CN108665419B - Image denoising method and device - Google Patents
Image denoising method and device Download PDFInfo
- Publication number
- CN108665419B CN108665419B CN201710204551.3A CN201710204551A CN108665419B CN 108665419 B CN108665419 B CN 108665419B CN 201710204551 A CN201710204551 A CN 201710204551A CN 108665419 B CN108665419 B CN 108665419B
- Authority
- CN
- China
- Prior art keywords
- pixel point
- size
- filtering
- current pixel
- template
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000001914 filtration Methods 0.000 claims abstract description 122
- 230000014759 maintenance of location Effects 0.000 claims abstract description 54
- 238000013507 mapping Methods 0.000 claims abstract description 14
- 238000004321 preservation Methods 0.000 claims description 21
- 238000000354 decomposition reaction Methods 0.000 claims description 12
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 4
- 230000002146 bilateral effect Effects 0.000 description 3
- 238000013519 translation Methods 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
Abstract
A method and a device for denoising an image comprise the steps of taking a current pixel point to be processed as a center, selecting a template with the size of p × q from the image, averagely decomposing the template with the size of p × q into s modules with the size of m × n, filtering a low-frequency component of the current pixel point according to a preset filtering mode to obtain a filtering value, and reconstructing the modules including the current pixel point according to a high-frequency component, a high-frequency retention parameter and the filtering value corresponding to each pixel point in the modules including the current pixel point and with the size of m × n to obtain a reconstructed template with the size of p × q, wherein a preset mapping relation exists between the high-frequency retention parameter and edge information of the current pixel point.
Description
Technical Field
The embodiment of the invention relates to the field of image processing, in particular to a method and a device for image denoising.
Background
With the rapid development of communication and network technologies and the increasing popularity of various digital products, images have become important carriers for people to obtain external information. However, in the process of acquisition and transmission, digital images are often polluted by noise, so that before subsequent processing such as segmentation, encoding or beautification is performed on the images, the images often need to be denoised. Therefore, image denoising is a crucial link in the field of image processing and is a long-term research direction. How to consider the protection of the edge and the denoising effect is always a difficult point of denoising.
At present, some conventional denoising methods for protecting edges exist, such as Non-local mean filtering (NLM), bilateral filtering (bilateral), etc., which have a certain edge-protecting function, but they still cannot completely separate noise from edge details, so that it is still difficult to achieve balance between protecting detailed edges and denoising, for example, if denoising is too strong, detailed edges are easily damaged by mistake, and if detail edges are protected, noise remains.
In recent years, many image denoising methods are based on wavelets, and because wavelets have good time-frequency characteristics, multi-resolution and other characteristics, the wavelets can be gathered to any details of images, so that the image denoising method is very suitable for image denoising. Wavelets, however, also have their serious limitations. Because the wavelet transformation does not have translation invariance, a pseudo Gibbs phenomenon is easy to generate, which is shown on a denoised image, the edge of the image can be translated, the color of the edge can overflow, and the image quality is seriously influenced. Although some methods for denoising by using improved translation-invariant wavelet can suppress the pseudo-gibbs phenomenon, the calculation is too complex, and is not beneficial to hardware implementation.
Disclosure of Invention
The embodiment of the invention solves the problem of how to protect edge details to the maximum extent while ensuring the denoising strength, and has the advantages of simple calculation and convenient realization.
In order to solve the problems, the embodiment of the invention provides an image denoising method which comprises the steps of taking a current pixel point to be processed as a center, selecting a template with the size of p × q from the image, wherein both p and q are odd numbers, averagely decomposing the template with the size of p × q into s modules with the size of m × n, wherein s is a positive integer and is larger than 1, and s, m and n are odd numbers, filtering a low-frequency component of the current pixel point according to a preset filtering mode to obtain a filtering value, wherein the strength of the filtering is related to edge information of the current pixel point, reconstructing the module comprising the current pixel point according to a high-frequency component, a high-frequency retention parameter and the filtering value corresponding to each pixel point in the module with the size of m × n, and obtaining the reconstructed template with the size of p × q, wherein a preset mapping relation exists between the high-frequency retention parameter and the edge information of the current pixel point.
Optionally, the high-frequency component corresponding to each pixel point in the module with the size of m × n including the current pixel point is calculated by using the following formula:
hqij＝pij-c
wherein: ij characterizing the identity of the pixel, hqijCharacterizing the high-frequency component, p, corresponding to each pixel pointijAnd c, representing the pixel value of each pixel point in the module comprising the current pixel point, and representing the average pixel value of all pixel points in the module comprising the current pixel point.
Optionally, a preset mapping relationship exists between the high-frequency retention parameter and the edge information of the current pixel point, and the preset mapping relationship satisfies the following formula:
wherein: hq ofratioCharacterizing the high frequency retention parameter, edge characterizing the edge information, edge1 characterizing a first edge information threshold, edge2 characterizing a second edge information threshold, hq ratio1 is the first high frequency preservation parameter threshold, and the high frequency preservation parameter corresponding to edge2 is 1.
Optionally, the intensity of the filtering is related to the edge information of the current pixel point, and satisfies the following formula:
wherein: filter denotes the strength of the filtering, filter1 denotes the first strength threshold of the filtering, edge _ th denotes the third edge information threshold, edge denotes the edge information.
Optionally, the module including the current pixel point is subjected to pixel reconstruction according to the high-frequency component, the high-frequency retention parameter, and the filter value corresponding to each pixel point in the module with the size of m × n including the current pixel point, and is adapted to adopt the following formula:
p′ij＝c’+hqij*hqratio
wherein: ij denotes the identity, p ', of the pixel'ijCharacterizing the pixel value after pixel reconstruction, c' characterizing the filtered value, hqijCharacterizing said high frequency component, hqratioCharacterizing the high frequency preservation parameter.
Optionally, the method further comprises the steps of taking the current pixel point as a center, selecting a template with the size of a × b from the reconstructed template with the size of p × q, wherein a is less than p, b is less than q, and a and b are odd numbers, decomposing the template into s templates with the size of m1 × 0n1, filtering the low-frequency component of the current pixel point according to a preset filtering mode on the template with the size of m1 × n1 to obtain a filtering value, wherein m1 and n1 are odd numbers, performing pixel reconstruction on the module including the current pixel point according to a high-frequency component, a high-frequency retention parameter and the filtering value corresponding to each pixel point in the m1 × n1 module, obtaining a template with the size of a × b after reconstruction, selecting a template with the size of t × d from the template with the size of a × b after reconstruction, decomposing the template into s templates with the size of m 356 n 42, performing pixel decomposition on the m 3527 n templates including the pixel values of m2 × and n2 ×, and performing pixel decomposition on the current pixel point according to the filtering mode, wherein the m2 × and n2 × and the filtering values include the pixel values of the pixel points.
The embodiment of the invention provides an image denoising device which comprises a selecting unit, a decomposing unit, a filtering unit and a reconstructing unit, wherein the selecting unit is suitable for selecting a template with the size of p × q from an image by taking a current pixel point to be processed as a center, the decomposing unit is suitable for averagely decomposing the template with the size of p × q into s modules with the size of m × n, s is a positive integer and is larger than 1, and s, m and n are odd numbers, the filtering unit is suitable for filtering a low-frequency component of the current pixel point according to a preset filtering mode to obtain a filtering value, the strength of filtering is related to edge information of the current pixel point, the reconstructing unit is suitable for reconstructing the module comprising the current pixel point according to a high-frequency component, a high-frequency retention parameter and the filtering value corresponding to each pixel point in the module with the size of m × n, and obtains the reconstructed template with the size of p × q, and the preset mapping relation exists between the high-frequency retention parameter and the edge information of the current pixel point.
Optionally, the reconstructing unit is adapted to calculate the high-frequency component corresponding to each pixel point in the module with the size of m × n and including the current pixel point by using the following formula:
hqij＝pij-c
wherein: ij characterizing the identity of the pixel, hqijCharacterizing the high-frequency component, p, corresponding to each pixel pointijAnd c, representing the pixel value of each pixel point in the module comprising the current pixel point, and representing the average pixel value of all pixel points in the module comprising the current pixel point.
Optionally, a preset mapping relationship exists between the high-frequency retention parameter and the edge information of the current pixel point, and the preset mapping relationship satisfies the following formula:
wherein: hq ofratioCharacterizing the high frequency retention parameter, edge characterizing the edge information, edge1 characterizing a first edge information threshold, edge2 characterizing a second edge information threshold, hq ratio1 is the first high frequency preservation parameter threshold, and the high frequency preservation parameter corresponding to edge2 is 1.
Optionally, the intensity of the filtering is related to the edge information of the current pixel point, and satisfies the following formula:
wherein: filter denotes the strength of the filtering, filter1 denotes the first strength threshold of the filtering, edge _ th denotes the third edge information threshold, edge denotes the edge information.
Optionally, the reconstructing unit is adapted to perform pixel reconstruction on the module including the current pixel point according to the high-frequency component, the high-frequency retention parameter, and the filter value corresponding to each pixel point in the module with size m × n and including the current pixel point by using the following formula:
p′ij＝c’+hqij*hqratio；
wherein: ij denotes the identity, p ', of the pixel'ijCharacterizing the pixel value after pixel reconstruction, c' characterizing the filtered value, hqijCharacterizing said high frequency component, hqratioCharacterizing the high frequency preservation parameter.
Optionally, the selecting unit is further adapted to select a template with a size of a × b from the reconstructed template with a size of p × q by taking the current pixel point as a center, wherein a is smaller than p, b is smaller than q, and a and b are odd numbers, the decomposing unit is further adapted to decompose the template with the size of a × b into s templates with a size of m1 × n1, the filtering unit is further adapted to perform filtering on the template with the size of m1 × n1 according to a preset filtering mode to obtain a filtering value of the low-frequency component of the current pixel point, wherein m1 and n1 are odd numbers, the reconstructing unit is further adapted to perform pixel reconstruction on the module including the current pixel point according to a high-frequency component corresponding to each pixel point in the module with the size of m 8 n1 including the current pixel point, a high-frequency retention parameter and the filtering value, obtain a template with a × b after reconstruction, select t from the reconstructed template with the size of a × b, select t is smaller than t, the pixel point t is smaller than m 4629, the filtering parameter, the m 4623 and perform pixel reconstruction on the template after reconstruction, the m2, the m 4623 and the filtering parameter, the m 468 n 3, the pixel point is smaller than the pixel point, the pixel value of the current pixel point, the pixel point is the template, the pixel point, the reconstructed is the reconstructed pixel value of the reconstructed template, the reconstructed pixel point, the reconstructed template, the reconstructed value of the reconstructed template, the reconstructed pixel point of the.
Compared with the prior art, the technical scheme of the invention has the following advantages:
according to the scheme, the pixels are filtered according to a preset filtering mode, the high-frequency pixels are reconstructed according to the high-frequency components, the high-frequency retention parameters and the filtering values which comprise the current pixel points and correspond to the pixel points in the m × n module, low-frequency noise removal and high-frequency edge detail retention can be considered, and therefore balance between noise removal and edge detail retention can be achieved, the whole calculation and logic are simple, and operability is high, and implementation is facilitated.
Further, the relationship of linear change exists between the high-frequency retention parameter and the edge information of the current pixel point within a certain interval, so that the retention effect of high-frequency edge details can be improved.
Furthermore, the intensity of the filtering is related to the edge information of the current pixel point, that is, the intensity of the filtering is linearly decreased to a certain extent along with the decrease of the edge information, so that the effect of removing the low-frequency noise can be improved.
Drawings
FIG. 1 is a schematic structural diagram of an image denoising method in an embodiment of the present invention;
FIG. 2 is a graph illustrating the relationship between the strength of filtering and edge information in the practice of the present invention;
FIG. 3 is a flow chart illustrating another method for denoising an image in accordance with an embodiment of the present invention;
FIG. 4 is a schematic, exploded view of a template in the practice of the present invention;
FIG. 5 is a schematic diagram of a mean template in the practice of the present invention;
FIG. 6 is a graph of the relationship between the high frequency preservation parameter and the edge information in the implementation of the present invention;
FIG. 7 is a schematic structural diagram of an apparatus for denoising an image according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of an apparatus for denoising an image according to an embodiment of the present invention.
Detailed Description
As mentioned above, the current image denoising method is easy to generate a pseudo Gibbs phenomenon, which is shown in the denoised image, the image edge can shift, the color of the edge can overflow, and the image quality is seriously affected. Although some improved methods for denoising based on translation invariant wavelets can suppress the pseudo-gibbs phenomenon, the calculation is too complex, and hardware implementation is not facilitated.
In order to solve the above problems, in the embodiments of the present invention, the pixels are filtered according to a preset filtering manner, and the high-frequency pixels are reconstructed according to the high-frequency component, the high-frequency retention parameter, and the filtering value corresponding to each pixel point in the module with the size of m × n including the current pixel point, so that both low-frequency noise removal and retention of high-frequency edge details can be considered, and therefore, balance between noise removal and retention of edge details can be achieved.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in detail below.
Fig. 1 shows a schematic structural diagram of a method for denoising an image in an embodiment of the present invention, and the method is described in detail in steps with reference to fig. 1, and may include the following steps:
and step S11, selecting a template with the size of p × q from the image by taking the current pixel point to be processed as the center.
Generally, images to be processed are large, but a pixel point which is relatively related to a specific pixel point is only a pixel in a certain range around the specific pixel point, so in order to reduce the calculation amount and improve the image quality, in a specific implementation, a template with the size of p × q can be selected from the images by taking the current pixel point to be processed as the center.
Step S12, the template with the size of p × q is averagely decomposed into S modules with the size of m × n.
In order to improve the retention of the edge color of the image, in an implementation, the template with the size p × q can be decomposed into s modules with the size m × n on average, and then the modules with the size m × n are processed respectively, it should be noted that s can be a positive integer, and s > 1, and s, m and n are all odd numbers.
Step S13: and filtering the low-frequency component of the current pixel point according to a preset filtering mode to obtain a filtering value.
In a specific implementation, the strength of filtering may be related to the edge information of the current pixel point. In detail, the intensity of the filtering and the edge information of the current pixel point satisfy the following formula (1):
wherein: filter denotes the strength of the filtering, filter1 denotes the first strength threshold of the filtering, edge _ th denotes the third edge information threshold, edge denotes the edge information.
For easy understanding, the relationship between the filtering strength and the edge information of the current pixel point is shown in fig. 2, where the horizontal axis of fig. 2 represents the edge information edge and the vertical axis represents the filtering strength filter.
Referring to fig. 2, when the edge information edge is smaller than the third edge information threshold edge _ th, the filtering strength filter is linearly and continuously decreased as the edge information edge is increased; when the edge information edge is greater than the third edge information threshold edge _ th, the edge information edge is kept unchanged, so that the denoising of the non-edge region can be improved, the denoising of the edge region is weakened, the low-frequency noise is removed, and the effect of retaining the edge information is achieved.
And S14, reconstructing the module comprising the current pixel point according to the high-frequency component, the high-frequency retention parameter and the filtering value corresponding to each pixel point in the module with the size of m × n, and obtaining a reconstructed template with the size of p × q, wherein a preset mapping relation exists between the high-frequency retention parameter and the edge information of the current pixel point.
In an embodiment of the present invention, the following formula (2) may be adopted to calculate the high frequency component corresponding to each pixel point in the module with the size of m × n and including the current pixel point:
hqij＝pij-c (2)
wherein: ij characterizing the identity of the pixel, hqijCharacterizing the high-frequency component, p, corresponding to each pixel pointijAnd c, representing the pixel value of each pixel point in the module comprising the current pixel point, and representing the average pixel value of all pixel points in the module comprising the current pixel point.
In an embodiment of the present invention, a preset mapping relationship between the high-frequency retention parameter and the edge information of the current pixel may satisfy the following formula (3):
wherein: hq ofratioCharacterizing the high frequency retention parameter, edge characterizing the edge information, edge1 characterizing a first edge information threshold, edge2 characterizing a second edge information threshold, hq ratio1 is the first high frequency preservation parameter threshold, and the high frequency preservation parameter corresponding to edge2 is 1.
In a specific implementation, the following formula (4) may be adopted to perform pixel reconstruction on the module including the current pixel point according to the high-frequency component, the high-frequency retention parameter, and the filter value corresponding to each pixel point in the module with size m × n and including the current pixel point:
pij′＝c’+hqij*hqratio(4)
wherein: ij denotes the identity, p ', of the pixel'ijCharacterizing the pixel value after pixel reconstruction, c' characterizing the filtered value, hqijCharacterizing said high frequency component, hqratioCharacterizing the high frequency preservation parameter.
In a specific implementation, after the reconstruction of the time is completed, a template with the size of a × b can be selected from the reconstructed templates with the size of p × q by taking the current pixel point as the center, wherein a is less than p, b is less than q, a and b are odd numbers and are decomposed into s templates with the size of m1 × n1, and the low-frequency component of the current pixel point is filtered by the template with the size of m1 × n1 according to a preset filtering mode to obtain a filtering value, wherein m1 and n1 are odd numbers.
And performing pixel reconstruction on the module comprising the current pixel point according to a high-frequency component, a high-frequency retention parameter and the filtering value corresponding to each pixel point in the module m1 × n1 comprising the current pixel point to obtain a reconstructed template with the size of a × b, selecting a template with the size of t × d from the reconstructed template with the size of a × b, decomposing the template into s templates with the size of m2 × n2, and filtering the low-frequency component of the current pixel point according to a preset filtering mode on the template with the size of m2 × n2 to obtain a filtering value, wherein t is less than a, d is less than b, and m2, n2, t and d are odd numbers.
In other words, after the reconstructed image, a smaller template is selected and decomposed again, the processing of steps S12 to S14 is repeated each time by using the template obtained after decomposition again until the image is decomposed or layered to the last layer, so that the decomposition cannot be continued again, and the operation is performed on all the pixel points, and the final result value is used as the result after image denoising.
At present, some improved methods for denoising images based on translation invariant wavelets exist, but the calculation is too complex, and hardware implementation is not facilitated.
In the embodiment of the invention, the pixels are filtered according to a preset filtering mode, the high-frequency pixels are reconstructed according to the high-frequency components, the high-frequency retention parameters and the filtering values which comprise the current pixel points and correspond to each pixel point in the module m × n, and low-frequency noise removal and high-frequency edge detail retention can be considered, so that the balance between the noise removal and the edge detail retention can be achieved, the whole calculation and logic are simple, and the operability is strong, so that the realization is convenient.
For a better understanding and realization of the present invention by those skilled in the art, fig. 3 shows a flow chart of another image denoising method in the implementation of the present invention, where the template size selected from the image is 15 × 15, and the template can be other sizes, and is not limited herein, it should be noted that if the color noise is severe, the template size is large to be effectively processed, and the method will be described in detail with reference to fig. 3:
step S31: and taking the current pixel point to be processed as a central point in the template, and performing first-layer decomposition on the selected template.
It should be noted that the image to be processed may be referred to as an image of layer0, and the image after one layer decomposition may be an image of layer 1.
In specific implementation, the current pixel point to be processed may be used as a central point in the template, and the selected template may be subjected to a first-layer decomposition, and various decomposition methods may be specifically adopted.
In detail, as shown in fig. 4, the 15 × 15 image is divided equally into 3 × 3 templates with the size of 5 × 5, and the template of the mean layer1 obtained by calculating the mean value of the 5 × 5 template in each 5 × 5 template is shown in fig. 5, and the central pixel point is m 11.
Step S32: calculate the high frequency component on layer 0:
for the template 5 × 5 where the center pixel point m11 is located, which may be referred to as a center template, in a specific implementation, the calculation of the high-frequency component on layer0, that is, the calculation of the high-frequency component of each pixel pij in the center template, may specifically adopt formula (5):
hqij＝pij-m11(5)
step S33: and calculating the edge information of the current pixel point.
If the current pixel point is in a flat area, the obtained high-frequency component is likely to be noise and needs to be eliminated. If the current pixel point is in the edge area, the obtained high-frequency component is likely to be real information and needs to be reserved. Therefore, in order to determine the retention coefficient of the high-frequency component, in a specific implementation, the edge information edge of the current pixel point may be calculated first to determine the adaptively-changing high-frequency retention coefficient, and the edge information edge may be calculated by using equations (6) to (10):
edgex1＝abs((2*m01+m11+m21)-(m00+m20+m10)-(m02+m12+m22)) (6)
edgey1＝abs((2*m10+m11+m12)-(m00+m02+m01)-(m20+m21+m22)) (7)
edgex2＝abs(m00+m10+m20-m02-m12-m22) (8)
edgey2＝abs(m00+m10+m02-m20-m21-m22) (9)
edge＝max(edgex1,edgey1,edgex2,edgey2) (10)
among them, edgex1 and edgey1 may be used to compute narrow edges, and edgex2 and edgey2 may be used to compute wide edges.
Step S34: and obtaining a high-frequency retention parameter according to a preset mapping relation.
FIG. 6 shows a relationship between the high frequency preservation parameter and the edge information, where the horizontal axis represents the edge information edge and the vertical axis represents the high frequency preservation parameter hqratio. It should be noted that the curve is not unique, and may be adjusted according to actual requirements, but the principle that the high frequency is reserved more when the edge information edge is larger, that is, the high frequency reservation parameter is larger may be followed; when the edge information edge is smaller, the high frequency reservation is less, i.e. the high frequency reservation parameter is smaller;
step S35: and (5) low-frequency filtering.
In a specific implementation, the mean template shown in fig. 5 may be used to perform bilateral filtering or gaussian filtering on the center point m11 to obtain m 11'. Wherein the filtering strength can be adaptively adjusted according to the curve shown in fig. 2.
Step S36: high frequency add back on layer 0.
In a specific implementation, the high frequency of the small template at center 5 × 5 can be added back at layer0 using equation (11):
p′ij＝m′11+hqij*hqratio(11)
step S37, decompose again on reconstructed layer0 to take the template of 9 × 9 around the center point.
In a specific implementation, similar to step S31, the template with size of 9 × 9 may be divided into 3 × 3 small templates with size of 3 × 3, as shown in fig. 7, and further, the average value in each small template may be calculated respectively, and filtering and reconstruction on layer1 are performed according to the methods of steps S32 to S36.
Step S38, decompose again on reconstructed layer0 to take a template of 3 × 3 around the center point.
In an embodiment of the present invention, the template with a size of 3 × 3 is the last layer, and can not be decomposed any more, so step S31 is not needed, the average value in each small template can be calculated respectively, filtering and reconstruction are performed on layer0 according to the methods of steps S32 to S36, and the final result is used as the denoising result value.
In summary, the embodiment of the method has better effects on removing low-frequency noise and retaining high-frequency edge details through low-frequency filtering and high-frequency adaptive add-back under different scales, and can better achieve the balance between noise removal and detail retention. The invention has the advantages of simple calculation and logic, strong operability and better practicability.
In order to make those skilled in the art better understand and implement the present invention, fig. 8 shows a schematic structural diagram of an apparatus for denoising an image in an embodiment of the present invention, as shown in fig. 8, the apparatus may include: a selecting unit 81, a decomposing unit 82, a filtering unit 83 and a reconstructing unit 84, wherein:
the selecting unit 81 is suitable for selecting a template with the size of p × q from the image by taking the current pixel point to be processed as the center, wherein p and q are both odd numbers;
a decomposition unit 82, adapted to decompose the template with size p × q into s modules with size m × n, wherein s is a positive integer and s > 1, and s, m and n are odd numbers;
the filtering unit 83 is adapted to filter the low-frequency component of the current pixel point according to a preset filtering manner to obtain a filtering value; wherein: the intensity of the filtering is related to the edge information of the current pixel point;
the reconstruction unit 84 is adapted to reconstruct the module including the current pixel point according to the high-frequency component, the high-frequency retention parameter and the filter value corresponding to each pixel point in the module with the size of m × n, so as to obtain a reconstructed template with the size of p × q, wherein a preset mapping relationship exists between the high-frequency retention parameter and the edge information of the current pixel point.
To sum up, the filtering unit 83 according to the embodiment of the present invention filters the pixels according to a preset filtering manner, and the reconstructing unit 84 reconstructs the high-frequency pixels according to the high-frequency component, the high-frequency retention parameter, and the filtering value corresponding to each pixel point in the m × n module including the current pixel point, so as to remove low-frequency noise and retain high-frequency edge details, thereby achieving balance between noise removal and edge detail retention, and the whole calculation and logic are simple, and have strong operability, so that the implementation is facilitated.
In a specific implementation, the reconstructing unit 84 is adapted to calculate the high-frequency component corresponding to each pixel point in the module with the size of m × n and including the current pixel point by using the following formula:
hqij＝pij-c
wherein: ij characterizing the identity of the pixel, hqijCharacterizing the high-frequency component, p, corresponding to each pixel pointijAnd c, representing the pixel value of each pixel point in the module comprising the current pixel point, and representing the average pixel value of all pixel points in the module comprising the current pixel point.
In specific implementation, a preset mapping relationship exists between the high-frequency retention parameter and the edge information of the current pixel point, and the following formula is satisfied:
wherein: hq ofratioCharacterizing the high frequency retention parameter, edge characterizing the edge information, edge1 characterizing a first edge information threshold, edge2 characterizing a second edge information threshold, hq ratio1 is the first high frequency preservation parameter threshold, and the high frequency preservation parameter corresponding to edge2 is 1.
In a specific implementation, the filtering strength is related to the edge information of the current pixel point, and satisfies the following formula:
wherein: filter denotes the strength of the filtering, filter1 denotes the first strength threshold of the filtering, edge _ th denotes the third edge information threshold, edge denotes the edge information.
In a specific implementation, the reconstructing unit 84 is adapted to perform pixel reconstruction on the module including the current pixel point according to the high-frequency component, the high-frequency retention parameter, and the filter value corresponding to each pixel point in the module with size m × n and including the current pixel point, by using the following formula:
p′ij＝c’+hqij*hqratio
wherein: ij denotes the identity, p ', of the pixel'ijCharacterizing the pixel value after pixel reconstruction, c' characterizing the filtered value, hqijCharacterizing said high frequency component, hqratioCharacterizing the high frequency preservation parameter.
In a specific implementation, the selecting unit 81 is further adapted to select a template with a size of a × b from the reconstructed templates with the size of p × q, with the current pixel point as a center, wherein a is less than p, b is less than q, and a and b are odd numbers;
the decomposition unit 82 is further adapted to decompose the template with the size a × b into s templates with the size m1 × n1, and the filtering unit 83 is further adapted to filter the low-frequency component of the current pixel point according to a preset filtering mode on the template with the size m1 × n1 to obtain a filtering value, wherein m1 and n1 are odd numbers;
the reconstruction unit 84 is further adapted to perform a pixel reconstruction operation on the module including the current pixel point according to a high-frequency component, a high-frequency retention parameter and the filtering value corresponding to each pixel point in the module m1 × n1 including the current pixel point to obtain a template with a reconstructed size of a × b, select a template with a size of t × d from the template with the reconstructed size of a × b, decompose the template into s templates with a size of m2 × n2, filter a low-frequency component of the current pixel point according to a preset filtering mode on the template with the size of m2 × n2 to obtain a filtering value, wherein t is smaller than a, d is smaller than b, m2, n2, t and d are odd numbers, and perform a pixel reconstruction operation on the module including the current pixel point according to the high-frequency component, the high-frequency retention parameter and the filtering value corresponding to each pixel point in the module m2 × n2 including the current pixel point until the pixel point cannot be decomposed again.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by hardware related to instructions of a program, which may be stored in a computer-readable storage medium, and the storage medium may include: ROM, RAM, magnetic or optical disks, and the like. Although the present invention is disclosed above, the present invention is not limited thereto. Various changes and modifications may be effected therein by one skilled in the art without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (10)
1. A method for denoising an image, comprising:
taking a current pixel point to be processed as a center, and selecting a template with the size of p × q from the image, wherein p and q are both odd numbers;
the template with the size of p × q is averagely decomposed into s modules with the size of m × n, wherein s is a positive integer and is more than 1, and s, m and n are odd numbers;
filtering the low-frequency component of the current pixel point according to a preset filtering mode to obtain a filtering value; wherein: the intensity of the filtering is related to the edge information of the current pixel point;
reconstructing the module comprising the current pixel point according to the high-frequency component, the high-frequency retention parameter and the filtering value which comprise the current pixel point and correspond to each pixel point in the module m × n in size, and obtaining a reconstructed template p × q in size, wherein a preset mapping relation exists between the high-frequency retention parameter and the edge information of the current pixel point;
the preset relationship satisfies the following formula:
wherein: hq ofratioCharacterizing the high frequency retention parameter, edge characterizing the edge information, edge1 characterizing a first edge information threshold, edge2 characterizing a second edge information threshold, hqratio1 is the first high frequency preservation parameter threshold, and the high frequency preservation parameter corresponding to edge2 is 1.
2. The method of denoising an image according to claim 1, wherein the high frequency component corresponding to each pixel point in a module of size m × n including the current pixel point is calculated using the following formula:
hqij＝pij-c
wherein: ij characterizing the identity of the pixel, hqijCharacterizing the high-frequency component, p, corresponding to each pixel pointijAnd c, representing the pixel value of each pixel point in the module comprising the current pixel point, and representing the average pixel value of all pixel points in the module comprising the current pixel point.
3. The method of image denoising of claim 1, wherein the intensity of the filtering is related to the edge information of the current pixel, and satisfies the following formula:
wherein: filter denotes the strength of the filtering, filter1 denotes the first strength threshold of the filtering, edge _ th denotes the third edge information threshold, edge denotes the edge information.
4. A method of denoising an image according to claim 1, wherein said reconstructing the block including the current pixel point according to the high frequency component, the high frequency preserving parameter and the filter value corresponding to each pixel point in the block of m × n size including the current pixel point is adapted to adopt the following formula:
p′ij＝c’+hqij*hqratio
wherein: ij denotes the identity, p ', of the pixel'ijCharacterizing the pixel value after pixel reconstruction, c' characterizing the filtered value, hqijCharacterizing said high frequency component, hqratioCharacterizing the high frequency preservation parameter.
5. The method of image denoising of any one of claims 1-4, further comprising:
selecting a template with the size of a × b from the reconstructed templates with the size of p × q by taking the current pixel point as the center, wherein a is less than p, b is less than q, and a and b are odd numbers;
decomposing the template into s templates with the size of m1 × n1, and filtering the low-frequency components of the current pixel points according to a preset filtering mode on the templates with the size of m1 × n1 to obtain filtering values, wherein m1 and n1 are odd numbers;
according to the high-frequency component, the high-frequency retention parameter and the filtering value corresponding to each pixel point in the m1 × n1 module comprising the current pixel point, carrying out pixel reconstruction on the module comprising the current pixel point to obtain a reconstructed template with the size of a × b, selecting a template with the size of t × d from the reconstructed template with the size of a × b, decomposing the template into s templates with the size of m2 × n2, and filtering the low-frequency component of the current pixel point according to a preset filtering mode on the template with the size of m2 × n2 to obtain a filtering value, wherein t is less than a, d is less than b, and m2, n2, t and d are odd numbers;
and according to the high-frequency component, the high-frequency retention parameter and the filtering value corresponding to each pixel point in the module with the size of the current pixel point m2 × n2, carrying out pixel reconstruction on the module with the current pixel point until the module can not be decomposed any more.
6. An apparatus for denoising an image, comprising:
the selecting unit is suitable for selecting a template with the size of p × q from the image by taking the current pixel point to be processed as the center, wherein p and q are both odd numbers;
the decomposition unit is suitable for decomposing the template with the size of p × q into s modules with the size of m × n on average, wherein s is a positive integer and is more than 1, and s, m and n are odd numbers;
the filtering unit is suitable for filtering the low-frequency component of the current pixel point according to a preset filtering mode to obtain a filtering value; wherein: the intensity of the filtering is related to the edge information of the current pixel point;
the reconstruction unit is suitable for reconstructing the module comprising the current pixel point according to the high-frequency component, the high-frequency retention parameter and the filtering value which comprise the current pixel point and correspond to each pixel point in the module with the size of m × n to obtain a reconstructed template with the size of p × q, wherein a preset mapping relation exists between the high-frequency retention parameter and the edge information of the current pixel point;
the preset relationship satisfies the following formula:
wherein: hq ofratioCharacterizing the high frequency guardLeave parameters, edge characterizing the edge information, edge1 characterizing a first edge information threshold, edge2 characterizing a second edge information threshold, hqratio1 is the first high frequency preservation parameter threshold, and the high frequency preservation parameter corresponding to edge2 is 1.
7. The apparatus for denoising the image according to claim 6, wherein the reconstructing unit is adapted to calculate the high frequency component corresponding to each pixel point in the module with size m × n, which includes the current pixel point, by using the following formula:
hqij＝pij-c
wherein: ij characterizing the identity of the pixel, hqijCharacterizing the high-frequency component, p, corresponding to each pixel pointijAnd characterizing the pixel value of each pixel point in the module of the current pixel point, and characterizing the average pixel value of all pixel points in the module including the current pixel point.
8. The apparatus for denoising image according to claim 6, wherein the intensity of the filtering is related to the edge information of the current pixel, and satisfies the following formula:
wherein: filter denotes the strength of the filtering, filter1 denotes the first strength threshold of the filtering, edge _ th denotes the third edge information threshold, edge denotes the edge information.
9. The apparatus for denoising the image according to claim 6, wherein the reconstructing unit is adapted to apply the following formula to reconstruct the pixel of the block including the current pixel according to the high frequency component, the high frequency preserving parameter and the filtered value corresponding to each pixel point in the block of m × n size including the current pixel:
p′ij＝c’+hqij*hqratio；
wherein: ij denotes the identity, p ', of the pixel'ijCharacterizing the pixel value after pixel reconstruction, c' characterizing the filtered value, hqijCharacterizing said high frequency component, hqratioCharacterizing the high frequency preservation parameter.
10. The apparatus for denoising the image according to any one of claims 6 to 9, wherein the selecting unit is further adapted to select a template with a size of a × b from the reconstructed templates with a size of p × q, with a < p, b < q, and a and b being odd numbers, centering on the current pixel point;
the decomposition unit is also suitable for decomposing the template with the size a × b into s templates with the size m1 × n 1;
the filtering unit is also suitable for filtering the low-frequency components of the current pixel points according to a preset filtering mode on the template with the size of m1 × n1 to obtain a filtering value, wherein m1 and n1 are odd numbers;
the reconstruction unit is further suitable for performing pixel reconstruction on the module comprising the current pixel point according to a high-frequency component, a high-frequency retention parameter and the filtering value corresponding to each pixel point in the module m1 × n1 comprising the current pixel point to obtain a reconstructed template with the size of a × b, selecting a template with the size of t × d from the reconstructed template with the size of a × b, decomposing the template into s templates with the size of m2 × n2, filtering the low-frequency component of the current pixel point according to a preset filtering mode on the template with the size of m2 × n2 to obtain a filtering value, wherein t is smaller than a, d is smaller than b, m2, n2, t and d are odd numbers, and performing pixel reconstruction on the module comprising the current pixel point according to the high-frequency component, the high-frequency retention parameter and the filtering value corresponding to each pixel point in the module m2 × n2 comprising the current pixel point until the reconstruction cannot be decomposed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710204551.3A CN108665419B (en) | 2017-03-30 | 2017-03-30 | Image denoising method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710204551.3A CN108665419B (en) | 2017-03-30 | 2017-03-30 | Image denoising method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108665419A CN108665419A (en) | 2018-10-16 |
| CN108665419B true CN108665419B (en) | 2020-07-03 |
Family
ID=63785642
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710204551.3A Active CN108665419B (en) | 2017-03-30 | 2017-03-30 | Image denoising method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108665419B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685730B (en) * | 2018-12-11 | 2023-06-02 | 上海集成电路研发中心有限公司 | Wavelet denoising method based on self-adaptive non-local mean value |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101957984A (en) * | 2010-08-30 | 2011-01-26 | 西安电子科技大学 | Image de-noising method based on parametric estimation of non-local shrinkage factor |
| CN101984461A (en) * | 2010-11-05 | 2011-03-09 | 西安电子科技大学 | Method for denoising statistical model image based on controllable pyramid |
| CN103345726A (en) * | 2013-06-14 | 2013-10-09 | 华为技术有限公司 | Image de-noising processing method, device and terminal |
| CN104766289A (en) * | 2015-03-20 | 2015-07-08 | 华南理工大学 | Image denoising and compressing method |
| CN106023112A (en) * | 2016-05-24 | 2016-10-12 | 中国科学院深圳先进技术研究院 | Image rain removing method and system based on wavelet analysis |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150363912A1 (en) * | 2014-06-12 | 2015-12-17 | Samsung Electronics Co., Ltd. | Rgbw demosaic method by combining rgb chrominance with w luminance |
-
2017
- 2017-03-30 CN CN201710204551.3A patent/CN108665419B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101957984A (en) * | 2010-08-30 | 2011-01-26 | 西安电子科技大学 | Image de-noising method based on parametric estimation of non-local shrinkage factor |
| CN101984461A (en) * | 2010-11-05 | 2011-03-09 | 西安电子科技大学 | Method for denoising statistical model image based on controllable pyramid |
| CN103345726A (en) * | 2013-06-14 | 2013-10-09 | 华为技术有限公司 | Image de-noising processing method, device and terminal |
| CN104766289A (en) * | 2015-03-20 | 2015-07-08 | 华南理工大学 | Image denoising and compressing method |
| CN106023112A (en) * | 2016-05-24 | 2016-10-12 | 中国科学院深圳先进技术研究院 | Image rain removing method and system based on wavelet analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108665419A (en) | 2018-10-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Knaus et al. | Progressive image denoising | |
| CN110889812B (en) | Underwater image enhancement method for multi-scale fusion of image characteristic information | |
| Rahman et al. | Gaussian noise reduction in digital images using a modified fuzzy filter | |
| Dewangan et al. | Image smoothening and sharpening using frequency domain filtering technique | |
| CN113724164A (en) | Visible light image noise removing method based on fusion reconstruction guidance filtering | |
| Makandar et al. | Computation pre-processing techniques for image restoration | |
| CN108665419B (en) | Image denoising method and device | |
| CN110175959B (en) | Typhoon cloud picture enhancement method | |
| Shetter et al. | Image de-noising algorithm based on filtering and histogram equalization | |
| Ono et al. | Optimized JPEG image decompression with super-resolution interpolation using multi-order total variation | |
| Sun et al. | Color image denoising based on guided filter and adaptive wavelet threshold | |
| Adhikari et al. | Image enhancement using successive mean quantization transform and homomorphic filtering | |
| Rajkumar et al. | An efficient image denoising approach for the recovery of impulse noise | |
| Charde | A review on image denoising using wavelet transform and Median filter over AWGN channel | |
| Wen et al. | Modified block-matching 3-D filter in Laplacian pyramid domain for speckle reduction | |
| Gerace et al. | An inpainting technique based on regularization to remove bleed-through from ancient documents | |
| Hanumantharaju et al. | Natural color image enhancement based on modified multiscale retinex algorithm and performance evaluation using wavelet energy | |
| Tyagi et al. | Comparative study of image enhancement and analysis of thermal images using image processing and wavelet techniques | |
| Siddeq et al. | Using Discrete Wavelet Transform and Wiener filter for Image De-nosing: Discrete Wavelet Transform | |
| Dhiman et al. | An improved threshold estimation technique for image denoising using wavelet thresholding techniques | |
| Singh | COMPARATIVE STUDY OF ALGORITHMS/TECHNIQUES FOR DENOISING OF GAUSSIAN NOISE. | |
| Ehsaeyan | A new shearlet hybrid method for image denoising | |
| Singh et al. | Study of various image enhancement techniques-a review | |
| Hsia et al. | Adaptive Wavelet Shrinkage Based On Intelligent FIS Learned Thresholding | |
| Sankaran et al. | Patch based image restoration using adaptive bilateral filtering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |