frequency wavelet coefficient LL1 after the noise reduction in step 5, to obtain the first layer low frequency wavelet coefficient LL1 after the second noise reduction; and the first layer high frequency wavelet
  • the coefficients HL1, LH1, HH1 perform similar operations as in step 3, and obtain the first layer of high frequency wavelet coefficients HL1, LH1, HH1 after noise reduction;
  • a component after noise reduction can be obtained, that is, wavelet reconstruction is performed on LL1, HL1, LH1, and HH1 after noise reduction in step 6, and the chrominance component u after noise reduction is obtained.
  • Step 1 For the v component, refer to Step 1 to Step 6 above to process the chrominance component v after noise reduction.
  • the luminance component y may be subjected to noise reduction processing by referring to steps 1 to 6 above.
  • luminance noise is removed.
  • an image obtained by removing luminance noise and color noise can be obtained, as shown in FIG. 7(c), it can be seen that the image is removed from noise. At the same time, the details are also very good.
  • the method may be the same or different, and may be adjusted as needed, which is not limited by the present invention.
  • the method for image noise reduction provided by the foregoing embodiments can be applied to the image denoising process when the terminal photographs, and the quality of the photographed image and the user experience can be improved.
  • the embodiment of the present invention further provides a terminal for image noise reduction, and as shown in FIG. 8 is an embodiment of a terminal provided by the present invention.
  • the terminal includes:
  • An image obtaining unit 800 configured to acquire image data of an image
  • the image decomposition unit 810 is configured to perform wavelet decomposition on at least one of the three components y, u, and v of the image data to obtain a high frequency wavelet coefficient and a low frequency wavelet coefficient of each component, where y is an image.
  • Luminance, u, v are the chromaticity of the image;
  • the image noise reduction processing unit 820 is configured to perform recursive noise reduction on the low frequency wavelet coefficients of each component to obtain low frequency wavelet coefficients of each component after noise reduction, according to high frequency wavelet coefficients of each component Performing wavelet reconstruction on the low-frequency wavelet coefficients of each component after noise reduction to obtain at least one component after noise reduction;
  • the noise reduction image acquisition unit 830 is configured to: when the at least one component after the noise reduction is three components, combine the three components after the noise reduction to obtain image data after noise reduction; when the noise reduction is performed When at least one component is one or two components, the noise-reduced at least one component is combined with other components of the three components to obtain noise-reduced image data.
  • the terminal further includes an image high-frequency processing unit 840, configured to perform noise reduction on the high-frequency wavelet coefficients of each component based on edge information, including a formula for denoising the high frequency wavelet coefficients of each of the components,
  • y ⁇ x+(1 ⁇ )h(x), where ⁇ is a parameter related to edge strength, and h(x) is an attenuation function with respect to x;
  • the image noise reduction processing unit 820 is configured to perform wavelet reconstruction according to the high frequency wavelet coefficients of each component and the low frequency wavelet coefficients of each of the component noise reductions, to obtain at least one component after noise reduction, specifically, And performing wavelet reconstruction on the low frequency wavelet coefficients after the noise reduction of each component and the low frequency wavelet coefficients after each component noise reduction, to obtain the at least one component after noise reduction.
  • the image decomposition unit 810 is specifically configured to perform n-layer wavelet decomposition on at least one of the luminance component y and the chrominance components u and v of the image data. Obtaining n layers of high frequency wavelet coefficients and n layers of low frequency wavelet coefficients for each component, where n ⁇ 2, n is an integer;
  • the image noise reduction processing unit 820 is specifically configured to perform the following processing on each component:
  • A recursively denoising the low-frequency wavelet coefficients of the nth layer to obtain the low-frequency wavelet coefficients of the n-th layer after noise reduction, according to the low-frequency wavelet coefficients of the n-th layer noise reduction and the high-frequency wavelet coefficients of the n-th layer Wavelet reconstruction, obtaining the low frequency wavelet coefficients after the n-1th layer of noise reduction;
  • wavelet reconstruction is performed according to the low frequency wavelet coefficient and the i-th high frequency wavelet coefficient after the second noise reduction of the i-th layer, and a component after noise reduction is obtained.
  • the terminal further includes an image high-frequency processing unit 840, configured to perform noise reduction on the high-frequency wavelet coefficients of each component based on edge information, including according to the following formula Denoising each layer of high frequency wavelet coefficients of the n layers of each component,
  • y j ⁇ j x j +(1- ⁇ j )h(x j ), j ⁇ 1, j is an integer;
  • y j is the value of the jth high frequency wavelet coefficient after noise reduction
  • x j is the value of the jth high frequency wavelet coefficient
  • h(x j ) is the attenuation function for x j
  • ⁇ j is the jth The edge intensity coefficient corresponding to the edge intensity of the pixel corresponding to the high-frequency wavelet coefficient, 0 ⁇ ⁇ j ⁇ 1.
  • the image noise reduction processing unit 820 is configured to perform wavelet reconstruction according to the low-frequency wavelet coefficients after the n-th noise reduction and the high-frequency wavelet coefficients of the n-th layer.
  • the low-frequency wavelet coefficient obtained after the n-1th layer of noise reduction is specifically configured to perform wavelet reconstruction according to the low-frequency wavelet coefficient after the n-th layer noise reduction and the high-frequency wavelet coefficient after the n-th noise reduction, to obtain the first Low frequency wavelet coefficient after noise reduction of n-1 layer;
  • the image noise reduction processing unit 820 is configured to perform wavelet reconstruction according to the low frequency wavelet coefficient and the ith layer high frequency wavelet coefficient of the ith layer secondary noise reduction, and is specifically configured to be used according to the ith layer
  • the low-frequency wavelet coefficients after noise reduction and the high-frequency wavelet coefficients after the i-th layer noise reduction are wavelet reconstructed.
  • the recursive noise reduction includes:
  • the noise reduction result of the kth low frequency wavelet coefficient the value of the kth low frequency wavelet coefficient + f (the noise reduction result of the k-1th low frequency wavelet coefficient - the value of the kth low frequency wavelet coefficient)
  • k>1 k is an integer
  • x represents the difference between the noise reduction result of the k-1th low frequency wavelet coefficient and the kth low frequency wavelet coefficient
  • y represents the drop Noise intensity.
  • the terminal further includes a detail restoring unit 821, configured to perform recursive noise reduction on the kth low frequency wavelet coefficient after the image noise reduction processing unit 820, according to the following formula Performing detail recovery on the kth low frequency wavelet coefficient;
  • the edge intensity of the pixel corresponding to the j-th high-frequency wavelet coefficient includes: y, u, v corresponding to the pixel corresponding to the j-th high-frequency wavelet coefficient The edge intensity corresponding to at least one of the three components.
  • an edge strength corresponding to the at least one component is an edge strength based on a high frequency wavelet coefficient and a low frequency wavelet coefficient of the at least one component.
  • the attenuation function is a wavelet threshold function, including at least one of the following: a hard threshold function, a soft threshold function.
  • the at least one direction includes the following One less: from left to right, from right to left, from top to bottom, from bottom to top.
  • FIG. 9 is an embodiment of a terminal provided by the present invention.
  • the terminal includes a memory, a processor, and a communication bus.
  • the processor is coupled to the memory via the communication bus.
  • the terminal may further include a communication interface, which is communicably connected to other devices (such as other terminals or access point devices, etc.) through the communication interface.
  • the memory may be one or more for storing image data acquired by the terminal, and storing instructions implementing a method for image noise reduction; wherein the image data and the instructions may be stored in the same memory Medium or separately stored in different memories;
  • the processor may be one or more, when the one or more processors retrieve image data stored in the one or more memories and implement instructions for a method for image noise reduction,
  • the image data performs the following steps:
  • the at least one component after the noise reduction is three components, combining the three components after the noise reduction to obtain image data after noise reduction;
  • the at least one component after the noise reduction is one or two components, combining the at least one component after the noise reduction with other components of the three components to obtain image data after noise reduction.
  • a processor may retrieve the image data and the instruction, The image data and the instructions can also be retrieved separately by different processors.
  • the processor when the processor retrieves image data stored in the one or more memories and an instruction to implement a method for image noise reduction, the image may also be The data performs the following steps: performing noise reduction on the high frequency wavelet coefficients of each component based on the attenuation function of the edge information;
  • the performing attenuation function based on the edge information based on the edge information of the high frequency wavelet coefficients of each component includes:
  • y ⁇ x+(1 ⁇ )h(x), where ⁇ is a parameter related to edge strength, and h(x) is an attenuation function with respect to x;
  • the processor performs wavelet reconstruction according to the high frequency wavelet coefficients of each component and the low frequency wavelet coefficients of each component after noise reduction, to obtain at least one component after noise reduction, specifically:
  • Wavelet reconstruction is performed according to the high frequency wavelet coefficients after each component noise reduction and the low frequency wavelet coefficients after each component noise reduction, to obtain the at least one component after noise reduction.
  • the processor performs wavelet decomposition on at least one of the luminance component y and the chrominance components u and v of the image data to obtain a high of each component.
  • Frequency wavelet coefficients and low frequency wavelet coefficients including:
  • n-layer wavelet decomposition on at least one of the luminance component y and the chrominance components u and v of the image data to obtain n-layer high-frequency wavelet coefficients and n-layer low-frequency wavelet coefficients of each component, where n ⁇ 2, n is an integer;
  • At least one component obtained after the noise reduction includes:
  • A Recursively denoising the low-frequency wavelet coefficients of the nth layer to obtain the low frequency of the nth layer after noise reduction
  • the wave coefficient is wavelet reconstructed according to the low-frequency wavelet coefficient of the n-th layer noise reduction and the high-frequency wavelet coefficient of the n-th layer, and the low-frequency wavelet coefficient after the n-1th layer noise reduction is obtained;
  • wavelet reconstruction is performed according to the low frequency wavelet coefficient and the i-th high frequency wavelet coefficient after the second noise reduction of the i-th layer, and a component after noise reduction is obtained.
  • the processor performs edge function-based attenuation function noise reduction on the high frequency wavelet coefficients of each component, including:
  • y j ⁇ j x j +(1- ⁇ j )h(x j ), j ⁇ 1, j is an integer;
  • y j is the value of the jth high frequency wavelet coefficient after noise reduction
  • x j is the value of the jth high frequency wavelet coefficient
  • h(x j ) is the attenuation function for x j
  • ⁇ j is the jth The edge intensity coefficient corresponding to the edge intensity of the pixel corresponding to the high-frequency wavelet coefficient, 0 ⁇ ⁇ j ⁇ 1.
  • the processor performs wavelet reconstruction according to the low-frequency wavelet coefficients after the n-th noise reduction and the high-frequency wavelet coefficients of the n-th layer to obtain an n-1th layer drop.
  • the low frequency wavelet coefficient after noise is specifically as follows:
  • the processor performs wavelet reconstruction according to the low frequency wavelet coefficient and the ith layer high frequency wavelet coefficient after the second noise reduction of the i-th layer:
  • the recursive noise reduction includes:
  • the noise reduction result of the kth low frequency wavelet coefficient the value of the kth low frequency wavelet coefficient + f (the noise reduction result of the k-1th low frequency wavelet coefficient - the value of the kth low frequency wavelet coefficient)
  • k>1 k is an integer
  • x represents the difference between the noise reduction result of the k-1th low frequency wavelet coefficient and the kth low frequency wavelet coefficient
  • y represents the drop Noise intensity.
  • the processor may further recover the kth low frequency wavelet coefficient according to the following formula. ;
  • the edge intensity of the pixel corresponding to the j-th high-frequency wavelet coefficient includes: y, u corresponding to the pixel corresponding to the j-th high-frequency wavelet coefficient, v Edge strength corresponding to at least one of the three components.
  • an edge strength corresponding to the at least one component is an edge strength based on a high frequency wavelet coefficient and a low frequency wavelet coefficient of the at least one component.
  • the attenuation function is a wavelet threshold function, including at least one of the following: a hard threshold function, a soft threshold function.
  • the at least one direction comprises at least one of: left to right, right to left, top to bottom, and bottom to top.
  • the processor retrieves the image data stored in the one or more memories and the instructions for implementing the method for image noise reduction, the steps that can be performed on the image data and For the specific content of each step, reference may be made to related parts in the foregoing method embodiments, and details are not described herein again.
  • the disclosed system, apparatus, and method may be implemented in other manners.
  • the device embodiments described above are merely illustrative.
  • the division of the modules or units is only a logical function division.
  • there may be another division manner for example, multiple units or components may be used. Combinations can be integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, and may be in an electrical, mechanical or other form.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)
  • Picture Signal Circuits (AREA)

Abstract

本发明实施例公开了一种用于图像降噪的方法及终端,其中方法包括:获取图像的图像数据;对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数;对所述每个分量的低频小波系数进行至少一个方向的递归降噪,得到所述每个分量降噪后的低频小波系数;根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量;当所述降噪后的至少一个分量为三个分量时,将所述降噪后的三个分量组合,得到降噪后的图像数据;当所述降噪后的至少一个分量为一个或两个分量时,将所述降噪后的至少一个分量与所述三个分量中的其他分量组合,得到降噪后的图像数据。

Description

一种用于图像降噪的方法及终端
本申请要求于2013年11月8日提交中国专利局,申请号为201310554182.2、发明名称为“一种用于图像降噪的方法及终端”的中国专利申请,其全部内容通过引用结合在本申请中。
技术领域
本发明涉及图像处理领域,尤指一种用于图像降噪的方法及终端。
背景技术
为了解决图像噪声问题,业界出现了很多图像降噪算法,如:与图像内容无关的窗口滤波法,基于图像结构相似性分析的Non-local Means(非局部均值)算法等。
与图像内容无关的窗口滤波法,计算量小,但图像的细节损失严重;基于图像结构相似性分析的Non-local Means算法在细节保持与色彩保护方面均表现很好,但算法复杂度高,效率低。降噪效果和效率的矛盾就更加突出。
发明内容
本发明实施例提供一种图像降噪的方法及终端,可以提高对图像进行降噪的降噪效果和效率。
第一方面,本发明实施例提供了一种图像降噪的方法,所述方法包括:
获取图像的图像数据;
对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数;
对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数;
根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量;
当所述降噪后的至少一个分量为三个分量时,将所述降噪后的三个分量 组合,得到降噪后的图像数据;
当所述降噪后的至少一个分量为一个或两个分量时,将所述降噪后的至少一个分量与所述三个分量中的其他分量组合,得到降噪后的图像数据。
结合第一方面,在第一方面的第一种实施方式中,所述方法还包括:
对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪;
所述对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪包括:
根据如下公式,对所述每个分量的高频小波系数进行降噪,
y=αx+(1-α)h(x),其中α是与边缘强度相关的参数,h(x)是关于x的衰减函数;
根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量,具体为:
根据所述每个分量降噪后的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的所述至少一个分量。
结合第一方面,在第一方面的第二种实施方式中,所述对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数,包括:
对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行n层小波分解,得到每个分量的n层高频小波系数和n层低频小波系数,其中n≥2,n为整数;
所述对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数,根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量包括:
对所述每个分量进行如下处理:
A:对第n层的低频小波系数进行递归降噪,得到第n层降噪后的低频小波系数,根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的的低频小波系数;
B:对第i层降噪后的低频小波系数进行递归降噪,得到第i层二次降噪后的低频小波系数,其中1≤i≤n-1,i为变量,i为整数,i初始值为n-1;
C:
当i>1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到第i-1层降噪后的低频小波系数,对i进行赋值,令i=i-1;,返回步骤B;
当i=1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到降噪后的一个分量。
结合第一方面的第二种实施方式,在第一方面的第三种实施方式中,所述方法还包括:对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪;
所述对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪包括:
根据如下公式,对所述每个分量的所述n层的每一层高频小波系数进行降噪,
yj=αjxj+(1-αj)h(xj),j≥1,j为整数;
其中,yj为第j个高频小波系数降噪后的值,xj为第j个高频小波系数的值,h(xj)是关于xj的衰减函数,αj是第j个高频小波系数所对应的像素点的边缘强度对应的边缘强度系数,0≤αj≤1。
结合第一方面的第三种实施方式,在第一方面的第四种实施方式中,所述根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波 重构,得到第n-1层降噪后的低频小波系数,具体为:
根据所述第n层降噪后的低频小波系数与第n层降噪后的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数;
所述根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构具体为:
根据所述第i层二次降噪后的低频小波系数与第i层降噪后的高频小波系数进行小波重构。
结合第一方面,或,第一方面的第一、第二、第三或第四种实施方式中任一种,在第一方面的第五种实施方式中,所述递归降噪,包括:
在至少一个方向上,第k个低频小波系数的降噪结果=第k个低频小波系数的值+f(第k-1个低频小波系数的降噪结果-第k个低频小波系数的值),其中k>1,k为整数,其中y=f(x)为降噪强度函数,x表示第k-1个低频小波系数的降噪结果与第k个低频小波系数的差,y表示降噪强度。
结合第一方面的第五种实施方式,在第一方面的第六种实施方式中,所述方法还包括:
在对所述第k个低频小波系数进行递归降噪后,根据如下公式,对所述第k个低频小波系数进行细节恢复;
第k个低频小波系数的细节恢复结果=第k个低频小波系数的降噪结果+g(第k个低频小波系数的值-第k个低频小波系数的降噪结果),其中y=g(x)为细节保持强度函数,x表示第k个低频小波系数的值与第k个低频小波系数的降噪结果的差,y表示细节保持强度的值。
结合第一方面的第三种或第四种实施方式,在第一方面的第七种实施方式中,所述第j个高频小波系数所对应的像素点的边缘强度包括:第j个高频小波系数所对应的像素点所对应的y、u、v三个分量中的至少一个分量所对应的边缘强度。
结合第一方面的第七种实施方式,在第一方面的第八种实施方式中,所述至少一个分量所对应的边缘强度为基于所述至少一个分量的高频小波系数和低频小波系数的边缘强度。
结合第一方面的第一种或第三种实施方式,在第一方面的第九种实施方式中,所述衰减函数为小波阈值函数,包括以下中的至少一个:硬阈值函数、软阈值函数。
结合第一方面的第五种实施方式,在第一方面的第十种实施方式中,所述至少一个方向包括以下中的至少一个:从左向右、从右向左、从上向下、从下向上。
第二方面,本发明实施例提供了一种图像降噪的终端,所述终端包括:
图像获取单元,用于获取图像的图像数据;
图像分解单元,用于对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数;
图像降噪处理单元,用于对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数,根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量;
降噪图像获取单元,用于当所述降噪后的至少一个分量为三个分量时,将所述降噪后的三个分量组合,得到降噪后的图像数据;
当所述降噪后的至少一个分量为一个或两个分量时,将所述降噪后的至少一个分量与所述三个分量中的其他分量组合,得到降噪后的图像数据。
结合第二方面,在第二方面的第一种实施方式中,所述终端还包括图像高频处理单元,用于对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪,包括根据如下公式,对所述每个分量的高频小波系数进行降噪,
y=αx+(1-α)h(x),其中α是与边缘强度相关的参数,h(x)是关于x的衰减函数;
所述图像降噪处理单元用于根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量具体为,用于根据所述每个分量降噪后的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的所述至少一个分量。
结合第二方面,在第二方面的第二种实施方式中,所述图像分解单元具体用于对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行n层小波分解,得到每个分量的n层高频小波系数和n层低频小波系数,其中n≥2,n为整数;
所述图像降噪处理单元具体用于对所述每个分量进行如下处理:
A:对第n层的低频小波系数进行递归降噪,得到第n层降噪后的低频小波系数,根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的的低频小波系数;
B:对第i层降噪后的低频小波系数进行递归降噪,得到第i层二次降噪后的低频小波系数,其中1≤i≤n-1,i为变量,i为整数,i初始值为n-1;
C:
当i>1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到第i-1层降噪后的低频小波系数,对i进行赋值,令i=i-1;,返回步骤B;
当i=1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到降噪后的一个分量。
结合第二方面的第二种实施方式,在第二方面的第三种实施方式中,所述终端还包括图像高频处理单元,用于对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪,包括根据如下公式,对所述每个分量的所述n层的每一层高频小波系数进行降噪,
yj=αjxj+(1-αj)h(xj),j≥1,j为整数;
其中,yj为第j个高频小波系数降噪后的值,xj为第j个高频小波系数的值,h(xj)是关于xj的衰减函数,αj是第j个高频小波系数所对应的像素点的边缘强度对应的边缘强度系数,0≤αj≤1。
结合第二方面的第三种实施方式,在第二方面的第四种实施方式中,所述图像降噪处理单元用于根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数具体为,用于根据所述第n层降噪后的低频小波系数与第n层降噪后的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数;
所述图像降噪处理单元用于根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构具体为,用于根据所述第i层二次降噪后的低频小波系数与第i层降噪后的高频小波系数进行小波重构。
结合第二方面,或第二方面的第一种、第二种、第三种或第四种实施方式中的任一种,在第二方面的第五种实施方式中,所述递归降噪,包括:
在至少一个方向上,第k个低频小波系数的降噪结果=第k个低频小波系数的值+f(第k-1个低频小波系数的降噪结果-第k个低频小波系数的值),其中k>1,k为整数,其中y=f(x)为降噪强度函数,x表示第k-1个低频小波系数的降噪结果与第k个低频小波系数的差,y表示降噪强度。
结合第二方面的第五种实施方式,在第二方面的第六种实施方式中,所述终端还包括细节恢复单元,用于在所述图像降噪处理单元对所述第k个低频小波系数进行递归降噪后,根据如下公式,对所述第k个低频小波系数进行细节恢复;
第k个低频小波系数的细节恢复结果=第k个低频小波系数的降噪结果+g(第k个低频小波系数的值-第k个低频小波系数的降噪结果),其中y=g(x)为细节保持强度函数,x表示第k个低频小波系数的值与第k个低频小波系数的降噪结果的差,y表示细节保持强度的值。
结合第二方面的第三种或第四种实施方式,在第二方面的第七种实施方式中,所述第j个高频小波系数所对应的像素点的边缘强度包括:第j个高频小波系数所对应的像素点所对应的y、u、v三个分量中的至少一个分量所对应的边缘强度。
结合第二方面的第七种实施方式,在第二方面的第八种实施方式中,所述至少一个分量所对应的边缘强度为基于所述至少一个分量的高频小波系数和低频小波系数的边缘强度。
结合第二方面的第一种或第三种实施方式,在第二方面的第九种实施方式中,所述衰减函数为小波阈值函数,包括以下中的至少一个:硬阈值函数、软阈值函数。
结合第二方面的第五种实施方式,在第二方面的第十种实施方式中,所述至少一个方向包括以下中的至少一个:从左向右、从右向左、从上向下、从下向上。
本发明实施例提供的图像降噪的方法及终端,通过对所述图像数据的y、u、v三个分量中的至少一个分量进行小波分解,对所述至少一个分量中的每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数;根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量;当所述降噪后的至少一个分量为三个分量时,将所述降噪后的三个分量组合,得到降噪后的图像数据;当所述降噪后的至少一个分量为一个或两个分量时,将所述降噪后的至少一个分量与所述三个分量中的其他分量组合,得到降噪后的图像数据。通过对图像进行小波分解,对低频小波系数进行递归降噪,降低了计算的数据量,计算复杂度低,提高了图像降噪的效率,同时通过递归方式降噪利用了图像中像素点之间的关系,保证了图像降噪的效果。
附图说明
为了更清楚地说明本发明实施例的技术方案,下面将对实施例中需要使用的附图作简要的介绍,显而易见地,下面描述中的附图仅仅是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1为本发明实施例一种用于图像降噪的方法流程图;
图2为本发明实施例另一种用于图像降噪的方法流程图;
图3为本发明实施例一种用于图像降噪的方法中对图像进行1层小波分解的示意图;
图4为本发明实施例一种用于图像降噪的方法中对图像进行2层小波分解的示意图;
图5为本发明实施例一种用于图像降噪的方法中对图像进行2层小波分解的示意图;
图6为本发明实施例又一种用于图像降噪的方法中降噪强度函数曲线及细节保持强度函数曲线的示意图;
图7为本发明实施例又一种用于图像降噪的方法的降噪效果示意图;
图8为本发明实施例一种用于图像降噪的终端的结构示意图;
图9为本发明实施例一种用于图像降噪的终端的另一结构示意图。
具体实施方式
下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。
本文中术语“和/或”,仅仅是一种描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B这三种情况。另外,本文中字符“/”,一般表示前后关联对象是 一种“或”的关系。
实施例一
如图1所示,在本发明提供的一种用于图像降噪的方法的一个实施例中,所述方法可以包括以下步骤:
101:获取一幅图像的图像数据;
102:对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数;
103:对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数;
104:根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量;
105:当所述降噪后的至少一个分量为三个分量时,将所述降噪后的三个分量组合,得到降噪后的图像数据;
当所述降噪后的至少一个分量为一个或两个分量时,将所述降噪后的至少一个分量与所述三个分量中的其他分量组合,得到降噪后的图像数据。
如,若所述降噪后的至少一个分量为三个分量,即得到降噪后的三个分量y、u、v时,将所述降噪后的三个分量y、u、v进行组合,即可得到降噪后的图像数据;若所述降噪后的至少一个分量为一个分量,如y分量时,将所述降噪后的y分量与所述三个分量中的其他两个分量即u、v分量进行组合,得到降噪后的图像数据;若所述降噪后的至少一个分量为两个分量,如y、u分量时,将所述降噪后的y、u分量与所述三个分量中的其他分量即v分量进行组合,得到降噪后的图像数据。
由上可以看出,通过对图像进行小波分解,对低频小波系数进行递归降噪,降低了计算的数据量,计算复杂度低,提高了图像降噪的效率,同时通过递归方式降噪利用了图像中像素点之间的关系,保证了图像降噪的效果。
实施例二
结合实施例一,在本发明提供的一种用于图像降噪的方法的另一实施例中,如图2所示,在步骤104之前还可以包括:
1031:对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪;
所述对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪包括:
根据如下公式,对所述每个分量的高频小波系数进行降噪,
y=αx+(1-α)h(x),其中α是与边缘强度相关的参数,h(x)是关于x的衰减函数。
本实施中,所述步骤104具体为104’:根据所述每个分量降噪后的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的所述至少一个分量。
所述步骤1031可以在步骤103之前或之后或同时执行,本发明对此不作限定。
通过对高频小波系数进行基于边缘信息的衰减函数降噪,参考了图像的边缘信息的边缘强度,根据边缘强度的不同,降噪程度也不同,从而可以在对高频小波系数进行降噪的同时,保留更多的图像的边缘信息等细节信息。对高频小波系数和低频小波系数采用了不同的方法来降噪,考虑了高频小波系数和低频小波系数各自包含的信息的特性,如高频部分一般包含边缘信息,在保证高频小波系数和低频小波系数各自的降噪效果的同时,提高了整幅图像综合的降噪效果,并保证了整幅图像的质量。
实施例三
结合实施例一,在本发明提供的一种用于图像降噪的方法的另一实施例中,所述步骤102可以包括:对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行n层小波分解,得到每个分量的n层高频小波系数和n层低频小波系数,其中n≥2,n为整数。
每一幅图像或照片,都可以具有y、u、v三个分量,对亮度分量y分量进行降噪,可以主要去除图像中存在的亮度噪声;对色度分量u、v分量进行降噪,可以主要去除图像中存在的彩色噪声。具体对哪些分量进行降噪,可以根据图像中存在的噪声的类型来进行选择。可以对图像数据的y、u、v三个分量中的至少一个分量进行一层或多层小波分解,得到每个分量的n层高频小波系数(其中,每一层可以有三个高频小波系数,分别为HL、LH和HH)和n层低频小波系数(其中,每一层可以有一个低频小波系数,为LL)。示例性的,第一层的低频小波系数可以记为LL1,第一层的高频小波系数可以分别记为:HL1、LH1和HH1。其它层可以依此类推。
本实施例中,所述步骤103、104可以包括:对所述每个分量进行如下处理:
A:对第n层的低频小波系数进行递归降噪,得到第n层降噪后的低频小波系数,根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的的低频小波系数;
B:对第i层降噪后的低频小波系数进行递归降噪,得到第i层二次降噪后的低频小波系数,其中1≤i≤n-1,i为变量,i为整数,i初始值为n-1;
C:
当i>1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到第i-1层降噪后的低频小波系数,对i进行赋值,令i=i-1;,返回步骤B;
当i=1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到降噪后的一个分量。
本发明实施例对含有噪声的图像的图像数据进行了小波分解,并从小波分解的最高层到第1层各层的低频小波系数,逐层进行了递归降噪,可以消除大面积的片状噪声,且计算复杂度小,小波分解每一层的低频小波系数的数据量为该层的总数据量的1/4,减少了需要处理的数据量,提高了效率,另 小波分解速度较快,进一步提高了图像降噪的效率。对于大面积的片状噪声,如果采用传统的窗口滤波的方式降噪,必须将窗口尺寸扩大到噪声尺寸的2倍以上,而随着窗口尺寸的增加,计算复杂度快速提升。本方案采用递归降噪,计算复杂度低,而不受到窗口尺寸的影响,像“推土机”一样,把噪声一步一步地逐渐蚕食掉。所述递归降噪,利用了低频小波系数之间(或图像的像素点之间)的关系来进行递归降噪,其效果相当于利用了整幅图像的信息来进行降噪,提高了降噪的效果,保证了图像的质量。
实施例四
结合实施例三,在本发明提供的一种用于图像降噪的方法的另一实施例中,在步骤104之前还可以包括:
1031’:对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪;
所述对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪包括:
根据如下公式,对所述每个分量的所述n层的每一层高频小波系数进行降噪,
yj=αjxj+(1-αj)h(xj),j≥1,j为整数;
其中,yj为第j个高频小波系数降噪后的值,xj为第j个高频小波系数的值,h(xj)是关于xj的衰减函数,αj是第j个高频小波系数所对应的像素点的边缘强度对应的边缘强度系数,0≤αj≤1。像素点的边缘强度越大,αj越趋近于1,从而yj越趋近于xj;像素点边缘强度越小,αj越趋近于0,从而yj越趋近于h(xj)。αj可以是第j个高频小波系数所对应的像素点的边缘强度进行归一化处理后得到的值。对每一层的高频小波系数进行处理时,j的取值与所对应的那一层的高频小波系数的个数相关。
所述步骤1031’可以在步骤103之前或之后或同时执行,本发明对此不作限定。
通过对高频小波系数进行基于边缘信息的衰减函数降噪,参考了图像的边缘信息的边缘强度,根据边缘强度的不同,降噪程度也不同,从而可以在对高频小波系数进行降噪的同时,保留更多的图像的边缘信息等细节信息。对高频小波系数和低频小波系数采用了不同的方法来降噪,考虑了高频小波系数和低频小波系数各自包含的信息的特性,如高频部分一般包含边缘信息,在保证高频小波系数和低频小波系数各自的降噪效果的同时,提高了整幅图像综合的降噪效果,并保证了整幅图像的质量。
实施例五
结合实施例四,在本发明提供的一种用于图像降噪的方法的另一实施例中,步骤A中所述根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数,具体为:
根据所述第n层降噪后的低频小波系数与第n层降噪后的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数。
步骤C中所述根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构具体为:
根据所述第i层二次降噪后的低频小波系数与第i层降噪后的高频小波系数进行小波重构。
实施例六
结合实施例一~五中任一实施例,在本发明提供的一种用于图像降噪的方法的另一实施例中,
所述递归降噪,可以包括:
在至少一个方向上,第k个低频小波系数的降噪结果=第k个低频小波系数的值+f(第k-1个低频小波系数的降噪结果-第k个低频小波系数的值),其中k>1,k为整数,其中y=f(x)为降噪强度函数,x表示第k-1个低频小波系数的降噪结果与第k个低频小波系数的差,y表示降噪强度。
可以根据所述图像的噪声估计水平按照预定规则得到该图像的若干个降 噪强度,即不同的降噪等级,也可以是系统预设的或用户设定的不同的降噪等级,所述降噪强度函数为根据所述图像的噪声估计水平和/或用户对该函数的设置来确定的函数。
所述递归降噪可以理解为,对每个低频小波系数,在至少一个方向上,若其之前有一个低频小波系数,则根据前一个低频小波系数降噪后的值来计算该低频小波系数降噪后的值。
对于一个方向上的第一个低频小波系数,其之前没有可供其参考的低频小波系数,第一个低频小波系数的降噪结果可以为该低频小波系数的值,或是采用降噪算法对该低频小波系数的值进行降噪后得到的值,所述降噪算法可以采用现有的降噪算法,本发明对比不作限定。
实施例七
结合实施例六,在本发明提供的一种用于图像降噪的方法的另一实施例中,所述方法还可以包括:
在对所述第k个低频小波系数进行递归降噪后,根据如下公式,对所述第k个低频小波系数进行细节恢复;
第k个低频小波系数的细节恢复结果=第k个低频小波系数的降噪结果+g(第k个低频小波系数的值-第k个低频小波系数的降噪结果),其中y=g(x)为细节保持强度函数,x表示第k个低频小波系数的值与第k个低频小波系数的降噪结果的差,y表示细节保持强度的值。
通过对每一层降噪后的低频小波系数,采用细节保持强度函数进行细节恢复,得到每一层细节恢复后的低频小波系数,可以对降噪后的低频小波系数进行调整,在降噪的同时保留更多的图像细节信息。
结合实施例四或五,在本发明提供的一种用于图像降噪的方法的另一实施例中,所述第j个高频小波系数所对应的像素点的边缘强度包括:第j个高频小波系数所对应的像素点所对应的y、u、v三个分量中的至少一个分量所对应的边缘强度。
通常,由于色度分量u、v的边缘信息不稳定,且不明显,所述第j个高频小波系数所对应的像素点的边缘强度可以既包括了对应于色度分量u、v的色度边缘强度(色度分量所对应的边缘强度),又包括了对应于y分量的亮度边缘强度(亮度分量所对应的边缘强度),即可以为y、u、v三个分量所对应的边缘强度之和。在具体实现时,可以取第j个高频小波系数所对应的像素点处的色度、亮度边缘强度中的最大值(即y、u、v三个分量所对应的边缘强度中的最大值),作为所述第j个高频小波系数所对应的像素点处的边缘强度,此种情况下,是对应的y、u、v三个分量中的一个分量的边缘强度。可选的,可以选y、u、v三个分量中的一个或多个分量所对应的边缘强度作为所述第j个高频小波系数所对应的像素点的边缘强度,如可以选y、u、v三个分量中边缘强度较大的两个,将所述两个的边缘强度之和作为所述第j个高频小波系数所对应的像素点的边缘强度。
进一步地,所述至少一个分量所对应的边缘强度为基于所述至少一个分量的高频小波系数和低频小波系数的边缘强度。
示例性的,对于亮度边缘强度的计算,由于其高频小波系数和低频小波系数均包含边缘信息,所以可以对其高频小波系数和低频小波系数均提取边缘强度(边缘强度的值可以采用经典的Sobel算子、Laplace算子等来计算),然后取高频小波系数和低频小波系数的边缘强度中(包括高频小波系数HLj、LHj、HHj及低频小波系数LLj各自对应的边缘强度)的最大值作为亮度边缘强度;对于色度边缘强度的计算,可以对其高频小波系数和低频小波系数均提取边缘强度,然后取高频小波系数和低频小波系数的边缘强度中的最大值作为色度边缘强度,实际操作中,因为色度的高频小波系数一般很弱,为节省计算量,可以只提取低频部分的边缘强度作为色度边缘强度。
进一步地,前述实施例中所述衰减函数具体为小波阈值函数,可以包括以下中的至少一个:硬阈值函数、软阈值函数。具体地,所述衰减函数可以为硬阈值或软阈值函数,或者,硬阈值和软阈值相结合的衰减函数等。
进一步地,前述实施例中所述至少一个方向包括以下中的至少一个:从左向右、从右向左、从上向下、从下向上。具体的,可以对每一层的低频小波系数进行上述四个方向的递归降噪,得到该层降噪后的低频小波系数。通过进行所述四个方向的递归降噪,可以保证降噪效果的对称性,提高图像的降噪效果的同时,保证了图像的质量。
从形式上看,对每个低频小波系数的递归降噪,在四个方向上仅利用了其上、下、左、右四个低频小波系数的信息,但是通过每一个低频小波系数的递归降噪,相当于间接地利用了整幅图像的信息,相当于提高了窗口滤波的尺寸。从而,可以很好地消除大面积的噪声,而且计算复杂度低,保证了降噪效果的同时提高了降噪效率。
下面,以对一幅图像进行3层小波分解(即n=3)为例来说明本发明的一种具体实施例。以主要对该图像进行彩色噪声降噪为例,由于彩色噪声降噪可以只涉及对u、v分量的处理,而且对u、v分量的处理过程可以相同或类似,以下以对u分量的处理为例进行阐述,对v分量的处理可以参考下面的过程。
步骤一:对该图像的色度分量u分量进行3层小波分解,得到第一层低频小波系数LL1和第一层高频小波系数HL1、LH1、HH1,参见图3;得到第二层低频小波系数LL2和第二层高频小波系数HL2、LH2、HH2,参见图4;得到第三层低频小波系数LL3和第二层高频小波系数HL3、LH3、HH3,参见图5。
步骤二:对小波分解的最高层(本实施例中为第三层)低频小波系数LL3进行四个方向的递归降噪;
1)第1遍,从左到右递归降噪
在从左到右的方向上,对每一行的每一个低频小波系数,参考其左侧参考点的值,按降噪强度函数y=f(x),从左到右进行递归运算,具体规则可以如下:
对该行第一个低频小波系数:
第一个低频小波系数的降噪结果=该低频小波系数的值对该行其他个低频小波系数:
参考点的值=左侧低频小波系数的降噪结果
该个低频小波系数的降噪结果=该个低频小波系数的值+f(参考点的值–该个低频小波系数的值)
即,在一个方向上,第一个低频小波系数的降噪结果为该低频小波系数的值,第k个低频小波系数的降噪结果=第k个低频小波系数的值+f(第k-1个低频小波系数的降噪结果-第k个低频小波系数的值),其中k>1,k为整数,其中y=f(x)为降噪强度函数,x表示第k-1个低频小波系数的降噪结果与第k个低频小波系数的差,y表示降噪强度,所述降噪强度函数为根据所述图像的噪声估计水平和/或用户对该函数的设置来确定的函数。
其中,y=f(x)可以为如图6实线所示的降噪强度函数曲线,水平轴x表示参考点的值与某个低频小波系数的差,竖直轴y表示降噪强度,可以根据所述图像的噪声估计水平按照预定规则得到该图像的若干个降噪强度,即不同的降噪等级,也可以是系统预设的或用户设定的不同的降噪等级,峰值点的位置(降噪强度最大的位置)可以根据所述图像的噪声估计水平和/或用户对该函数的设置来得到,峰值点的幅度可以是用户自由设定或设备预设的。例如,通过对所述图像的噪声水平进行估计,因为噪声最大的位置对应降噪强度的峰值,从而可以根据噪声最大的位置来确定该降噪强度函数曲线的峰值点的位置(即峰值点所对应的x值),当然也可以同时结合对所述图像的噪声估计水平和用户对该函数的设置,或者,仅根据用户对该函数的设置(如,用户选择该图像中自己感兴趣的部分来重点降噪,或者,用户根据自己的视觉感知,选择出该图像中用户认为噪声较大的需要重点降噪的部分),来得到y=f(x)函数峰值点的位置。y=f(x)函数峰值点的幅度,也即最大的降噪强度,可以是用户自由设定或设备预设的,例如,在设备上实现该方法时,设备可以根据预定的规则来设定峰值点的幅度;当然,如果降噪效果没有达到用户的预期, 用户也可以通过设置来对峰值点的幅度进行调节。
在该降噪强度函数曲线中,x越趋近于两端(表示边缘越强),y越趋近于0,从而该低频小波系数的降噪结果越趋近于该低频小波系数的原始值,边缘保持越好;x越趋近于峰值点的位置,y越大,从而该低频小波系数的降噪结果越趋近于参考点,降噪强度越强。
可以将y=f(x)以LUT(Look Up Table,查找表)的形式存入内存,将y=f(x)的计算转换为一次查表操作,这样可以提高算法的速度。
2)第2遍,从上到下递归降噪
在从上到下的方向上,对每一列的每一个低频小波系数,参考其上侧参考点的值,按降噪强度函数y=f(x),从上到下进行递归运算,具体规则可以与第1遍类似。
3)第3遍,从右到左递推降噪
在从右到左的方向上,对每一行的每一个低频小波系数,参考其右侧参考点的值,按降噪强度函数y=f(x),从右到左进行递归运算,具体规则可以与第1遍类似。
4)第4遍,从下到上递归降噪
在从下到上的方向上,对每一列的每一个低频小波系数,参考其下侧参考点的值,按降噪强度函数y=f(x),从下到上进行递归运算,具体规则可以与第1遍类似。
上述1)~4)分别从四个不同方向上,对图像的第3层小波分解的低频小波系数LL3进行了递归降噪,得到了第3层降噪后的低频小波系数。其中,上述四个方向在执行时的顺序(也即第1遍~第4遍在执行时的顺序)可以根据需要进行调整,本发明对此不作限定,本发明实施例仅例举了其中一种情况。本发明实施例例举的是4个方向的递归降噪,方向的多少可以根据需要而不同,如可以仅对2个方向上进行递归降噪(如对称的两个方向:从左到右和从右到左),也可以有更多的方向,如斜对角方向上也可以形成四个方向:左 下到右上,右上到左下,左上到右下,右下到左上。
经过上述1)~4)4遍操作后,从形式上看,对每个低频小波系数的降噪滤波,仅利用了其上、下、左、右四个低频小波系数的信息,但是实际上间接地利用了整幅图像的信息,相当于提高了窗口滤波的尺寸,从而,可以很好地消除图像中存在的大面结的片状彩色噪声,而且计算复杂度低。
可选的,在对图像的第3层小波分解的低频小波系数LL3进行了上述递归降噪后,还可以参考该图像第3层原始的低频小波系数和细节保持强度函数,对所述得到的第3层降噪后的低频小波系数进行细节恢复,具体如下:
5)第5遍,参考该图像原始的低频小波系数进行细节恢复
在对所述第k个低频小波系数进行递归降噪后,根据如下公式,对所述第k个低频小波系数进行细节恢复;
第k个低频小波系数的细节恢复结果=第k个低频小波系数的降噪结果+g(第k个低频小波系数的值-第k个低频小波系数的降噪结果),其中y=g(x)为细节保持强度函数,x表示第k个低频小波系数的值与第k个低频小波系数的降噪结果的差,y表示细节保持强度的值。
其中,y=g(x)可以为如图6虚线所示的细节保持强度函数曲线,水平轴x表示所述图像第k个低频小波系数的值与第k个低频小波系数的降噪结果的差,竖直轴y表示细节保持强度,可以根据所述图像的噪声估计水平按照预定规则得到该图像的若干个细节保持强度,即不同的细节保持等级,也可以是系统预设的或用户设定的不同的细节保持等级,所述细节保持强度函数为根据所述降噪强度函数和/或用户对该函数的设置来确定的函数。例如,其峰值点的位置(细节保持强度最大的位置)可以根据所述图像的噪声估计水平和/或用户对该函数的设置来得到,峰值点的幅度可以是用户自由设定或设备预设的。例如,通过对所述图像的噪声水平进行估计,因为噪声最大的位置对应降噪强度的峰值,可能损失较多图像的细节,从而可以根据噪声最大的位置来确定细节保持强度函数曲线的峰值点的位置(即峰值点所对应的x值), 当然也可以同时结合对所述图像的噪声估计水平和用户对该函数的设置,或者,仅根据用户对该函数的设置(如,用户选择该图像中自己感兴趣的部分来重点进行细节恢复,或者,用户根据自己的视觉感知,选择出该图像中用户认为细节损失较大的需要重点进行细节恢复的部分),来得到y=g(x)函数峰值点的位置。y=g(x)函数峰值点的幅度,也即最大的细节保持强度,可以是用户自由设定或设备预设的,例如,在设备上实现该方法时,设备可以根据预定的规则来设定峰值点的幅度,所述预定的规则可以是与所述y=f(x)函数峰值点的幅度之间预定的数量关系,即可以根据y=f(x)函数峰值点的幅度来确定y=g(x)函数峰值点的幅度,如图6中所示,y=g(x)函数峰值点的幅度约为y=f(x)函数峰值点的幅度的1/2;当然,如果细节恢复效果没有达到用户的预期,用户也可以通过设置来对y=g(x)函数峰值点的幅度进行调节。
同理,也可以将y=g(x)以LUT的形式存入内存,将y=g(x)的计算转换为一次查表操作,以提高算法速度。
步骤三:对第三层高频小波系数HL3、LH3、HH3进行基于边缘信息的衰减函数降噪;
所述基于边缘信息的衰减函数降噪包括按如下公式对该层的高频小波系数进行降噪:
yj=αjxj+(1-αj)h(xj),j≥1,j为整数;
其中,yj为第j个高频小波系数降噪后的值,xj为第j个高频小波系数的值,h(xj)是关于xj的衰减函数,αj是第j个高频小波系数所对应的像素点的边缘强度对应的边缘强度系数,0≤αj≤1。像素点的边缘强度越大,αj越趋近于1,从而yj越趋近于xj,边缘保持越好;像素点的边缘强度越小,αj越趋近于0,从而yj越趋近于h(xj),降噪效果越好。αj可以是第j个高频小波系数所对应的像素点的边缘强度进行归一化处理后得到的值。
其中,所述第j个高频小波系数所对应的像素点的边缘强度可以包括:第j个高频小波系数所对应的像素点所对应的y、u、v三个分量中的至少一个分 量所对应的边缘强度。
如,在本发明的一个实施例中,色度分量u、v的边缘信息可能不稳定,或不明显,这种情况下,所述第j个高频小波系数所对应的像素点的边缘强度可以既包括了对应于色度分量u、v的色度边缘强度(色度分量所对应的边缘强度),又包括了对应于y分量的亮度边缘强度(亮度分量所对应的边缘强度),即可以为y、u、v三个分量所对应的边缘强度之和。在具体实现时,可以取第j个高频小波系数所对应的像素点处的色度、亮度边缘强度中的最大值(即y、u、v三个分量所对应的边缘强度中的最大值),作为所述第j个高频小波系数所对应的像素点处的边缘强度,此种情况下,是对应的y、u、v三个分量中的一个分量的边缘强度。可选的,可以选y、u、v三个分量中的一个或多个分量所对应的边缘强度作为所述第j个高频小波系数所对应的像素点的边缘强度,如可以选y、u、v三个分量中边缘强度较大的两个,将所述两个的边缘强度之和作为所述第j个高频小波系数所对应的像素点的边缘强度。本实施例在对图像的高频小波系数降噪时,参考了亮度和色度的边缘信息,因此,在滤除高频噪声的同时,仍然可以较好地保留图像的细节信息。
可选的,所述至少一个分量所对应的边缘强度可以为基于所述至少一个分量的高频小波系数和低频小波系数的边缘强度。对于一个分量,其高频小波系数和低频小波系数均包含边缘信息,可以对高频小波系数和低频小波系数均提取边缘强度,基于高频小波系数和低频小波系数的边缘强度,来得到该分量对应的边缘强度。例如,对于亮度边缘强度的计算,由于其高频小波系数和低频小波系数均包含边缘信息,可以对高频小波系数和低频小波系数均提取边缘强度(如,边缘强度的值可以采用经典的Sobel算子、Laplace算子等来进行边缘检测),然后取高频小波系数和低频小波系数的边缘强度中(包括高频小波系数HLm、LHm、HHm及低频小波系数LLm各自对应的边缘强度,m指小波系数所对应的层数,本实施例中针对第三层时m取值为3)的最大值作为亮度边缘强度。可选的,对于色度边缘强度(u或v所对应的边缘强度) 的计算,可以对高频小波系数和低频小波系数均提取边缘强度,然后取高频小波系数和低频小波系数的边缘强度中的最大值作为色度边缘强度,色度分量的高频小波系数一般很弱,也可以只提取低频小波系数的边缘强度,以节省计算量。本实施例在对图像的高频小波系数降噪时,参考了至少一个分量所对应的边缘强度,且计算一个分量的边缘强度时,参考了其高频小波系数和低频小波系数,从而在滤除高频噪声的同时,可以保留图像更多的细节信息。
进一步的,所述衰减函数具体为小波阈值函数,可以包括以下中的至少一个:硬阈值函数、软阈值函数。具体地,所述衰减函数可以为硬阈值或软阈值函数,或者,硬阈值和软阈值相结合的衰减函数等。本实施例中以软阈值函数为例来进行处理。如下式子为软阈值函数的一种:
Figure PCTCN2014090421-appb-000001
其中,T为阈值,阈值的确定方法可以采用已有的各种方法,本发明实施例对此不作限定。
对图像的高频小波系数,参考了图像的亮度和色度的边缘强度,进行软阈值函数降噪,在滤除高频噪声的同时,仍然可以较好地保留图像的细节信息。
步骤四:对步骤二处理后的第三层低频小波系数LL3、步骤三处理后的第三层高频小波系数HL3、LH3、HH3进行小波重构,得到降噪后的第二层(即第n-1层)低频小波系数LL2;
步骤五:对步骤四中所述降噪后的第二层(即第i层,i初始值为n-1)低频小波系数LL2进行与步骤二类似的操作,得到二次降噪后的第二层低频小波系数LL2(注意:由于LL3与LL2代表了不同尺度(层次)、不同频率的信息,从而,这样可以对不同尺度、不同频率分别降噪);对第二层高频小波系数HL2、LH2、HH2进行与步骤三类似的操作,得到降噪后的第二层高频小波系 数HL2、LH2、HH2;
此时i的值为2,即i>1,则对步骤五降噪后的LL2、HL2、LH2、HH2进行小波重构,得到降噪后的第一层(即第i-1层)低频小波系数LL1;
步骤六:对步骤五中所述降噪后的第一层低频小波系数LL1进行与步骤二类似的操作,得到二次降噪后的第一层低频小波系数LL1;对第一层高频小波系数HL1、LH1、HH1进行与步骤三类似的操作,得到降噪后的第一层高频小波系数HL1、LH1、HH1;
上述步骤相当于在步骤五执行完后,对i进行赋值,令i=i-1;,将此时i的值为1代入步骤二和步骤三。执行完步骤二和步骤三后,i的值为1,则当i=1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,可以得到降噪后的一个分量,即对步骤六降噪后的LL1、HL1、LH1、HH1进行小波重构,得到降噪后的色度分量u。
对v分量可以参考上述步骤一~步骤六进行处理,得到降噪后的色度分量v。
将亮度分量和上述降噪后的色度分量组合,得到去除了彩色噪声后的图像,如图7(a)和图7(b)所示,分别为原始含噪图像和对该图像去彩色噪声后的图像。可以看出,原始图像中车的颜色本身是灰色的,但图7(a)中包含了大量的红、蓝、绿等彩色噪声,显得花花绿绿,但图7(b)因为去除了这些彩色噪声,图像的噪声就干净很多。
进一步可选的,在本发明的另一个实施例中,可以参考上述步骤一~步骤六对亮度分量y进行降噪处理,这种情况下,去除的是亮度噪声。将降噪后的亮度分量和上述降噪后的色度分量组合,可以得到去除了亮度噪声和彩色噪声后的图像,如图7(c)所示意的,可以看出,图像在去除了噪声的同时,细节也保持很好。
需要说明的是,对不同的分量进行处理时,所涉及的各个函数,包括y=f(x)、y=g(x)以及衰减函数等,可以相同,也可以不同,确定这些函数的方 法也可以相同或不同,具体可根据需要进行调整,本发明对此不作限定。
以上各实施例提供的用于图像降噪的方法,可以应用于终端拍照时对图像进行降噪处理,可以提升拍照图像的质量和用户的体验。
本发明实施例还提供了一种用于图像降噪的终端,如图8所示为本发明提供的终端的一个实施例,在该实施例中,所述终端包括:
图像获取单元800,用于获取图像的图像数据;
图像分解单元810,用于对所述图像数据的y、u、v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数,其中,y为图像的亮度,u、v为图像的色度;
图像降噪处理单元820,用于对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数,根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量;
降噪图像获取单元830,用于当所述降噪后的至少一个分量为三个分量时,将所述降噪后的三个分量组合,得到降噪后的图像数据;当所述降噪后的至少一个分量为一个或两个分量时,将所述降噪后的至少一个分量与所述三个分量中的其他分量组合,得到降噪后的图像数据。
在本发明提供的终端的一个实施例中,所述终端还包括图像高频处理单元840,用于对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪,包括根据如下公式,对所述每个分量的高频小波系数进行降噪,
y=αx+(1-α)h(x),其中α是与边缘强度相关的参数,h(x)是关于x的衰减函数;
所述图像降噪处理单元820用于根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量具体为,用于根据所述每个分量降噪后的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的所述至少一个分量。
本发明提供的终端的一个实施例中,所述图像分解单元810具体用于对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行n层小波分解,得到每个分量的n层高频小波系数和n层低频小波系数,其中n≥2,n为整数;
所述图像降噪处理单元820具体用于对所述每个分量进行如下处理:
A:对第n层的低频小波系数进行递归降噪,得到第n层降噪后的低频小波系数,根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的的低频小波系数;
B:对第i层降噪后的低频小波系数进行递归降噪,得到第i层二次降噪后的低频小波系数,其中1≤i≤n-1,i为变量,i为整数,i初始值为n-1;
C:
当i>1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到第i-1层降噪后的低频小波系数,对i进行赋值,令i=i-1;,返回步骤B;
当i=1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到降噪后的一个分量。
本发明提供的终端的一个实施例中,所述终端还包括图像高频处理单元840,用于对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪,包括根据如下公式,对所述每个分量的所述n层的每一层高频小波系数进行降噪,
yj=αjxj+(1-αj)h(xj),j≥1,j为整数;
其中,yj为第j个高频小波系数降噪后的值,xj为第j个高频小波系数的值,h(xj)是关于xj的衰减函数,αj是第j个高频小波系数所对应的像素点的边缘强度对应的边缘强度系数,0≤αj≤1。
本发明提供的终端的一个实施例中,所述图像降噪处理单元820用于根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构, 得到第n-1层降噪后的低频小波系数具体为,用于根据所述第n层降噪后的低频小波系数与第n层降噪后的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数;
所述图像降噪处理单元820用于根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构具体为,用于根据所述第i层二次降噪后的低频小波系数与第i层降噪后的高频小波系数进行小波重构。
本发明提供的终端的一个实施例中,所述递归降噪,包括:
在至少一个方向上,第k个低频小波系数的降噪结果=第k个低频小波系数的值+f(第k-1个低频小波系数的降噪结果-第k个低频小波系数的值),其中k>1,k为整数,其中y=f(x)为降噪强度函数,x表示第k-1个低频小波系数的降噪结果与第k个低频小波系数的差,y表示降噪强度。
本发明提供的终端的一个实施例中,所述终端还包括细节恢复单元821,用于在所述图像降噪处理单元820对所述第k个低频小波系数进行递归降噪后,根据如下公式,对所述第k个低频小波系数进行细节恢复;
第k个低频小波系数的细节恢复结果=第k个低频小波系数的降噪结果+g(第k个低频小波系数的值-第k个低频小波系数的降噪结果),其中y=g(x)为细节保持强度函数,x表示第k个低频小波系数的值与第k个低频小波系数的降噪结果的差,y表示细节保持强度的值。
本发明提供的终端的一个实施例中,所述第j个高频小波系数所对应的像素点的边缘强度包括:第j个高频小波系数所对应的像素点所对应的y、u、v三个分量中的至少一个分量所对应的边缘强度。
本发明提供的终端的一个实施例中,所述至少一个分量所对应的边缘强度为基于所述至少一个分量的高频小波系数和低频小波系数的边缘强度。
本发明提供的终端的一个实施例中,所述衰减函数为小波阈值函数,包括以下中的至少一个:硬阈值函数、软阈值函数。
本发明提供的终端的一个实施例中,所述至少一个方向包括以下中的至 少一个:从左向右、从右向左、从上向下、从下向上。
本发明提供的终端的各实施例中各单元执行的步骤及各步骤的具体内容可以参考上述方法实施例中的相关部分,在此不再赘述。
本发明实施例还提供了一种用于图像降噪的终端,如图9所示为本发明提供的终端的一个实施例,在该实施例中,所述终端包括存储器、处理器和通信总线,所述处理器通过所述通信总线与所述存储器连接。进一步地,所述终端还可以包括通信接口,通过通信接口与其他设备(例如其他终端或接入点设备等)通信连接。
所述存储器可以为一个或多个,用于存储所述终端获取的图像数据,以及存储实现用于图像降噪的方法的指令;其中,所述图像数据和所述指令可以存储在同一个存储器中或是分别存储在不同的存储器中;
所述处理器可以为一个或多个,当所述一个或多个处理器调取所述一个或多个存储器中存储的图像数据以及实现用于图像降噪的方法的指令时,可以对所述图像数据执行如下步骤:
对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数;
对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数;
根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量;
当所述降噪后的至少一个分量为三个分量时,将所述降噪后的三个分量组合,得到降噪后的图像数据;
当所述降噪后的至少一个分量为一个或两个分量时,将所述降噪后的至少一个分量与所述三个分量中的其他分量组合,得到降噪后的图像数据。
其中,所述一个或多个处理器调取所述图像数据以及实现用于图像降噪的方法的指令时,可以是一个处理器调取所述图像数据以及所述指令,所述 图像数据和所述指令也可以由不同的处理器分别去调取。
在本发明提供的终端的一个实施例中,所述处理器在调取所述一个或多个存储器中存储的图像数据以及实现用于图像降噪的方法的指令时,还可以对所述图像数据执行如下步骤:对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪;
所述对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪包括:
根据如下公式,对所述每个分量的高频小波系数进行降噪,
y=αx+(1-α)h(x),其中α是与边缘强度相关的参数,h(x)是关于x的衰减函数;
所述处理器根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量,具体为:
根据所述每个分量降噪后的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的所述至少一个分量。
在本发明提供的终端的一个实施例中,所述处理器对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数,包括:
对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行n层小波分解,得到每个分量的n层高频小波系数和n层低频小波系数,其中n≥2,n为整数;
所述对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数,根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量包括:
对所述每个分量进行如下处理:
A:对第n层的低频小波系数进行递归降噪,得到第n层降噪后的低频小 波系数,根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的的低频小波系数;
B:对第i层降噪后的低频小波系数进行递归降噪,得到第i层二次降噪后的低频小波系数,其中1≤i≤n-1,i为变量,i为整数,i初始值为n-1;
C:
当i>1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到第i-1层降噪后的低频小波系数,对i进行赋值,令i=i-1;,返回步骤B;
当i=1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到降噪后的一个分量。
在本发明提供的终端的一个实施例中,所述处理器对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪包括:
根据如下公式,对所述每个分量的所述n层的每一层高频小波系数进行降噪,
yj=αjxj+(1-αj)h(xj),j≥1,j为整数;
其中,yj为第j个高频小波系数降噪后的值,xj为第j个高频小波系数的值,h(xj)是关于xj的衰减函数,αj是第j个高频小波系数所对应的像素点的边缘强度对应的边缘强度系数,0≤αj≤1。
在本发明提供的终端的一个实施例中,所述处理器根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数,具体为:
根据所述第n层降噪后的低频小波系数与第n层降噪后的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数;
在本发明提供的终端的一个实施例中,所述处理器根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构具体为:
根据所述第i层二次降噪后的低频小波系数与第i层降噪后的高频小波系 数进行小波重构。
在本发明提供的终端的一个实施例中,所述递归降噪,包括:
在至少一个方向上,第k个低频小波系数的降噪结果=第k个低频小波系数的值+f(第k-1个低频小波系数的降噪结果-第k个低频小波系数的值),其中k>1,k为整数,其中y=f(x)为降噪强度函数,x表示第k-1个低频小波系数的降噪结果与第k个低频小波系数的差,y表示降噪强度。
在本发明提供的终端的一个实施例中,所述处理器在对所述第k个低频小波系数进行递归降噪后,还可以根据如下公式,对所述第k个低频小波系数进行细节恢复;
第k个低频小波系数的细节恢复结果=第k个低频小波系数的降噪结果+g(第k个低频小波系数的值-第k个低频小波系数的降噪结果),其中y=g(x)为细节保持强度函数,x表示第k个低频小波系数的值与第k个低频小波系数的降噪结果的差,y表示细节保持强度的值。
在本发明提供的终端的一个实施例中,所述第j个高频小波系数所对应的像素点的边缘强度包括:第j个高频小波系数所对应的像素点所对应的y、u、v三个分量中的至少一个分量所对应的边缘强度。
在本发明提供的终端的一个实施例中,所述至少一个分量所对应的边缘强度为基于所述至少一个分量的高频小波系数和低频小波系数的边缘强度。
在本发明提供的终端的一个实施例中,所述衰减函数为小波阈值函数,包括以下中的至少一个:硬阈值函数、软阈值函数。
在本发明提供的终端的一个实施例中,所述至少一个方向包括以下中的至少一个:从左向右、从右向左、从上向下、从下向上。
本发明提供的终端的各实施例中处理器在调取所述一个或多个存储器中存储的图像数据以及实现用于图像降噪的方法的指令时,可以对所述图像数据执行的步骤及各步骤的具体内容可以参考前述方法实施例中的相关部分,在此不再赘述。
还需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分步骤是可以通过程序来指令相关硬件来完成,所述的程序可以存储于一终端的可读存储介质中,该程序在执行时,包括上述全部或部分步骤,所述的存储介质,如:FLASH、EEPROM等。上述实施例中的方法也可以通过图像处理的芯片来实现。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,仅以上述各功能模块的划分进行举例说明,实际应用中,可以根据需要而将上述功能分配由不同的功能模块完成,即将装置的内部结构划分成不同的功能模块,以完成以上描述的全部或者部分功能。上述描述的系统,装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
在本申请所提供的几个实施例中,应该理解到,所揭露的系统,装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述模块或单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,装置或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本发明各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用软件功能单元的形式实现。
以上所述的具体实施方式,对本发明的目的、技术方案和有益效果进行了进一步详细说明,所应理解的是,不同的实施例可以进行组合,以上所述仅为本发明的具体实施方式而已,并不用于限定本发明的保护范围,凡在本发明的精神和原则之内,所做的任何组合、修改、等同替换、改进等,均应包含在本发明的保护范围之内。

Claims (22)

  1. 一种图像降噪的方法,其特征在于:
    获取图像的图像数据;
    对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数;
    对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数;
    根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量;
    当所述降噪后的至少一个分量为三个分量时,将所述降噪后的三个分量组合,得到降噪后的图像数据;
    当所述降噪后的至少一个分量为一个或两个分量时,将所述降噪后的至少一个分量与所述三个分量中的其他分量组合,得到降噪后的图像数据。
  2. 如权利要求1所述的方法,其特征在于,所述方法还包括:
    对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪;
    所述对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪包括:
    根据如下公式,对所述每个分量的高频小波系数进行降噪,
    y=αx+(1-α)h(x),其中α是与边缘强度相关的参数,h(x)是关于x的衰减函数;
    所述根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量,具体为:
    根据所述每个分量降噪后的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的所述至少一个分量。
  3. 如权利要求1所述的方法,其特征在于,所述对所述图像数据的亮度分 量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数,包括:
    对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行n层小波分解,得到每个分量的n层高频小波系数和n层低频小波系数,其中n≥2,n为整数;
    所述对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数,根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量包括:
    对所述每个分量进行如下处理:
    A:对第n层的低频小波系数进行递归降噪,得到第n层降噪后的低频小波系数,根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的的低频小波系数;
    B:对第i层降噪后的低频小波系数进行递归降噪,得到第i层二次降噪后的低频小波系数,其中1≤i≤n-1,i为变量,i为整数,i初始值为n-1;
    C:
    当i>1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到第i-1层降噪后的低频小波系数,对i进行赋值,令i=i-1;,返回步骤B;
    当i=1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到降噪后的一个分量。
  4. 如权利要求3所述的方法,其特征在于,所述方法还包括:对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪;
    所述对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪包括:
    根据如下公式,对所述每个分量的所述n层的每一层高频小波系数进行降噪,
    yj=αjxj+(1-αj)h(xj),j≥1,j为整数;
    其中,yj为第j个高频小波系数降噪后的值,xj为第j个高频小波系数的值,h(xj)是关于xj的衰减函数,αj是第j个高频小波系数所对应的像素点的边缘强度对应的边缘强度系数,0≤αj≤1。
  5. 如权利要求4所述的方法,其特征在于,所述根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数,具体为:
    根据所述第n层降噪后的低频小波系数与第n层降噪后的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数;
    所述根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构具体为:
    根据所述第i层二次降噪后的低频小波系数与第i层降噪后的高频小波系数进行小波重构。
  6. 如权利要求1-5任一所述的方法,其特征在于,所述递归降噪,包括:
    在至少一个方向上,第k个低频小波系数的降噪结果=第k个低频小波系数的值+f(第k-1个低频小波系数的降噪结果-第k个低频小波系数的值),其中k>1,k为整数,其中y=f(x)为降噪强度函数,x表示第k-1个低频小波系数的降噪结果与第k个低频小波系数的差,y表示降噪强度。
  7. 如权利要求6所述的方法,其特征在于,所述方法还包括:
    在对所述第k个低频小波系数进行递归降噪后,根据如下公式,对所述第k个低频小波系数进行细节恢复;
    第k个低频小波系数的细节恢复结果=第k个低频小波系数的降噪结果+g(第k个低频小波系数的值-第k个低频小波系数的降噪结果),其中y=g(x)为细节保持强度函数,x表示第k个低频小波系数的值与第k个低频小波系数的降噪结果的差,y表示细节保持强度的值。
  8. 如权利要求4或5所述的方法,其特征在于,所述第j个高频小波系数 所对应的像素点的边缘强度包括:第j个高频小波系数所对应的像素点所对应的y、u、v三个分量中的至少一个分量所对应的边缘强度。
  9. 如权利要求8所述的方法,其特征在于,所述至少一个分量所对应的边缘强度为基于所述至少一个分量的高频小波系数和低频小波系数的边缘强度。
  10. 如权利要求2或4所述的方法,其特征在于,所述衰减函数为小波阈值函数,包括以下中的至少一个:硬阈值函数、软阈值函数。
  11. 如权利要求6所述的方法,其特征在于,所述至少一个方向包括以下中的至少一个:从左向右、从右向左、从上向下、从下向上。
  12. 一种图像降噪的终端,其特征在于,所述终端包括:
    图像获取单元,用于获取图像的图像数据;
    图像分解单元,用于对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行小波分解,得到每个分量的高频小波系数和低频小波系数;
    图像降噪处理单元,用于对所述每个分量的低频小波系数进行递归降噪,得到所述每个分量降噪后的低频小波系数,根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量;
    降噪图像获取单元,用于当所述降噪后的至少一个分量为三个分量时,将所述降噪后的三个分量组合,得到降噪后的图像数据;
    当所述降噪后的至少一个分量为一个或两个分量时,将所述降噪后的至少一个分量与所述三个分量中的其他分量组合,得到降噪后的图像数据。
  13. 如权利要求12所述的终端,其特征在于,所述终端还包括图像高频处理单元,用于对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪,包括根据如下公式,对所述每个分量的高频小波系数进行降噪,
    y=αx+(1-α)h(x),其中α是与边缘强度相关的参数,h(x)是关于x的衰减函数;
    所述图像降噪处理单元,用于根据所述每个分量的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的至少一个分量具体为,用于根据所述每个分量降噪后的高频小波系数和所述每个分量降噪后的低频小波系数进行小波重构,得到降噪后的所述至少一个分量。
  14. 如权利要求12所述的终端,其特征在于,所述图像分解单元具体用于对所述图像数据的亮度分量y、色度分量u和v三个分量中的至少一个分量进行n层小波分解,得到每个分量的n层高频小波系数和n层低频小波系数,其中n≥2,n为整数;
    所述图像降噪处理单元具体用于对所述每个分量进行如下处理:
    A:对第n层的低频小波系数进行递归降噪,得到第n层降噪后的低频小波系数,根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的的低频小波系数;
    B:对第i层降噪后的低频小波系数进行递归降噪,得到第i层二次降噪后的低频小波系数,其中1≤i≤n-1,i为变量,i为整数,i初始值为n-1;
    C:
    当i>1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到第i-1层降噪后的低频小波系数,对i进行赋值,令i=i-1;,返回步骤B;
    当i=1时,根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构,得到降噪后的一个分量。
  15. 如权利要求14所述的终端,其特征在于,所述终端还包括图像高频处理单元,用于对所述每个分量的高频小波系数进行基于边缘信息的衰减函数降噪,包括根据如下公式,对所述每个分量的所述n层的每一层高频小波系数进行降噪,
    yj=αjxj+(1-αj)h(xj),j≥1,j为整数;
    其中,yj为第j个高频小波系数降噪后的值,xj为第j个高频小波系数的值, h(xj)是关于xj的衰减函数,αj是第j个高频小波系数所对应的像素点的边缘强度对应的边缘强度系数,0≤αj≤1。
  16. 如权利要求15所述的终端,其特征在于,所述图像降噪处理单元用于根据所述第n层降噪后的低频小波系数与第n层的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数具体为,用于根据所述第n层降噪后的低频小波系数与第n层降噪后的高频小波系数进行小波重构,得到第n-1层降噪后的低频小波系数;
    所述图像降噪处理单元用于根据所述第i层二次降噪后的低频小波系数与第i层高频小波系数进行小波重构具体为,用于根据所述第i层二次降噪后的低频小波系数与第i层降噪后的高频小波系数进行小波重构。
  17. 如权利要求12-16任一所述的终端,其特征在于,所述递归降噪,包括:
    在至少一个方向上,第k个低频小波系数的降噪结果=第k个低频小波系数的值+f(第k-1个低频小波系数的降噪结果-第k个低频小波系数的值),其中k>1,k为整数,其中y=f(x)为降噪强度函数,x表示第k-1个低频小波系数的降噪结果与第k个低频小波系数的差,y表示降噪强度。
  18. 如权利要求17所述的终端,其特征在于,所述终端还包括细节恢复单元,用于在所述图像降噪处理单元对所述第k个低频小波系数进行递归降噪后,根据如下公式,对所述第k个低频小波系数进行细节恢复;
    第k个低频小波系数的细节恢复结果=第k个低频小波系数的降噪结果+g(第k个低频小波系数的值-第k个低频小波系数的降噪结果),其中y=g(x)为细节保持强度函数,x表示第k个低频小波系数的值与第k个低频小波系数的降噪结果的差,y表示细节保持强度的值。
  19. 如权利要求15或16所述的终端,其特征在于,所述第j个高频小波系数所对应的像素点的边缘强度包括:第j个高频小波系数所对应的像素点所对应的y、u、v三个分量中的至少一个分量所对应的边缘强度。
  20. 如权利要求19所述的终端,其特征在于,所述至少一个分量所对应的 边缘强度为基于所述至少一个分量的高频小波系数和低频小波系数的边缘强度。
  21. 如权利要求13或15所述的终端,其特征在于,所述衰减函数为小波阈值函数,包括以下中的至少一个:硬阈值函数、软阈值函数。
  22. 如权利要求17所述的方法,其特征在于,所述至少一个方向包括以下中的至少一个:从左向右、从右向左、从上向下、从下向上。
PCT/CN2014/090421 2013-11-08 2014-11-06 一种用于图像降噪的方法及终端 WO2015067186A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
US15/033,550 US9904986B2 (en) 2013-11-08 2014-11-06 Image denoising method and terminal
JP2016524563A JP6216987B6 (ja) 2013-11-08 2014-11-06 画像ノイズ除去方法および端末
EP14860150.3A EP3051485B1 (en) 2013-11-08 2014-11-06 Method and terminal used for image noise reduction

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201310554182.2A CN104639800B (zh) 2013-11-08 2013-11-08 一种用于图像降噪的方法及终端
CN201310554182.2 2013-11-08

Publications (1)

Publication Number Publication Date
WO2015067186A1 true WO2015067186A1 (zh) 2015-05-14

Family

ID=53040910

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/090421 WO2015067186A1 (zh) 2013-11-08 2014-11-06 一种用于图像降噪的方法及终端

Country Status (4)

Country Link
US (1) US9904986B2 (zh)
EP (1) EP3051485B1 (zh)
CN (1) CN104639800B (zh)
WO (1) WO2015067186A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111696076A (zh) * 2020-05-07 2020-09-22 杭州电子科技大学 一种新型立体图像舒适度预测方法
CN116894951A (zh) * 2023-09-11 2023-10-17 济宁市质量计量检验检测研究院(济宁半导体及显示产品质量监督检验中心、济宁市纤维质量监测中心) 一种基于图像处理的珠宝在线监测方法

Families Citing this family (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105824186A (zh) * 2016-05-05 2016-08-03 杭州电子科技大学 一种物体三维透视成像方法和系统
US10983236B2 (en) 2017-06-20 2021-04-20 Saudi Arabian Oil Company Super-resolution radon transform based on thresholding
JP6904842B2 (ja) 2017-08-03 2021-07-21 キヤノン株式会社 画像処理装置、画像処理方法
CN110246088B (zh) * 2018-03-07 2021-07-13 舜宇光学(浙江)研究院有限公司 基于小波变换的图像亮度降噪方法及其图像降噪系统
CN109242799B (zh) * 2018-09-19 2021-10-12 安徽理工大学 一种变阈值的小波去噪方法
CN111292250B (zh) * 2018-12-29 2023-01-13 展讯通信(上海)有限公司 图像处理方法及装置
CN110349101B (zh) * 2019-06-26 2023-11-14 上海集成电路研发中心有限公司 一种多尺度联合的小波图像去噪方法
CN111127365B (zh) * 2019-12-26 2023-08-29 重庆矢崎仪表有限公司 基于三次样条曲线拟合的hud畸变矫正方法
CN111161188B (zh) * 2019-12-30 2023-09-08 珠海全志科技股份有限公司 降低图像色彩噪声的方法、计算机装置及可读存储介质
CN111242856B (zh) * 2020-01-06 2023-03-24 西安工程大学 一种基于剪切波的双重图像去噪方法
CN113160063A (zh) * 2020-11-05 2021-07-23 北京工业大学 一种基于小波及中值滤波对隧道图像去噪的方法
CN112348031A (zh) * 2020-11-17 2021-02-09 安徽理工大学 一种去除指纹图像混合噪声的改进小波阈值去噪方法
CN112634164B (zh) * 2020-12-29 2022-09-30 哈尔滨工业大学芜湖机器人产业技术研究院 一种基于多尺度哈尔小波变换的图像去噪方法
CN112884674A (zh) * 2021-03-12 2021-06-01 汉得利(常州)电子股份有限公司 一种基于contourlet变换的红外阵列图像增强系统及方法
CN113837042B (zh) * 2021-09-14 2024-04-19 中国铁路设计集团有限公司 一种基于局部均值分解的小波阈值自动化变形监测信号去噪方法
CN115804568B (zh) * 2022-05-05 2024-10-15 湖南万脉医疗科技有限公司 一种智能呼吸信号处理方法和装置
CN117221458B (zh) * 2023-10-23 2024-05-03 上海为旌科技有限公司 一种分频段去除图像色度噪声的方法和系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101561925A (zh) * 2009-03-27 2009-10-21 天津大学 基于全相位邻域加窗的小波图像去噪方法
EP1519564B1 (en) * 2003-09-29 2010-01-27 Xerox Corporation Method for embedding color image information into a black-and-white image
CN101719267A (zh) * 2009-11-09 2010-06-02 中兴通讯股份有限公司 一种带噪图像降噪处理方法和系统
CN101944230A (zh) * 2010-08-31 2011-01-12 西安电子科技大学 基于多尺度的自然图像非局部均值去噪方法
US8457433B2 (en) * 2010-01-28 2013-06-04 Texas Instruments Incorporated Methods and systems for image noise filtering

Family Cites Families (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5352942B2 (ja) 2005-03-31 2013-11-27 株式会社ニコン 画像処理方法、画像処理プログラム、および画像処理装置
JP2007316983A (ja) 2006-05-26 2007-12-06 Matsushita Electric Ind Co Ltd 画像処理装置、画像処理方法、及びそのプログラム
WO2008105268A1 (ja) * 2007-02-28 2008-09-04 Nikon Corporation 画像処理方法
WO2009098546A1 (en) 2008-02-06 2009-08-13 Zoran (France) Video enhancement using recursive bandlets
US8284271B2 (en) 2009-06-05 2012-10-09 Apple Inc. Chroma noise reduction for cameras
CN101882305B (zh) * 2010-06-30 2014-02-05 中山大学 一种图像增强处理的方法
US8836824B2 (en) 2011-06-02 2014-09-16 Apple Inc. Systems and methods for threshold-based luma channel noise reduction
CN102999884B (zh) 2011-09-16 2015-11-25 株式会社东芝 图像处理设备和方法
KR101828411B1 (ko) * 2011-09-21 2018-02-13 삼성전자주식회사 영상 처리 방법 및 영상 처리 장치
CN102332155A (zh) * 2011-10-19 2012-01-25 江苏技术师范学院 一种图像增强方法
US9242661B2 (en) 2013-05-24 2016-01-26 Spx International Limited Railway point crank system
CA2913246A1 (en) 2013-05-24 2014-12-04 Spx International Limited Railway point crank system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1519564B1 (en) * 2003-09-29 2010-01-27 Xerox Corporation Method for embedding color image information into a black-and-white image
CN101561925A (zh) * 2009-03-27 2009-10-21 天津大学 基于全相位邻域加窗的小波图像去噪方法
CN101719267A (zh) * 2009-11-09 2010-06-02 中兴通讯股份有限公司 一种带噪图像降噪处理方法和系统
US8457433B2 (en) * 2010-01-28 2013-06-04 Texas Instruments Incorporated Methods and systems for image noise filtering
CN101944230A (zh) * 2010-08-31 2011-01-12 西安电子科技大学 基于多尺度的自然图像非局部均值去噪方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111696076A (zh) * 2020-05-07 2020-09-22 杭州电子科技大学 一种新型立体图像舒适度预测方法
CN111696076B (zh) * 2020-05-07 2023-07-07 杭州电子科技大学 一种新型立体图像舒适度预测方法
CN116894951A (zh) * 2023-09-11 2023-10-17 济宁市质量计量检验检测研究院(济宁半导体及显示产品质量监督检验中心、济宁市纤维质量监测中心) 一种基于图像处理的珠宝在线监测方法
CN116894951B (zh) * 2023-09-11 2023-12-08 济宁市质量计量检验检测研究院(济宁半导体及显示产品质量监督检验中心、济宁市纤维质量监测中心) 一种基于图像处理的珠宝在线监测方法

Also Published As

Publication number Publication date
CN104639800A (zh) 2015-05-20
EP3051485A4 (en) 2016-11-16
EP3051485B1 (en) 2018-09-19
JP2016539404A (ja) 2016-12-15
US20160284067A1 (en) 2016-09-29
US9904986B2 (en) 2018-02-27
JP6216987B2 (ja) 2017-10-25
EP3051485A1 (en) 2016-08-03
CN104639800B (zh) 2017-11-24

Similar Documents

Publication Publication Date Title
WO2015067186A1 (zh) 一种用于图像降噪的方法及终端
Liu et al. Efficient single image dehazing and denoising: An efficient multi-scale correlated wavelet approach
EP4109392A1 (en) Image processing method and image processing device
US6801672B1 (en) Removing noise from a color image using wavelets
CN105096280B (zh) 处理图像噪声的方法及装置
US8878963B2 (en) Apparatus and method for noise removal in a digital photograph
CN104125374B (zh) 图像处理装置及其影像噪声抑制方法
WO2018086289A1 (zh) 显示面板自动光学检测中的背景抑制方法及检测装置
JP4862897B2 (ja) 画像処理方法
Mao et al. Multi-directional laplacian pyramid image fusion algorithm
CN109978775A (zh) 颜色去噪方法及装置
JP5256582B2 (ja) 画像処理方法、画像処理プログラム、および画像処理装置
CN110322404B (zh) 一种图像增强方法及系统
CN113362338B (zh) 铁轨分割方法、装置、计算机设备和铁轨分割处理系统
US20080285868A1 (en) Simple Adaptive Wavelet Thresholding
US11915392B2 (en) Image enhancement method and apparatus
Zhao et al. Learning a deep convolutional network for subband image denoising
Ponomaryov et al. Fuzzy color video filtering technique for sequences corrupted by additive Gaussian noise
CN106447616B (zh) 一种实现小波去噪的方法和装置
CN111383299A (zh) 图像的处理方法、装置及计算机可读存储介质
TWI460667B (zh) 模糊指紋影像重建方法
Parekh et al. Comparative study and analysis of medical image fusion techniques
JP6216987B6 (ja) 画像ノイズ除去方法および端末
Talukder et al. Enhancement of discrete wavelet transform (dwt) for image transmission over internet
Hsieh Dehazed Image Enhancement by a Gamma Correction with Global Limits

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: 14860150

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2016524563

Country of ref document: JP

Kind code of ref document: A

REEP Request for entry into the european phase

Ref document number: 2014860150

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 2014860150

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 15033550

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE