WO2015135208A1 - 对图像进行噪声抑制的方法及装置 - Google Patents

对图像进行噪声抑制的方法及装置 Download PDF

Info

Publication number
WO2015135208A1
WO2015135208A1 PCT/CN2014/073461 CN2014073461W WO2015135208A1 WO 2015135208 A1 WO2015135208 A1 WO 2015135208A1 CN 2014073461 W CN2014073461 W CN 2014073461W WO 2015135208 A1 WO2015135208 A1 WO 2015135208A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
noise
periodic noise
periodic
constraint condition
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.)
Ceased
Application number
PCT/CN2014/073461
Other languages
English (en)
French (fr)
Inventor
赖勇铨
邹耀贤
林穆清
许�鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Mindray Bio Medical Electronics Co Ltd
Original Assignee
Shenzhen Mindray Bio Medical Electronics Co Ltd
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 Shenzhen Mindray Bio Medical Electronics Co Ltd filed Critical Shenzhen Mindray Bio Medical Electronics Co Ltd
Priority to PCT/CN2014/073461 priority Critical patent/WO2015135208A1/zh
Priority to CN201480077057.XA priority patent/CN106464778B/zh
Publication of WO2015135208A1 publication Critical patent/WO2015135208A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00Details of television systems
    • H04N5/14Picture signal circuitry for video frequency region
    • H04N5/21Circuitry for suppressing or minimising disturbance, e.g. moiré or halo

Definitions

  • the present application relates to image processing techniques, and more particularly to a method and apparatus for noise suppression of an image, in particular, the image is an image that is interfered by periodic noise.
  • Background technique
  • Digital Radiography is a digital sensor
  • X-ray technology for photographing medical photographs which digitally stores and processes medical images, has the advantages of fast collection and transmission speed, easier image enhancement and display, etc., and has been widely used in various medical examinations and diagnoses.
  • DR products add an anti-scatter grid between the X-ray receiver and the human body.
  • the function of the grid is to allow the X-rays passing through the human body to pass through the front side, so that the X-rays scattered from the human body are filtered out, so as to avoid the same photosensitive point on the flat plate from receiving different tissues from the human body.
  • X-ray signals to improve image contrast and spatial resolution.
  • the spatial frequency of the grid is not an integer multiple of the spatial sampling frequency of the image, streaky noise is displayed on the image, which is a grating 13 as shown in Fig. 1, which is usually rendered as periodic noise. This phenomenon is called the Moir effect; the resulting streaks are also called moiré stripes or moiré.
  • method 1) and method 2) have higher requirements on the structure and installation of the grid
  • method 3) reduces the structure and installation method of the grid, but needs to remove the streaks generated during imaging, and the requirements cannot be Affected by defects in the imaging plate itself (such as bad lines, bad spots).
  • a method for performing noise suppression on an image includes: an image acquisition step: acquiring an image to be processed, the image to be processed being an image interfered by periodic noise;
  • a period determining step determining a period of periodic noise in the image to be processed
  • the optimization solving step substituting the image data of the image to be processed and the period of the periodic noise into a signal separation model, and optimally solving the signal separation model to obtain a clean image signal corresponding to the image to be processed.
  • an apparatus for performing noise suppression on an image includes: an image acquisition module, configured to acquire an image to be processed, where the image to be processed is an image that is interfered by periodic noise;
  • a period determining step determining a period of periodic noise in the image to be processed
  • the optimization solving step substituting the image data of the image to be processed and the period of the periodic noise into a signal separation model, and optimally solving the signal separation model to obtain a clean image signal corresponding to the image to be processed.
  • the method for performing noise suppression on an image of the present invention is implemented by a signal separation model, and when the model is used, processing is performed in combination with a period of the determined periodic noise, so that the periodic noise signal and the non-periodic noise signal are separated.
  • the three components of the clean image signal can take into account the periodic effects of noise, so that the clean image does not contain the ringing effect, thereby improving the image quality.
  • Fig. 1 is a schematic diagram of the generation of the astigmatism in the operation of the anti-scatter grid, wherein 11 is an X-ray receiving plate, 12 is a grid, 13 is a grating, short arrow 14 is a straight ray, and long arrow 15 is a scattering line;
  • Figure 2 and Figure 3 are schematic diagrams of the bad line image generated by the bad line ringing and the low-pass filtered image.
  • FIG. 4 is a flow chart showing a method for suppressing a DR image grid shadow with a wire grid according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram of decomposing an input DR image signal into three signal components according to an embodiment of the present invention
  • FIG. 6 is a schematic diagram showing an example of a DCT frequency domain and its filter coefficients in an embodiment of the present invention
  • FIG. 7 is a signal component shown in FIG. Schematic diagram
  • FIG. 8 is a schematic flowchart of a gate shadow direction detecting algorithm according to an embodiment of the present invention.
  • 9a and 9b are schematic diagrams of segmentation of an image to be processed in an embodiment of the present invention. detailed description
  • the present invention proposes a noise suppression method for an image, in particular, the image is an image that is interfered by periodic noise.
  • the method can be applied to digital radiographic images (referred to as DR images) including, but not limited to, with anti-scatter grids.
  • DR images digital radiographic images
  • the method of performing noise suppression on the image of the present invention will be described in detail by taking the shadow suppression of the DR image with the filter grid (that is, the periodic noise as the gate shadow). It should be understood that the following method can be applied to various methods. An image that needs to suppress periodic noise.
  • the method for performing noise suppression on an image proposed by the embodiment of the present invention is based on signal processing in such a situation: a clean signal X is subjected to periodic noise; Additive interference, and the introduction of sparse noise s (also known as non-periodic noise) due to sensor defects during observation, requires the estimation of the three components of ⁇ , and X from the sensor output ⁇ .
  • this embodiment proposes a gate shadow suppression method with a filter grid DR image, by which the pair of images proposed by the present invention (especially an image interfered by periodic noise such as a gate shadow) is embodied.
  • the general procedure for noise suppression is to first determine the gate shadow period and then perform gate shadow suppression (the solid line portion in the figure).
  • gate shadow suppression can be performed.
  • the obtaining of the SPX model may be constructed by the following method; or directly calling a previously stored model, and the stored model may be constructed in advance according to the following method.
  • the SPX model constructed here includes a cost function and a constraint, wherein the cost function is a function that satisfies the constraint and is related to at least one of a clean signal component, a periodic signal component, and a non-periodic signal component, the clean signal component being about a clean image signal
  • the cost function is a function that satisfies the constraint and is related to at least one of a clean signal component, a periodic signal component, and a non-periodic signal component, the clean signal component being about a clean image signal
  • the measure of smoothness, the periodic signal component is a measure of the periodic noise and its period, and the aperiodic signal component is a measure of the non-periodic noise.
  • the expression of the clean signal component is the expression of the periodic signal component is e 2
  • the expression of the non-periodic signal component is e3 g( w )4
  • the constraint condition includes a fixed constraint condition, or includes a fixed condition and Specify a constraint where the fixed constraint is
  • the cost function expression of this embodiment is:
  • the cost term in the cost function may become a constraint term, for example:
  • the cost function may be represented as ⁇
  • the cost function can be expressed as e 3 , and the fixed constraint and the specified constraint are satisfied, the specified constraint is ei
  • the cost function can be expressed as
  • the specified constraint is e 3 ⁇ r 2 ;
  • the cost function is , and the fixed constraint and the specified constraint are satisfied, and the specified constraint is 3 ⁇ 4 ⁇ ; or, the cost function is And satisfying the fixed constraint and the specified constraint, the specified constraint is e x
  • the measure functions of , and X are used or norm, such as the norm for the S term (ie, the non-periodic noise term), and the X term (ie, the clean image signal term);
  • the term (i.e., periodic noise term) 2 uses 2 norm; of course, in other embodiments, other obvious approximations or equivalent measures may be used instead of norms and/or norms.
  • the norm can be used for the ⁇ term (ie, the periodic noise term), and the norm can be used for the X term (ie, the clean image signal term) and the S term (ie, the non-periodic noise term), or it can be the pair S.
  • item preclude the use of the norm, X-item (i.e., item a clean image signal), and;? 2-norm with items (i.e., items of periodic noise) Bian, etc., and so on.
  • f is the number of segments set to segment the periodic noise
  • is the period of each periodic noise
  • ⁇ 0
  • ⁇ . is the number of ⁇ in each piece of periodic noise.
  • the optimized solution of the constructed SPX model is to solve the following mathematical optimization problem, thereby achieving grid shadow suppression:
  • the SPX model minimizes the cost function as the optimization target.
  • the clean image signal is obtained by solving the cost function. Periodic noise and aperiodic noise. At this time, the cost function is, then the math
  • the X and U terms use the 2 norm.
  • the 2 norm is calculated by taking the sum of the squares of the elements of the vector and then finding the square root.
  • e R" is an input signal (ie, an input DR image to be subjected to noise suppression processing)
  • s is a non-periodic noise
  • X is a clean signal.
  • is a weighted DCT (Discrete Cosine Transform) matrix, the purpose is to filter, that is, the component of a certain frequency is suppressed to minimize the X term, and its calculation method is shown in equation (5).
  • Q diag(c)Q ( 5 )
  • Q is the DCT amplitude spectrum of the input signal (ie, the DR image to be subjected to noise suppression processing) (which can be obtained by DCT transforming the image to obtain the amplitude spectrum)
  • ce R is a weight vector (or filter coefficient vector).
  • the shape of c is set to one or more peaks, corresponding to the peaks due to periodic noise on the DCT spectrum, as shown in Figure 6. The larger the coefficient, the greater the suppression of periodic noise; when the filter coefficient is 0, the corresponding spectrum will not be attenuated.
  • " c is the number of filter coefficients not equal to zero, which is equivalent to the number of rows of the matrix ⁇ .
  • is a weighted DCT matrix
  • 3 ⁇ 4 may also be, for example, a weighted DFT (Discrete Fourier Transform) matrix or an FFT (Fast Fourier Transform) matrix.
  • U Discrete Fourier Transform
  • FFT Fast Fourier Transform
  • each signal can be decomposed into f segments, each segment being represented by a fundamental wave, the length of the fundamental wave being ⁇ , repeating T Q times, ⁇ is the period of the gate shadow, can pass The empirical value is obtained or calculated by a subsequent estimation algorithm for the gate shadow period. The purpose of this segmentation is to increase robustness.
  • u [u , u 2 T ,...,u ,...u K TT GR Kt , where . eR,
  • weR is the input weight, which can be reduced or increased by increasing or decreasing the value of ⁇ .
  • the method for performing noise suppression on an image subjected to periodic noise interference in this embodiment is implemented based on an SPX model, and a periodic noise signal, a non-periodic noise signal, and a clean image signal can be separated by the model.
  • a periodic noise signal, a non-periodic noise signal, and a clean image signal can be separated by the model.
  • the processing of using a DR image with a grid it is possible to effectively remove the astigmatism in the image, and because the period in which the noise is combined in the SPX model is processed, the periodic noise signal is separated.
  • the non-periodic noise signal and the clean image signal are used, the periodic influence of the noise can be taken into consideration, so that the ringing effect is not included in the clean image, thereby improving the image quality.
  • the embodiment further provides an apparatus for performing noise suppression on an image, which includes:
  • An image acquisition module configured to acquire an image to be processed, where the image to be processed is an image interfered by periodic noise
  • a period determining module configured to determine a period of periodic noise in the image to be processed
  • the optimization solution module is configured to substitute the image data of the image to be processed and the determined periodic noise into the signal separation model, and optimize the signal separation model to obtain a clean image signal corresponding to the image to be processed.
  • This embodiment still takes the example of performing gate shadow suppression on the grid image of the filter grid as an example, and assumes that there is a large noise at the bad line, and other pixel noise fields are small or 0 (ie, the noise is large and sparse.
  • the noise is large and sparse noise
  • it before performing gate shadow suppression, it is first considered to interpolate the non-periodic noise to restore the pixel value there, so as to reduce the influence of the non-periodic noise on the gate shadow suppression. As shown in FIG.
  • the general flow of the method for suppressing the grid shadow with the filter grid DR image proposed in this embodiment is as follows: First, the direction of the grid shadow is obtained (ie, periodic noise) The direction can be understood as the visual distribution direction of the periodic noise in the image. According to the direction of the grating shadow, it is determined whether interpolation is needed, and then the period of the gate shadow is estimated, so that the period is considered in the suppression of the grid shadow. The effect of the component is finally removed by the SPX model.
  • the acquisition of the shadow shadow direction can be automatically detected by the image processing method; of course, it can also be obtained by artificial means, such as by manual input, hardware switch, etc., the direction information can be obtained. These artificial methods can be implemented, for example, by means of interface input in human-computer interaction, and will not be described in detail herein.
  • the direction of the bad line ie, the direction of the non-periodic noise, it can be understood as the non-periodic noise such as the visual distribution direction of the bad line in the image), which can be observed by the human eye, or it can be calibrated.
  • the method is calculated.
  • the specific calibration method can refer to the existing calibration method, which will not be described in detail here.
  • the embodiment provides an image processing method for automatically performing gate shadow direction detection, so that the input image is divided into three types, that is, an image without a gate shadow and a first direction gate shadow (ie, a direction of the gate shadow) Parallel to the direction of the bad line, referred to herein as the lateral gaze image and the second directional shadow (ie, the direction of the shadow is parallel to the direction of the bad line, referred to herein as the vertical sash) image.
  • the basic flow of the shadow shadow direction detection algorithm is as shown in FIG. 8. First, the wavelet feature is extracted from the input DR image, and then the first supported support vector machine (SVM) is used.
  • SVM support vector machine
  • the classifier (Classifier 1 in the figure) performs classification with and without gates; for the rasterized DR image, the extracted features are again input to the second SVM classifier (Classifier 2 in the figure), It is judged whether the shadow is horizontal or vertical.
  • SVM classifier A method of feature extraction and SVM classifier design involved in an embodiment of the present invention is described below.
  • the purpose of feature extraction is to extract eigenvectors which are favorable for sag detection and sash direction classification.
  • the present embodiment mainly extracts wavelet features for classification.
  • the extraction algorithm includes the following steps S11-S13:
  • Step S11 calculating two-level two-dimensional wavelet transform for the input image, and selecting a Haar wavelet or a db wavelet to obtain a horizontal decomposition cHl, cH2 and a vertical decomposition cVl, cV2;
  • Step S12 for each of the obtained four decompositions, calculate the horizontal and vertical gradients in this way, that is, for each pixel position (w), calculate the gradients of the horizontal direction and the vertical direction, and count all
  • the number of pixels in the pixel position in which the horizontal direction gradient is larger than the vertical threshold by a certain threshold ⁇ , the number of pixels in which the vertical direction gradient is larger than the horizontal gradient certain threshold ⁇ in all pixel positions, and the image pixels for N1 and ⁇ 2 are used.
  • the number is normalized and returned as two eigenvalues of the present decomposition;
  • step S13 eight eigenvalues obtained by the four decompositions are returned as feature vectors of the input image.
  • the classifier of this embodiment is trained according to the training samples, and the classifier parameters mainly include the support vector, the parameters of the kernel function, and the bias.
  • the classification formula is:
  • _y is the transformed eigenvector of the input image
  • ⁇ X l
  • m is the support vector, which is the weighting coefficient
  • b is the offset, which is the kernel function.
  • Classifier 1 uses the rbf kernel function, ie
  • Classifier 2 uses a polynomial kernel function, ie
  • the parameters of classifier 1 include support vector, weighting coefficient, offset, and rbf parameters.
  • the parameters of classifier 2 include support vectors, weighting coefficients, and offsets. These parameters are all obtained during training.
  • the basic flow of training can usually be described as follows: First, prepare three parts of the sample, which are image samples without astigmatism, image samples of transverse gaze and image samples of vertical astigmatism; then train classifier 1, which will have a shadow The image samples (including the lateral shadow and the vertical shadow) are taken as positive samples with the label +1, and the image samples without the astigmatism are used as the negative samples, and the label is -1. The features of the positive and negative samples are extracted according to the above feature extraction algorithm.
  • the input classifier training algorithm is trained to obtain the parameters of the classifier 1; finally, the classifier 2 is trained, that is, the image sample of the lateral raster image is taken as a positive sample, the label is +1, and the image sample of the vertical grid shadow is used as the negative sample, the label For the -1, the feature extraction algorithm is used to extract the features of the positive and negative samples respectively, and the classifier training algorithm is input for training, and the parameters of the classifier 2 are obtained.
  • the feature extraction algorithm extracts the feature according to the foregoing feature extraction algorithm, and the presence or absence of the gate shadow detection and the shadow shadow direction detection is performed according to the flow of FIG.
  • the discriminant formula of the classifier is shown in the above formula (6). Judging when the output of classifier 1 calculated according to formula (6) is greater than 0 In order to have a shadow, when it is less than zero, it is judged as no shadow. When the output of the classifier 2 calculated according to the formula (6) is greater than 0, it is judged as a lateral raster, and when it is less than 0, it is judged as a vertical raster. It should be noted that the labels of positive or negative samples can be interchanged during training, and the corresponding labels of online classification should also be interchanged.
  • the wavelet features used in this embodiment are classified by SVM, and the specific wavelet features are obtained by two-level two-dimensional wavelet transform and the kernel function of the specific SVM classifier.
  • the feature of the wavelet transform of the present embodiment such as the Gabor wavelet feature, the statistical method of the feature, or the like, may be used to extract the feature.
  • the support vector machine of the embodiment may also include a kernel including but not limited to changing the support vector machine. , kernel functions, etc.
  • the detection algorithm of the shadow direction may also use known image features such as invariant moments, and may also use known classifiers such as neural networks, fisher classifiers, etc., as long as It is suitable for obtaining features and classifications with or without grid shadow and grid shadow.
  • the method of determining the gate shadow period and the method of suppressing the shadow of the present embodiment can be referred to the related description in the foregoing Embodiment 1.
  • One of the effects of the calculation of the shadow direction on the gate shadow suppression is that since the pixel value at the bad line has been restored by interpolation when it is judged that the direction of the gate shadow is perpendicular to the direction of the bad line, when performing gate shadow suppression, The value of the aforementioned input weight ⁇ may not be set to be large.
  • the present embodiment not only has the advantages of Embodiment 1, but also detects the direction of the gate shadow during processing, determines whether to interpolate the bad line according to the direction of the gate shadow to recover the pixel value at the bad line, and minimize the bad line. The effect on subsequent gate shadow suppression.
  • the embodiment further provides an apparatus for performing noise suppression on an image, including:
  • An image acquisition module configured to acquire an image to be processed, where the image to be processed is an image interfered by periodic noise
  • a direction detecting module configured to determine a direction of the periodic noise
  • a period determining module configured to determine a period of periodic noise in the image to be processed
  • An optimization solution module for using the image data of the image to be processed and the determined periodic noise
  • the signal is separated into a signal separation model, and the signal separation model is optimized to obtain a clean image signal corresponding to the image to be processed.
  • Embodiment 1 or 2 assuming that the cost function is about the cost term of X, (3 ⁇ 4 is the DCT matrix, ie, Q ⁇ R ⁇ ", as the length of each line of the image is increased, the calculation of the cost function based on the SPX model is performed.
  • the complexity increases squarely, that is, the progressive complexity is (( « 2 ). It can be understood that (3 ⁇ 4 is a similar complexity problem for FFT matrix or DFT matrix, etc.
  • This embodiment is for the existence of embodiment 1 or 2 It is possible to improve the complexity of the algorithm in order to reduce the complexity of the algorithm based on the SPX model.
  • the complexity is processed in a segmented manner, and if it is divided into segments, the progressive complexity becomes. Therefore, when ⁇ is closer to ", the complexity of the algorithm is connected.
  • the segmentation in this embodiment is different from the segmentation of the variable u in Embodiment 1.
  • the segmentation in this embodiment refers to segmenting a row of image data, and then using the algorithm in Embodiment 1. The calculation is performed to combine the present embodiment with the embodiment 1. It is apparent that the data processing for the segmentation of the variable u in the embodiment 1 at this time is a certain piece of image data which has been segmented in the present embodiment.
  • the present embodiment processes the boundary problem as follows: A line of image data is divided into segments, each segment is left with a small portion as an overlapping region, and the last image data is derived from the first cycle and the last segment discarded in each segment. The combination of cycles. As shown in Figures 9a and 9b, it can both illustrate the segmentation processing of the image data of the present invention.
  • a line of image data 90 is segmented, and two adjacent segments of image data overlap for two periods, the overlapping portions being as shown by the thick black line portion and the thickest black line portion in the figure.
  • each segment of the image data is discarded for its first and last cycles.
  • the second segment discards the thick black line segment with the overlapping portion number 201 and the thick black line segment labeled 204
  • the third segment discards the overlapping portion label 301 and label 304.
  • the thick black line segment The result of the above segmentation process is shown in Fig. 9b.
  • the second segment and the third segment respectively discard the overlap of the first segment and the end, since the second segment complements the discarding period 301, the third segment complements the discarding period. 204. There is no discontinuity in the image data at the intersection of the two segments. Further, as shown in FIG.
  • the first block of data of each segment of image data (marked as the thickest black in the figure) Line part) overlaps with several pieces of data at the end of the previous piece of image data (marked as the thickest black line part in the figure), and several pieces of data at the end of each piece of image data (marked as the thickest black line part in the figure) and The first pieces of data of a piece of image data (marked as the thickest black line portion in the figure) overlap.
  • the noise suppression method for the image subjected to periodic noise interference of the present embodiment is large.
  • the volume flow is: first, for the input image to be processed that is interfered by the periodic noise, determine the period (or direction and period) of the periodic noise according to the foregoing Embodiment 1 (or Embodiment 2), and press the image to be processed
  • the segmentation method of the embodiment performs segmentation processing to obtain a plurality of pieces of image data to be processed, and then processes the image data to be processed for each segment by using the sag suppression method in Embodiment 1, and finally obtains noise suppression of the image to be processed. the result of.
  • the complexity of the algorithm is reduced because of the segmentation method.
  • the method for suppressing the grid shadow with the grid DR image proposed in this embodiment can be implemented based on the foregoing embodiments, but the method for determining the gate shadow period is an algorithm based on the SPX model.
  • the size of the grief period in the embodiment is based on the aforementioned SPX model and its mathematical optimization problem, and can be obtained by the following algorithm:
  • the energy distribution of several frequency points is higher than the surrounding area, so the higher frequency amplitude corresponds to the frequency region where the grating shadow is located; based on this, the grating shadow region, that is, the region with relatively obvious period, can be selected.
  • This area is the area of interest.
  • the best period is measured as follows: The stronger the periodic signal is calculated, the closer the period is to the real period; that is, the 2 norm of the periodic signal is calculated;
  • the calculation method of the 2 norm is: Take the square sum of each element of the vector and then open the root number. Therefore, the period corresponding to the model with the strongest periodic signal is the period of the gate shadow.
  • the cost function of the above formula (3) is taken as an example, and a slight modification is made in the mathematical game:
  • represents the image data of the region of interest
  • t represents The candidate period is preset, and the candidate period in a certain range is substituted into the formula and expression involved here, and the candidate period corresponding to the model with the strongest periodic noise signal is
  • the optimal period that is, the period of periodic noise required to be taken
  • D is the difference matrix, and its form is as follows:
  • ⁇ 3 ⁇ 44 is called total-variation, which is a measure for measuring the smoothness of image signals; "the length of image data for the region of interest; the meanings of other parameters are described in the previous embodiment, and will not be repeated here. . Since the bad line can be avoided when estimating the period, the weight vector ⁇ can be set to all ones. It can be understood that the optimization function of the formula (3) is taken as an example to describe the period calculation. In other embodiments, the optimization function may be a constraint item as described above, which may become a constraint item, and each cost item. The former coefficients can be modified accordingly, and the 2 norm can be changed to other equivalent measures, etc., as long as it is suitable for calculating a function of a better period.
  • an embodiment of the present invention further provides a method for determining a period of periodic noise in an image, including:
  • An image acquisition step acquiring an image to be processed, the image to be processed being an image interfered by periodic noise, and acquiring image data of the region of interest from the image to be processed;
  • Optimizing the solution step Substituting the image data of the region of interest and the preset candidate period into the signal separation model, and optimizing the signal separation model to obtain a clean image signal and periodic noise corresponding to the image data of the region of interest And aperiodic noise;
  • the period determining step determining the candidate period corresponding to the model with the strongest periodic noise signal as the period of the periodic noise.
  • another embodiment further provides an apparatus for determining a period of periodic noise in an image, comprising:
  • An image acquisition module configured to acquire an image to be processed, where the image to be processed is an image interfered by periodic noise, and acquire image data of a region of interest from the image to be processed;
  • the optimization solution module is configured to substitute the image data of the region of interest and the preset candidate period into the signal separation model, and optimize the signal separation model to obtain a clean image signal and a period corresponding to the image data of the region of interest.
  • a period determining module configured to determine a candidate period corresponding to the model with the strongest periodic noise signal as the period of the periodic noise.
  • the noise suppression method or apparatus for images subjected to periodic noise interference proposed by the present invention can separate periodic noise and separate sparse noise by constructing an SPX model and performing an optimization solution (ie, Non-periodic noise), so that clean image noise can be obtained; in addition, in one embodiment, the segmentation processing method can significantly reduce the complexity of the algorithm based on the SPX model, and can effectively eliminate the boundary effect of the segmentation.
  • a method and apparatus for noise suppression of an image according to an embodiment of the present invention may be implemented in a current X-ray imaging system by hardware, software, firmware, or a combination thereof, thereby enabling the X-ray imaging system to employ an embodiment in accordance with the present invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Image Processing (AREA)

Abstract

一种对图像进行噪声抑制的方法及装置,方法包括:获取待处理图像,待处理图像为被周期性噪声干扰的图像;确定周期性噪声的周期;将所述周期性噪声的周期代入信号分离模型,对信号分离模型进行优化求解,得到对应于所述待处理图像的干净图像信号。基于SPX模型实现,在该模型时结合了确定出的周期性噪声的周期进行处理,使得在分离出周期性噪声信号、非周期性噪声信号和干净图像信号时,能够考虑到噪声的周期性影响,使得干净的图像中不包含振铃效应,进而提高图像质量。

Description

对图像进行噪声抑制的方法及装置 技术领域
本申请涉及图像处理技术, 尤其涉及一种对图像进行噪声抑制的方法及装 置, 特别地, 该图像为被周期性噪声干扰的图像。 背景技术
数字 X射线照相术( Digital Radiography, DR )是一种釆用数字传感器检测
X 光进行拍摄医学相片的技术, 其以数字的方式存储和处理医学图像, 具有釆 集和传输速度快、 更容易进行图像增强和显示等优点, 目前已经广泛应用于各 种医学检查和诊断。
为了提高 DR图像的分辨率, 大多数 DR产品都会在 X光接收平板和人体 之间加入一块防散射滤线栅。 该滤线栅的作用是, 让来自正面的穿过人体的 X 光通过的同时, 使得从人体散射出来的 X光被滤除, 从而尽量避免平板上的同 一感光点接收到来自人体不同组织的 X光信号, 以提高图像对比度和空间分辨 率。 当滤线栅的空间频率不是图像的空间釆样频率的整数倍时, 图像上会显示 出条紋状的噪声, 为如图 1所示的栅影 13 , 其通常呈现为周期性噪声。 这种现 象称之为摩尔效应 ( Moir effect ); 所产生的条紋又称为摩尔条紋或云紋。 摩尔 条紋的存在会影响正常人体组织显示, 进而影响医生对病症的诊断。 为了避免 这种摩尔条紋, 通常可釆用如下几种方法, 即: 1 )调整滤线栅的线密度, 使得 其频率是平板釆样频率(即像素点间隔的倒数) 的整数倍; 2 )釆用运动栅, 当 栅的运动速度足够大, 使得栅经过像素点的时间小于该像素点的曝光时间时, 栅影将减弱或消失; 3 )使用图像处理的方法去除这种云紋。 其中, 方法 1 )和 方法 2 )对滤线栅的结构和安装要求较高, 而方法 3 )对滤线栅的结构和安装方 式要求降低, 但需要去除成像时产生的条紋, 并且要求不能受到成像平板本身 缺陷 (如坏线、 坏点) 带来的影响。
对于使用图像处理的方法, 大多数是使用基于空间域卷积滤波或频域直接 进行抑制。 这一类方法容易受成像平板坏线的影响, 滤波过程中在坏线周围产 生扩散式的振铃效应, 参见图 2的坏线图像和图 3的低通滤波后的图像。 为了 降低这种振铃效应, 通常有两种做法。 一是在设计滤波时在尽量降低滤波器的 阶数, 通过牺牲频率特性来换取较低的振铃效应。 二是尽量减小坏线或坏点处 周围像素点的突变, 即釆用好的插值算法计算出坏线或坏点的可能值。 因为对 坏线处釆用常用的插值方法如线性插值、 三次插值等进行插值, 只能得到平滑 的过渡, 破坏了栅影的周期性, 导致频域滤波产生振铃效应而不能达到有效降 低振铃的效果。对于带栅影的 DR图像,第一种方法通常无法达到在彻底滤除栅 影的同时而不在坏线处产生振铃效应, 而第二种方法则需要在插值的同时考虑 栅影的影响, 即不能只考虑简单的邻域插值, 还应将栅影的周期性规律考虑进 去。 然而, 目前尚未有研究给出既能有效去除栅影同时又能将振铃的产生最小 化的算法。 发明内容
根据本发明的一个方面, 提供一种对图像进行噪声抑制的方法, 包括: 图像获取步骤: 获取待处理图像, 所述待处理图像为被周期性噪声干扰的 图像;
周期确定步骤: 确定所述待处理图像中周期性噪声的周期;
优化求解步骤: 将待处理图像的图像数据和所述周期性噪声的周期代入信 号分离模型, 对信号分离模型进行优化求解, 得到对应于所述待处理图像的干 净图像信号。
根据本发明的另一个方面, 提供一种对图像进行噪声抑制的装置, 包括: 图像获取模块, 用于获取待处理图像, 所述待处理图像为被周期性噪声干 扰的图像;
周期确定步骤: 确定所述待处理图像中周期性噪声的周期;
优化求解步骤: 将待处理图像的图像数据和所述周期性噪声的周期代入信 号分离模型, 对信号分离模型进行优化求解, 得到对应于所述待处理图像的干 净图像信号。
本发明的对图像进行噪声抑制的方法是通过信号分离模型来实现, 在使用 该模型时结合了确定出的周期性噪声的周期进行处理, 使得在分离出周期性噪 声信号、 非周期性噪声信号和干净图像信号这三个分量时能够考虑到噪声的周 期性影响, 使得干净的图像中不包含振铃效应, 进而提高图像质量。 附图说明
图 1为防散射滤线栅工作时栅影产生的示意图, 其中 11为 X光接收平板, 12为滤线栅, 13为栅影, 短箭头 14为直射线, 长箭头 15为散射线;
图 2和图 3分别为坏线振铃产生的坏线图像示意图和低通滤波后的图像示 意图;
图 4为本发明一种实施例的带滤线栅的 DR图像栅影抑制方法的流程示意 图;
图 5为本发明一种实施例中输入的 DR图像信号分解为三个信号分量的示意 图;
图 6为本发明一种实施例中 DCT频域及其滤波器系数的举例示意图; 图 7为图 5所示信号分量;?的示意图;
图 8为本发明一种实施例中的栅影方向检测算法的流程示意图;
图 9a和 9b为本发明一种实施例中对待处理图像进行分段的示意图。 具体实施方式
本发明提出一种对图像进行噪声抑制方法, 特别地, 该图像为被周期性噪 声干扰的图像。 该方法可以应用于包括但不限于带防散射滤线栅的数字放射图 像 (简称 DR图像)中。 以下以对带滤线栅的 DR图像进行栅影抑制 (即周期性 噪声为栅影) 为例, 对本发明的对图像进行噪声抑制方法进行详细说明, 应理 解, 以下的方法能适用于各种需要抑制周期性噪声的图像。
本发明实施例提出的对图像进行噪声抑制的方法是基于这样一种情形下的 信号处理: 一个干净的信号 X受到周期性噪声;?的加性干扰, 并且在观测时由 于传感器缺陷引入稀疏的噪声 s (也称为非周期性噪声), 此时需要从传感器输 出 ί来进行 ·、 和 X三个分量的估计。 简言之, 即根据式子 ί =·ν+/?+χ分别求出 s、p和 X三个未知分量,,通过求解式子 ί = +/?+Χ能够分离出周期性噪声信号(即 ρ分量, 为栅影引起的噪声)、 非周期性噪声信号 (即 S分量, 为如坏线或坏点 导致的噪声)和干净的图像信号(即 X分量, 为去除了栅影和坏线坏点等之后 的图像信号), 从而可以得到干净的图像信号。
以下通过具体实施例结合附图对本发明作进一步详细说明。
实施例 1 :
如图 4所示,本实施例提出了一种带滤线栅 DR图像的栅影抑制方法,通过 该方法体现出本发明提出的对图像(尤其指被周期性噪声如栅影干扰的图像) 进行噪声抑制的方法, 其大体流程是: 首先确定栅影周期, 然后进行栅影抑制 (图示中的实线部分)。
对于栅影周期的确定, 通常有两种方法可以获得该周期参数, 一是通过目 测图像而得到, 二是釆用数值方法进行自动计算, 关于后者的一种具体实现可 参考下文中的实施例 4。
在得到估计的栅影周期后, 可进行栅影抑制。 本实施例中基于前述的信号 处理方式进行栅影抑制, 即把信号 ί (即待处理图像)分离成三个分量(见图 5 ), 包括未知的干净图像信号 x, 周期性噪声;?和非周期性噪声 这三个分量满足 x+s+p=d, 据此构建信号分离模型(简称 SPX模型), 并将栅影周期代入 SPX模 型进行优化求解, 从而可以得到干净图像信号 x。 实施例中, SPX模型的获得可 通过如下方法构建而得; 或者是直接调用事先已存储的模型, 该已存储的模型 可以是按照如下方法事先构建而得。
这里所构建的 SPX模型包括代价函数和约束条件, 其中代价函数为满足约 束条件且与干净信号分量、 周期信号分量和非周期信号分量中的至少一个相关 的函数, 干净信号分量为关于干净图像信号的平滑度的测度函数, 周期信号分 量为关于周期性噪声及其周期的测度函数, 非周期信号分量为关于非周期性噪 声的测度函数。
本实施例中, 干净信号分量的表达式为 周期信号分量的表达式为 e2 , 非周期信号分量的表达式为 e3 g(w)4;约束条件包括固定约束条件, 或者包括固定约 条件和指定约束条件,其中固定约束条件为
Figure imgf000006_0001
本实施例的代价函数表达式为:
ei iQ ti + e2 IHE + es I go I!, ( 1 ) subject to x+s+Mu=d, ( 2 ) 其中 为周期性噪声, 为转换矩阵, M为用于重建周期信号的变量, ei、 e2和 e3分别为用于控制干净图像信号、 周期性噪声和非周期性噪声的大小的权 重, ^c =diag(c) , β为待处理图像的幅度谱矩阵, c为权重向量, ^为用于降低 或提高非周期性噪声的值的权重。
其它实施例中, 代价函数中的代价项可以变为约束项, 例如: 代价函数可 以表示为^|^¾, 且该代价函数还满足固定约束条件和指定约束条件, 该指定 约束条件为 e2
Figure imgf000006_0002
<r2 , 或者该指定约束条件为
Figure imgf000006_0003
<r3 或者, 代价函数可以表示为 e21 , 且满足固定约束条件 和指定约束条件,该指定约束条件为 £1|0^ < 4且^^ )4<^或者该指定约 束条件为 |gc g+e3
Figure imgf000006_0004
<r5 或者, 该代价函数可以表示为 e3 , 且 满足固定约束条件和指定约束条件, 该指定约束条件为 ei| x < 4£2 < , 或者该指定约束条件为 ei| x +e2K<r6; 或者, 该代价函数可表示为
^Ιαί+^ΙΗΕ ' 且满足固定约束条件和指定约束条件, 该指定约束条件为 e3
Figure imgf000006_0005
<r2 ; 或者, 该代价函数为 , 且满足固定约束条件 和指定约束条件, 该指定约束条件为 ¾< ; 或者, 该代价函数为
Figure imgf000006_0006
, 且满足固定约束条件和指定约束条件, 该指定约束条件为 ex ||ec g < r4 , 其中, 、 r r3、 r4、 r5和 r6为预设值如经验值或试验值等。
本实施例中, 对 、 和 X的测度函数是釆用 或 范数, 如对 S项(即非 周期性噪声项)釆用 范数, 对 X项 (即干净图像信号项)和;?项 (即周期性 噪声项)釆用 2范数; 当然, 在其它实施例中, 可以釆用其它显而易见的近似 或等价测度来替代 范数和 /或 范数。 例如, 可以对 ρ项 (即周期性噪声项 ) 釆用 范数, 对 X项 (即干净图像信号项)和 S项 (即非周期性噪声项)釆用 2范数, 也可以是对 S项 (即非周期性噪声项)釆用 范数, 对 X项 (即干净 图像信号项)和;?项 (即周期性噪声项)釆用 2范数, 等等, 以此类推。
本实施例中, 设 ei =丄, e2 = e3 = - ' 其中《c为矩阵 β的行数且不等于 nc Kt n
零, "和 分别为用于控制周期性噪声和非周期性噪声与干净图像信号的大小的 权重, f 为将周期性噪声分段而设定的段数, ί为每一段周期性噪声的周期, η=Κίΐ0, Τ。为每一段周期性噪声中 ί的个数。
以本实施例的代价函数为例(即式( 1 ) ), 对构建的 SPX模型进行优化求解 则是求解下面的数学优化问题, 由此可以实现栅影抑制:
minw ( 3 )
Figure imgf000007_0001
subject to x+s+Mu=d ( 4 ) 即 SPX模型以最小化代价函数为优化目标, 在满足约束条件的前提下, 结 合周期性噪声的周期作为参数, 通过求解代价函数得到干净图像信号、 周期性 噪声和非周期性噪声。 此时代价函数为 , 则该数学
Figure imgf000007_0002
优化问题是求使代价函数为最小时的 x、 u、 s的值, 且其约束条件是 χ+·ν+Μ = 其中对 s项釆用 范数, 范数的计算方法为取向量各元素的绝对值的和, 对
X项和 U项釆用 2范数, 2范数的计算方法为取向量各元素的平方和然后求平 方根。 如前述, 在式子 (3 )和(4 ) 中, e R"为输入信号 (即输入的待进行噪 声抑制处理的 DR图像), s为非周期性噪声,;?= M为周期信号, X为干净的信 号。 ^为带权重的 DCT (离散余弦变换)矩阵, 目的是滤波, 即将某个频率的 分量抑制掉以使得 X项最小, 其计算方法见式(5 )。
Qc = diag(c)Q ( 5 ) 其中 Q为输入信号(即待进行噪声抑制处理的 DR图像)的 DCT幅度谱(其 可通过对图像进行 DCT变换后求频谱,进而求得幅度谱), c e R"为权重向量(或 称滤波器系数向量)。 c的形状设为一个或多个波峰的形式, 分别对应 DCT频谱 上由于周期性噪声造成的波峰, 如图 6 所示。 滤波器系数越大, 则对周期性噪 声的抑制越大; 滤波器系数为 0 时, 对应的频谱将不会衰减。 《c为不等于零的 滤波器系数的个数,相当于是矩阵 ^的行数。该具体实现中 ^为带权重的 DCT 矩阵, 其它具体实现中(¾还可以是例如带权重的 DFT (离散傅里叶变换)矩阵 或 FFT (快速傅里叶变换)矩阵等。 对于用于重建周期信号的变量 U,每一个信号可以分解为 f段,每段用一个 基波表示, 该基波的长度为 ί, 重复 TQ次, ί即为栅影的周期, 可以通过经验值 得到, 或者通过后续对栅影周期的估计算法计算而得。 这种分段的目的是增加 鲁棒性。 如图 7所示, 以图 5所示周期信号;?为例, 周期信号;?可分解为 f段, 每段用一个基波表示, 该基波的长度为 ί, 重复 Τ。次(图示中 Τ。=3)。 Μ可以表 示如下:
u = [u ,u2 T,...,u ,...uK T T G RKt , 其中 . eR,
是信号转换矩阵, 使得 = [ ,...,«,..., ..., ,..., ] R"。 其中", Κ, T0有如下的关系: n=K T0。 当信号长度不是 ίΤ0的整数倍时, 可以对 进 行截断处理。
weR"是输入权重, 通过增加或减少 ^的值可以降低或提高噪声项 s在位置
, (即 的值。 ", 分别控制周期信号;?和噪声项 s与信号 的大小关系。 增加 "可以降低周期信号 ρ相对于 s和 X的幅值,同时由于等式约束的关系,即式( 4 ), •V和 /或 X的幅值相应降值。 对 也同理。 这里提及的参数^、 a. 等在实际使 用中可根据经验值或是试验值得到。
上述式(3)和(4)涉及到 范数和 2范数的数学优化, 可通过现有相关 数学知识进行求解, 即可得到分离出的三个分量;?、 s和 X, X即为分离后的干净 的图像信号, 即去除了栅影和坏线坏点后的图像信号。
综上, 本实施例的对受周期性噪声干扰的图像进行噪声抑制的方法是基于 SPX模型实现, 通过该模型能分离出周期性噪声信号、 非周期性噪声信号和干 净图像信号,将该方法应用于使用带滤线栅的 DR图像的处理中, 能够在有效地 去除图像中的栅影, 而且由于在釆用 SPX模型时结合了噪声的周期进行处理, 使得在分离出周期性噪声信号、 非周期性噪声信号和干净图像信号时, 能够考 虑到噪声的周期性影响, 使干净的图像中不包含振铃效应, 进而提高图像质量。
基于上述方法, 本实施例还提供了一种对图像进行噪声抑制的装置, 其包 括:
图像获取模块, 用于获取待处理图像, 该待处理图像为被周期性噪声干扰 的图像;
周期确定模块, 用于确定待处理图像中周期性噪声的周期;
优化求解模块, 用于将待处理图像的图像数据和确定出的周期性噪声的周 期代入信号分离模型, 对信号分离模型进行优化求解, 得到对应于所述待处理 图像的干净图像信号。
以上各模块的具体实现可参考上述方法实施例中相应的描述, 在此不作重 述。
实施例 2:
本实施例仍以对带滤线栅 DR图像的进行栅影抑制为例进行说明 ,并假设坏 线处存在较大的噪声, 而其它像素噪声场较小或为 0 (即噪声为大的稀疏噪声 large and sparse noise ), 所以本实施例在进行栅影抑制之前, 首先考虑对非周期 性噪声处进行插值以恢复该处的像素值, 以便减小非周期性噪声对栅影抑制的 影响。 如图 4 所示, 包括图示中的虚线部分和实线部分, 本实施例提出的带滤 线栅 DR图像的栅影抑制方法的大体流程是: 首先获取栅影方向(即周期性噪声 的方向, 可以理解为是周期性噪声呈现在图像中的视觉上的分布方向), 根据栅 影方向情况确定是否需要进行插值, 然后估计出栅影的周期, 以便在栅影抑制 时考虑这种周期分量的影响, 最后根据 SPX模型去除栅影。
栅影方向的获取可釆用图像处理的方法自动进行检测得到; 当然, 也可以 通过人为的方式得到, 如通过人工输入、 硬件开关等均可以得到方向信息。 这 些人为的方式可以参考例如人机交互中通过界面输入的方式实现, 在此不做详 述。 至于坏线的方向 (即非周期性噪声的方向, 可以理解为是非周期性噪声如 坏线呈现在图像中的视觉上的分布方向), 其可以通过人眼观察得到, 也可以是 通过标定的方式计算得到, 具体的标定方法可以参考现有的标定方法, 此处不 做详述。
本实施例提供了一种釆用图像处理的方法自动进行栅影方向检测, 以使得 将输入图像分为三种类型, 即不存在栅影的图像、 第一方向栅影 (即栅影的方 向与坏线的方向平行, 本文中称之为横向栅影) 图像和第二方向栅影 (即栅影 的方向与坏线的方向平行, 本文中称之为竖向栅影) 图像。 在本实施例中, 栅 影方向检测算法的基本流程如图 8所示, 首先对输入的 DR图像提取小波特征, 然后釆用预先训练好的第一个支持向量机 ( Support Vector Machine , SVM )分 类器 (图示中的分类器 1 )进行有栅和无栅的分类; 针对有栅的 DR图像, 将提 取的特征再次输入到第二个 SVM分类器(图示中的分类器 2 ),判断出栅影是横 向还是竖向。以下介绍本发明一种实施例中涉及的特征提取和 SVM分类器设计 的方法。
特征提取的目的是从图像中提取出有利于栅影检测和栅影方向分类的特征 向量, 本实施例主要是提取小波特征进行分类, 其提取算法包括如下步骤 S11-S13 :
步骤 S11 , 对于输入图像计算两级二维小波变换, 可选用 Haar小波或者 db 小波, 得到水平方向的分解 cHl、 cH2和竖直方向的分解 cVl、 cV2; 步骤 S12,对于得到的四个分解中的每一个分解, 按这种方式计算水平和竖 直的梯度, 即, 对于每个像素位置 (w), 计算水平方向和竖直方向的梯度, 统计 所有像素位置中水平方向梯度大于竖直方向梯度一定门限 ε的像素个数 N1, 统 计所有像素位置中竖直方向梯度大于水平方向梯度一定门限 ε的像素个数 Ν2, 对 N1和 Ν2用图像像素个数进行归一化并返回, 作为本分解的两个特征值; 步骤 S13, 返回四个分解得到的八个特征值作为输入图像的特征向量。
对于 SVM分类器设计, 本实施例的分类器根据训练样本训练得到, 分类器 参数主要包括支持向量、 核函数的参数和偏置 (bias)。 分类的公式为:
f(y) = sign\ Targ{y y) + b ( 6 )
i=l
其中 _y为输入图像的变换后特征向量, {X =l,...m为支持向量, 为加权系数, b为偏置, 为核函数。 特征向量的每一个值需要经过平移和缩放后再输入, 即
Figure imgf000010_0001
其中 _y为从图像中直接提取的特征向量, z和 /分别为缩放和平移参数向量。 分类器 1釆用 rbf核函数, 即
^( ^) = 6χρ〔- ^ly- 〕 (8) 其中 F为 rbf参数。 分类器 2釆用多项式核函数, 即
g{yl,y) = y y(y y+^) (9) 分类器 1的参数包括支持向量、 加权系数、 偏置、 rbf参数。 分类器 2的参 数包括支持向量、 加权系数和偏置。 这些参数都是在训练中得到。
训练的基本流程通常可以描述为: 首先准备好三部分样本, 分别为无栅影 的图像样本、 横向栅影的图像样本和竖向栅影的图像样本; 然后训练分类器 1, 即将有栅影 (包括横向栅影和竖向栅影) 的图像样本作为正样本, 标号为 +1, 无栅影的图像样本作为负样本, 标号为 -1,按前述特征提取算法分别提取正负样 本的特征, 输入分类器训练算法进行训练, 得到分类器 1 的参数; 最后训练分 类器 2, 即将横向栅影的图像样本作为正样本, 标号为 +1, 竖向栅影的图像样本 作为负样本, 标号为 -1, 按前述特征提取算法分别提取正负样本的特征, 输入分 类器训练算法进行训练, 得到分类器 2的参数。
在线分类时, 对于新输入的图像(即前述待处理图像)按前述特征提取算 法提取特征, 按照图 8 的流程进行有无栅影检测和栅影方向检测。 分类器的判 别公式见上述公式( 6 )。 当分类器 1的按照公式( 6 )计算的输出大于 0时判断 为有栅影, 小于零时判断为无栅影。 当分类器 2的按照公式(6 )计算的输出大 于 0时判断为横向栅影, 小于 0时判断为竖向栅影。 需要注意的是, 训练时正 样本或者负样本的标号可以互换, 相应地在线分类的得到的标号也应该互换。
本实施例中釆用的小波特征并结合 SVM进行分类,其中给出了具体的小波 特征为通过两级二维小波变换而得到以及具体的 SVM分类器的核函数,可以理 解, 在其它实施例中,还可以通过改变本实施例的小波变换的类型如釆用 Gabor 小波特征、 特征的统计方式等来提取特征, 也可以对本实施例的支持向量机进 行包括但不限于改变支持向量机的核、 核函数等。 此外, 在另外一些实施例中, 栅影方向的检测算法还可以釆用已知的图像特征如不变矩等, 也可以釆用已知 的分类器如神经网络、 fisher分类器等, 只要是适于得到有无栅影及栅影方向的 特征和分类均可。
在通过上述方法获取栅影方向后,对栅影方向进行判断。 若是竖向栅影(即 栅影的方向与坏线的方向垂直), 则直接进行常规的差值处理如线性插值, 插值 后进行栅影周期计算及后续的栅影抑制处理, 可以理解,插值后的 DR图像中相 当于不存在坏线, 从而在后续的处理中可以据此设定相关的权重; 若是横向栅 影(即栅影的方向与坏线的方向平行), 则不需要对坏线处进行插值处理, 直接 进行栅影周期计算及后续的栅影抑制处理。 在此进一步说明的是, 当检测到的 非周期性噪声为坏点时, 无论栅影方向如何, 均认为此时栅影的方向与坏点平 行(即为横向栅影), 直接进行栅影周期计算及后续的栅影抑制处理。
本实施例的栅影周期确定方法以及栅影抑制方法可参考前述实施例 1 中的 相关描述。 栅影方向的计算对栅影抑制的影响之一在于, 由于在判断出栅影方 向与坏线方向垂直时, 已通过插值方式恢复坏线处的像素值, 所以, 在进行栅 影抑制时, 前述的输入权重 ^的值可以不用设置为很大。
综上, 本实施例不仅具有实施例 1 的优点, 而且在处理时首先检测栅影方 向, 根据栅影方向确定是否对坏线进行插值以恢复坏线处的像素值, 尽量减小 了坏线对后续栅影抑制的影响。
基于本实施例的方法, 本实施例还提供了一种对图像进行噪声抑制的装置, 其包括:
图像获取模块, 用于获取待处理图像, 该待处理图像为被周期性噪声干扰 的图像;
方向检测模块, 用于确定所述周期性噪声的方向;
周期确定模块, 用于确定待处理图像中周期性噪声的周期;
优化求解模块, 用于将待处理图像的图像数据和确定出的周期性噪声的周 期代入信号分离模型, 对信号分离模型进行优化求解, 得到对应于所述待处理 图像的干净图像信号。
以上各模块的具体实现可参考上述方法实施例中相应的描述, 在此不作重 述。
实施例 3 :
在实施例 1或 2中, 假设代价函数的关于 X的代价项中, (¾为 DCT矩阵, 即 Q^ R^" , 随着图像每一行长度《的增加, 基于 SPX模型的代价函数的计算复 杂度呈平方增加, 即渐进复杂度为( («2)。 可以理解, (¾为 FFT矩阵或 DFT矩 阵等也存在类似的复杂度问题。 本实施例即是针对实施例 1或 2存在的可能较 高的算法复杂度进行改进, 目的是降低基于 SPX模型的算法复杂度。
本实施例中, 杂度, 釆用分段的方式进行处理, 假设分为 段, 则渐进复杂度变为 。 因此, 当 ^越接近于《时, 算法的复杂度接
Figure imgf000012_0001
Figure imgf000012_0002
近 0(n) , 即线性复杂度。 应理解, 本实施例中说的分段与实施例 1 中变量 u的 分段不同, 本实施例中的分段是指对一行图像数据进行分段, 然后再釆用实施 例 1 中的算法进行计算, 将本实施例和实施例 1结合, 艮显然, 此时实施例 1 中变量 u的分段所针对的数据处理^ ^于本实施例已经分段得到的某一段图像 数据。
一般地, 如果采用分段处理, 则需要解决由于频域滤波的副作用而导致的 分段边界的不连续问题。 对此, 本实施例这样处理边界问题: 将一行图像数据 分成 段, 每一段均前后留出一小部分作为交叠区域, 最后的图像数据来源于 每一段丟弃的第一个周期和最后一个周期的组合。 如图 9a和 9b所示, 其均可 示出本发明对图像数据的分段处理。 在图 9a中, 对一行图像数据 90进行分段, 前后两段相邻图像数据重叠两个周期, 重叠部分如图中粗黑线部分和最粗黑线 部分所示。 分段过程中, 除第一段图像数据和最后一段图像数据外, 每一段图 像数据均丟弃其第一个周期和最后一个周期。 以第二段和第三段图像数据为例, 第二段丟弃重叠部分标号为 201 的粗黑线段和标号为 204的粗黑线段, 第三段 丟弃重叠部分标号为 301和标号为 304的粗黑线段。 上述分段处理的结果见图 9b , 尽管第二段和第三段分别丟弃了首段和末端重叠, 但由于第二段补足了丟 弃的周期 301 , 第三段补足了丟弃的周期 204 , 在两段交接处图像数据并不存在 不连续的现象。 进一步如图 9b所示, 除第一段图像数据的起始若干块数据与最 后一段图像数据的末尾若干块数据外, 每一段图像数据的起始若干块数据(图 示中标记为最粗黑线部分) 与前一段图像数据的末尾若干块数据(图示中标记 为最粗黑线部分) 重叠, 每一段图像数据的末尾若干块数据(图示中标记为最 粗黑线部分) 与后一段图像数据的起始若干块数据(图示中标记为最粗黑线部 分)重叠。
从整体来看, 本实施例的对受周期性噪声干扰的图像的噪声抑制方法的大 体流程是: 首先对输入的受周期性噪声干扰的待处理图像, 按前述实施例 1 (或 者实施例 2 )确定其周期性噪声的周期 (或者是方向和周期), 将该待处理图像 按本实施例的分段方法进行分段处理, 得到多段待处理图像数据, 然后针对每 一段待处理图像数据釆用实施例 1 中的栅影抑制方法进行处理, 最终得到待处 理图像的噪声抑制后的结果。 如前述, 由于釆用了分段的方式, 降低了算法的 复杂度。
实施例 4:
本实施例提出的一种带滤线栅 DR 图像的栅影抑制方法可基于前述各实施 例实现, 但其中涉及的确定栅影周期采用的是一种基于 SPX模型实现的算法。
实施例中栅影周期的大小基于前述的 SPX模型及其数学优化问题实现, 可 通过如下算法得到:
对于输入的待处理图像, 选取其中一块感兴趣区域进行周期扫描, 寻找最 佳的周期, 候选周期为经验值, 例如可以设置为 5-10的范围, 或根据实际情况 进行调整。 感兴趣区域最好应选择周期比较明显的区域, 避开由于剂量不足或 过剩导致周期信号太弱的区域, 或者釆用随机选取的方式。 由于包含栅影的图 像经过傅里叶(DFT )变换到频域后, 栅影在频域内分布在多个频率(高频、 中 频、 低频) 与图像信息变换后的部分频率重合, 导致在频谱上可见数个频率点 能量分布较其周围高, 所以这个频率幅度较高之处对应地即为栅影所处的频率 区域; 基于此, 可以选择出栅影区域, 即周期比较明显的区域, 该区域即为感 兴趣区域。
最佳周期的衡量方法为: 最终计算出来的周期信号越强, 则周期越接近真 实周期; 即计算周期信号的 2范数 |;4。 2范数的计算方法为: 取向量各元素 的平方和再开根号。 因此, 取分解出来的周期信号最强的模型对应的周期即为 栅影的周期。 本实施例中以前述式(3 ) 的代价函数为例, 稍作变形, 在数学上 的表戏 'ά疋:
首先计算
Figure imgf000013_0001
( 10 ) subject to x+s+Mu=d ( 11 ) 其中 argmin表示的是使目标函数取最小值时的变量值, 体现在这里是指, 求使代价函数取最小值时候的 M值;
然后根据计算得到的 M值计算
Figure imgf000013_0002
这里涉及到的公式和表达式中, ί表示的是感兴趣区域的图像数据; t表示 的是预先设定的候选周期, 将预先设定的某个范围内的候选周期代入此处涉及 的公式和表达式进行计算, 求解出来的周期性噪声信号最强的模型对应的那个 候选周期为最佳周期, 即需要求取的周期性噪声的周期; D 为差分矩阵, 其形 式如下:
Figure imgf000014_0001
差分矩阵的作用是用于求信号的差分, 即/) x = [Xi - x2, x2 - x3, ..., x„— x„f , 其中
|ί¾4被称为总变差(total-variation ), 是用于衡量图像信号平滑度的测度; 《为感 兴趣区域的图像数据的长度; 其它参数的含义参见前述实施例, 在此不作重述。 由于估计周期时可以避开坏线处, 权重向量^可以设为全 1。 可以理解, 这里是 以式(3 ) 的优化函数为例稍作变形来进行周期计算的说明, 其它实施例中该优 化函数可以是如前所述的代价项可以变为约束项, 各代价项前的系数可以相应 修改, 以及 2范数也可以改为其它的等价测度等, 只要其适合于计算出较好的 周期的函数。
为了增加鲁棒性, 可以选取多行进行计算, 找所有行周期的范数的平方和 的累加值最大的对应周期作为当前图像的栅影估计周期。
根据本实施例, 本发明一种实施例还提供了一种确定图像中周期性噪声的 周期的方法, 包括:
图像获取步骤: 获取待处理图像, 所述待处理图像为被周期性噪声干扰的 图像, 从所述待处理图像中获取感兴趣区域的图像数据;
优化求解步骤: 将感兴趣区域的图像数据和预先设定的候选周期代入信号 分离模型, 对信号分离模型进行优化求解, 得到对应于所述感兴趣区域的图像 数据的干净图像信号、 周期性噪声和非周期性噪声;
周期确定步骤: 将求解出的周期性噪声信号最强的模型对应的候选周期确 定为周期性噪声的周期。
以上各步骤的具体实现可参考前述确定周期性噪声的周期中的相关描述, 在此不作重述。
基于该确定图像中周期性噪声的周期的方法的实施例, 另一种实施例还提 供了一种确定图像中周期性噪声的周期的装置, 包括:
图像获取模块, 用于获取待处理图像, 所述待处理图像为被周期性噪声干 扰的图像, 从所述待处理图像中获取感兴趣区域的图像数据;
优化求解模块, 用于将感兴趣区域的图像数据和预先设定的候选周期代入 信号分离模型, 对信号分离模型进行优化求解, 得到对应于所述感兴趣区域的 图像数据的干净图像信号、 周期性噪声和非周期性噪声; 周期确定模块, 用于将求解出的周期性噪声信号最强的模型对应的候选周 期确定为周期性噪声的周期。
以上各模块的具体实现可参考前述确定周期性噪声的周期中的相关描述, 在此不作重述。
基于以上实施例, 本发明提出的对受周期性噪声干扰的图像的噪声抑制方 法或装置中, 其通过构建 SPX模型并进行优化求解既能分离出周期性噪声, 还 能分离出稀疏噪声 (即非周期性噪声), 从而可以得到干净图像噪声; 此外, 一 种实施例中通过分段处理方式, 能显著降低基于 SPX模型的算法复杂度, 并能 有效消除分段的边界效应。 按照本发明实施例的对图像进行噪声抑制的方法及 其装置, 可以通过硬件、 软件、 固件或者其组合是现在 X射线成像系统中, 从 而使得 X射线成像系统可以釆用按照本发明实施例的噪声抑制的方法, 或者包 括按照本发明实施例的噪声抑制的装置。
本领域技术人员可以理解, 上述实施方式中各种方法的全部或部分步骤可 以通过程序来指令相关硬件完成, 该程序可以存储于一计算机可读存储介质中, 存储介质可以包括: 只读存储器、 随机存储器、 磁盘或光盘等。
以上内容是结合具体的实施方式对本申请所作的进一步详细说明, 不能认 定本申请的具体实施只局限于这些说明。 对于本申请所属技术领域的普通技术 人员来说, 在不脱离本申请构思的前提下, 还可以做出若干简单推演或替换。

Claims

权 利 要 求 书
1. 一种对图像进行噪声抑制的方法, 其特征在于, 包括:
图像获取步骤: 获取待处理图像, 所述待处理图像为被周期性噪声干扰的 图像;
周期确定步骤: 确定所述待处理图像中周期性噪声的周期;
优化求解步骤: 将待处理图像的图像数据和所述周期性噪声的周期代入信 号分离模型, 对信号分离模型进行优化求解, 得到对应于所述待处理图像的干 净图像信号。
2. 根据权利要求 1所述的方法, 其特征在于, 所述信号分离模型包括代 价函数和约束条件, 所述代价函数为满足所述约束条件且与干净信号分量、 周 期信号分量和非周期信号分量中的至少一个相关的函数;
所述干净信号分量为关于干净图像信号的平滑度的测度函数, 所述周期信 号分量为关于周期性噪声及其周期的测度函数, 所述非周期信号分量为关于非 周期性噪声的测度函数。
3. 根据权利要求 2所述的方法, 其特征在于, 所述约束条件包括固定约 束条件, 或者所述约束条件包括固定约束条件和指定约束条件,
所述固定约束条件为: x+s+Mu=d,
所述干净信号分量的表达式包括:
所述周期信号分量的表达式包括: e2 K ,
所述非周期信号分量的表达式包括: e3
Figure imgf000016_0001
,
其中 ί表示待处理图像, X表示干净图像信号, s表示非周期性噪声, u表 示用于重建周期性噪声的变量, Mu表示周期性噪声, M为转换矩阵, ei、 e2 和 e3分别为用于控制干净图像信号、周期性噪声和非周期性噪声的大小的权重, Qc = diag(c)Q , β为待处理图像的幅度谱矩阵, c为权重向量, ^为用于降低或提 高非周期性噪声的值的权重;
所述代价函数为 | x + , 且满足固定约束条件; 或者, 所述代价函数为^ 定约束条件和指定约束条件, 所 述指定约束条件为 e2 Κ < ^
Figure imgf000016_0002
, 或者所述指定约束条件为 或者, 所述代价函数为 e2 K , 且满足固定约束条件和指定约束条件, 所述 指定约束条件为 l x < r4且 e3
Figure imgf000016_0003
< r2 , 或者所述指定约束条件为 ei lfex 1 < ;
或者, 所述代价函数为^ )^ , 且满足固定约束条件和指定约束条件, 所述指定约束条件为
Figure imgf000016_0004
ri , 或者所述指定约束条件为 ei |feX + e2 IHI2 < r6;
或者, 所述代价函数为£1|2^ +£2|^ , 且满足固定约束条件和指定约束条 件, 所述指定约束条件为
Figure imgf000017_0001
< r2
或者, 所述代价函数为^^^^ +^^^^)^ , 且满足固定约束条件和指定约 束条件, 所述指定约束条件为 ;
或者, 所述代价函数为
Figure imgf000017_0002
,且满足固定约束条件和指定约束 条件, 所述指定约束条件为 ei | x < r4
其中, 、 2、 3、 4、 5和 6为予贞设值。
4. 根据权利要求 3所述的方法, 其特征在于, ei =丄, e2 = , e3 = - '
nc Kt n
«c为矩阵 ρ的行数且不等于零, "和 分别为用于控制周期性噪声和非周期性噪 声与干净图像信号的大小的权重, f为将周期性噪声分段而设定的段数, t为每 一段周期性噪声的周期, η=Κίΐ0, Τ。为每一段周期性噪声中 ί的个数。
5. 根据权利要求 1所述的方法, 其特征在于, 在确定所述周期性噪声的 周期这一步骤之前, 所述方法还包括: 确定所述周期性噪声的方向, 如果确定 出所述周期性噪声的方向与非周期性噪声的方向垂直, 对所述待处理图像进行 插值处理, 继续执行周期确定步骤, 如果确定出所述周期性噪声的方向与非周 期性噪声的方向平行, 直接执行周期确定步骤;
其中所述确定周期性噪声的方向这一步骤具体为:
对所述待处理图像进行特征提取;
根据提取到的特征 , 釆用预先训练得到的第一分类器对所述待处理图像进 行分类, 将所述待处理图像识别为带周期性噪声的图像或不带周期性噪声的图 像, 所述第一分类器是通过将带周期性噪声的图像列入正样本训练集、 将不带 周期性噪声的图像列入负样本训练集训练得到, 或者所述笫一分类器是通过将 带周期性噪声的图像列入负样本训练集、 将不带周期性噪声的图像列入正样本 训练集训练得到;
根据所述提取到的特征, 采用预先训练得到的第二分类器对分类得到的带 周期性噪声的图像进行分类, 将所述带周期性噪声的图像识别为第一方向噪声 图像和第二方向噪声图像, 其中所述第一方向噪声图像中周期性噪声的方向与 非周期性噪声的方向平行, 所述第二方向噪声图像中周期性噪声的方向与非周 期性噪声的方向垂直, 所述第二分类器是通过将第一方向噪声图像列入正样本 训练集、 将第二方向噪声图像列入负样本训练集训练得到, 或者所述第二分类 器是通过将第一方向噪声图像列入负样本训练集、 将第二方向噪声图像列入正 样本训练集训练得到。
6. 根据权利要求 5所述的方法, 其特征在于, 所述提取到的特征包括小 波特征; 所述第一分类器和所述第二分类器均为支持向量机分类器。
7. 根据权利要求 1所述的方法, 其特征在于, 所述方法还包括对所述待 处理图像进行分段处理, 得到多段图像数据, 针对每一段待处理图像数据, 执 行所述模型构建步骤和优化求解步骤;
其中, 所述分段处理包括: 将待处理图像分为多段, 其中, 除第一段图像 数据的起始若干块数据与最后一段图像数据的末尾若干块数据外, 每一段图像 数据的起始若干块数据与前一段图像数据的末尾若干块数据重叠, 每一段图像 数据的末尾若干块数据与后一段图像数据的起始若干块数据重叠。
8. 根据权利要求 1所述的方法, 其特征在于, 所述周期性噪声的周期被 确定为预设值, 或者所述周期性噪声的周期通过以下步骤计算而确定包括: 获取感兴趣区域的图像数据;
将感兴趣区域的图像数据和预先设定的候选周期代入信号分离模型, 对信 号分离模型进行干净图像信号、 周期性噪声和非周期性噪声优化求解;
将求解出来的周期性噪声信号最强的模型对应的候选周期确定为周期性噪 声的周期。
9. 根据权利要求 8所述的方法, 其特征在于,
所述对信号分离模型进行干净图像信号、 周期性噪声和非周期性噪声优化 求解的公式包括:
Λ 1 cc 2 β w
u{t) = arg min Dx H u H—— < "g(w)S
1 且约 x,u ,s n— l Kt n
束条件为 x+s+Mu=d;
所述将求解出来的周期性噪声信号最强的模型对应的候选周期确定为周期 性噪声的周期这一步骤对应的表达式为:
Figure imgf000018_0001
其中, 表示感兴趣区域的图像数据, X表示干净图像信号, S表示非周期 性噪声, Μ表示用于重建周期性噪声的变量, 表示周期性噪声, 为转换矩 阵, D为差分矩阵, ^为用于降低或提高非周期性噪声的值的权重, 《为所述感 兴趣区域的图像数据的长度, "和 分别为用于控制周期性噪声和非周期性噪声 与干净图像信号的大小的权重, f为将周期性噪声分段而设定的段数, ί为所述 候选周期, n=KfT0 , Τ。为每一段周期性噪声中 ί的个数, 为所述周期性噪声的 周期。
10. 如权利要求 1所述的方法, 其特征在于, 所述待处理图像包括数字放 射图像, 所述周期性噪声包括栅影。
11. 一种对图像进行噪声抑制的装置, 其特征在于, 包括:
图像获取模块, 用于获取待处理图像, 所述待处理图像为被周期性噪声干 扰的图像;
周期确定模块, 用于确定所述待处理图像中周期性噪声的周期;
优化求解模块, 用于将待处理图像的图像数据和所述周期性噪声的周期代 入信号分离模型, 对信号分离模型进行优化求解, 得到对应于所述待处理图像 的干净图像信号。
12. 根据权利要求 11 所述的装置, 其特征在于, 包括: 所述信号分离模 型包括代价函数和约束条件, 所述代价函数为满足所述约束条件且与干净信号 分量、 周期信号分量和非周期信号分量中的至少一个相关的函数;
所述干净信号分量为关于干净图像信号的平滑度的测度函数, 所述周期信 号分量为关于周期性噪声及其周期的测度函数, 所述非周期信号分量为关于非 周期性噪声的测度函数。
13. 根据权利要求 12 所述的装置, 其特征在于, 所述约束条件包括固定 约束条件, 或者所述约束条件包括固定约束条件和指定约束条件,
所述固定约束条件为: x+s+Mu=d,
所述干净信号分量的表达式包括:
所述周期信号分量的表达式包括: e2 K ,
所述非周期信号分量的表达式包括: e3
Figure imgf000019_0001
,
其中 ί表示待处理图像, X表示干净图像信号, s表示非周期性噪声, u表 示用于重建周期性噪声的变量, Mu表示周期性噪声, M为转换矩阵, ei、 e2 和 e3分别为用于控制干净图像信号、周期性噪声和非周期性噪声的大小的权重, Qc = diag(c)Q , β为待处理图像的幅度谱矩阵, c为权重向量, ^为用于降低或提 高非周期性噪声的值的权重;
所述代价函数为 | x + , 且满足固定约束条件; 或者, 所述代价函数为^ 定约束条件和指定约束条件, 所 述指定约束条件为 e2 Κ < ^
Figure imgf000019_0002
, 或者所述指定约束条件为 或者, 所述代价函数为 e2 K , 且满足固定约束条件和指定约束条件, 所述 指定约束条件为 l x < r4且 e3
Figure imgf000019_0003
< r2 , 或者所述指定约束条件为 ei lfex 1 < ;
或者, 所述代价函数为^ )^ , 且满足固定约束条件和指定约束条件, 所述指定约束条件为
Figure imgf000019_0004
ri , 或者所述指定约束条件为 ei |feX + e2 IHI2 < r6;
或者, 所述代价函数为£1 |2^ +£2 |^ , 且满足固定约束条件和指定约束条 件, 所述指定约束条件为
Figure imgf000020_0001
< r2
或者, 所述代价函数为^^^^ +^^^^)^ , 且满足固定约束条件和指定约 束条件, 所述指定约束条件为 ;
或者, 所述代价函数为
Figure imgf000020_0002
,且满足固定约束条件和指定约束 条件, 所述指定约束条件为 ei | x < r4
其中, 、 2、 3、 4、 5和 6为予贞设值。
14. 根据权利要求 11所述的装置, 其特征在于, 还包括: 方向检测模块, 用于确定所述周期性噪声的方向;
其中所述确定周期性噪声的方向具体为:
对所述待处理图像进行特征提取;
根据提取到的特征 , 釆用预先训练得到的第一分类器对所述待处理图像进 行分类, 将所述待处理图像识别为带周期性噪声的图像或不带周期性噪声的图 像, 所述第一分类器是通过将带周期性噪声的图像列入正样本训练集、 将不带 周期性噪声的图像列入负样本训练集训练得到, 或者所述笫一分类器是通过将 带周期性噪声的图像列入负样本训练集、 将不带周期性噪声的图像列入正样本 训练集训练得到;
根据所述提取到的特征, 采用预先训练得到的第二分类器对分类得到的带 周期性噪声的图像进行分类, 将所述带周期性噪声的图像识别为第一方向噪声 图像和第二方向噪声图像, 其中所述第一方向噪声图像中周期性噪声的方向与 非周期性噪声的方向平行, 所述第二方向噪声图像中周期性噪声的方向与非周 期性噪声的方向垂直, 所述第二分类器是通过将第一方向噪声图像列入正样本 训练集、 将第二方向噪声图像列入负样本训练集训练得到, 或者所述第二分类 器是通过将第一方向噪声图像列入负样本训练集、 将第二方向噪声图像列入正 样本训练集训练得到。
15. 根据权利要求 11 所述的装置, 其特征在于, 所述周期性噪声的周期 被确定为预设值, 或者所述周期性噪声的周期的计算包括:
获取感兴趣区域的图像数据;
将感兴趣区域的图像数据和预先设定的候选周期代入信号分离模型, 对信 号分离模型进行干净图像信号、 周期性噪声和非周期性噪声优化求解;
将求解出来的周期性噪声信号最强的模型对应的候选周期确定为周期性噪 声的周期。
16. 如权利要求 11 所述的装置, 其特征在于, 所述待处理图像包括数字 放射图像, 所述周期性噪声包括栅影。
PCT/CN2014/073461 2014-03-14 2014-03-14 对图像进行噪声抑制的方法及装置 Ceased WO2015135208A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2014/073461 WO2015135208A1 (zh) 2014-03-14 2014-03-14 对图像进行噪声抑制的方法及装置
CN201480077057.XA CN106464778B (zh) 2014-03-14 2014-03-14 对图像进行噪声抑制的方法及装置

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2014/073461 WO2015135208A1 (zh) 2014-03-14 2014-03-14 对图像进行噪声抑制的方法及装置

Publications (1)

Publication Number Publication Date
WO2015135208A1 true WO2015135208A1 (zh) 2015-09-17

Family

ID=54070835

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/073461 Ceased WO2015135208A1 (zh) 2014-03-14 2014-03-14 对图像进行噪声抑制的方法及装置

Country Status (2)

Country Link
CN (1) CN106464778B (zh)
WO (1) WO2015135208A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108111240A (zh) * 2017-12-19 2018-06-01 中国科学院上海天文台 一种射频干扰的数据处理方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101101338A (zh) * 2007-07-10 2008-01-09 廊坊开发区中油油田科技工贸有限责任公司 一种油气勘探方法及系统
CN102725715A (zh) * 2009-10-20 2012-10-10 赛普拉斯半导体公司 减少触控屏幕控制器中的耦合噪声影响的方法和设备
US8346008B2 (en) * 2009-06-09 2013-01-01 Aptina Imaging Corporation Systems and methods for noise reduction in high dynamic range imaging
CN103020913A (zh) * 2012-12-18 2013-04-03 武汉大学 基于分段校正的遥感影像条带噪声去除方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102262775B (zh) * 2010-05-26 2014-12-10 深圳迈瑞生物医疗电子股份有限公司 一种图像降噪方法、装置及图像处理设备
CN102279971B (zh) * 2011-04-06 2013-01-30 中国人民解放军理工大学气象学院 基于变分方法的中分辨率成像光谱仪条带噪声去除方法
CN102750683B (zh) * 2012-06-18 2014-10-29 常州大学 Modis遥感影像中海面条带噪声和条状云的过滤方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101101338A (zh) * 2007-07-10 2008-01-09 廊坊开发区中油油田科技工贸有限责任公司 一种油气勘探方法及系统
US8346008B2 (en) * 2009-06-09 2013-01-01 Aptina Imaging Corporation Systems and methods for noise reduction in high dynamic range imaging
CN102725715A (zh) * 2009-10-20 2012-10-10 赛普拉斯半导体公司 减少触控屏幕控制器中的耦合噪声影响的方法和设备
CN103020913A (zh) * 2012-12-18 2013-04-03 武汉大学 基于分段校正的遥感影像条带噪声去除方法

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108111240A (zh) * 2017-12-19 2018-06-01 中国科学院上海天文台 一种射频干扰的数据处理方法及装置

Also Published As

Publication number Publication date
CN106464778B (zh) 2019-11-05
CN106464778A (zh) 2017-02-22

Similar Documents

Publication Publication Date Title
EP2819091B1 (en) Method and apparatus for processing a gray image
EP3132418B1 (en) Non local image denoising
Ikhsan et al. An analysis of x-ray image enhancement methods for vertebral bone segmentation
Ahmed et al. Medical image denoising system based on stacked convolutional autoencoder for enhancing 2-dimensional gel electrophoresis noise reduction
Gao et al. A deep learning based approach to classification of CT brain images
US9589328B2 (en) Globally dominant point spread function estimation
CN104915930B (zh) 对图像进行灰度补偿和噪声抑制的方法及装置
US11645736B2 (en) Image processing methods, apparatuses and systems
CN104599239A (zh) 一种医学图像中的金属伪影消除方法及装置
CN110428385A (zh) 一种基于无监督对抗神经网络的sd-oct去噪方法
Xu et al. Efficient low‐dose CT artifact mitigation using an artifact‐matched prior scan
CN111612741A (zh) 一种基于失真识别的精确无参考图像质量评价方法
CN120144800B (zh) 人工智能辅助的ct影像质量增强与噪声降低方法
JP5106047B2 (ja) 画像処理方法及び装置
US7508994B2 (en) Method for detecting streaks in digital images
CN114663424B (zh) 基于边云协同的内窥镜视频辅助诊断的方法、系统、设备及介质
CN119417988B (zh) 乳腺结节三维重建方法及系统
CN115359000A (zh) 一种无参考图像质量评价方法、装置及存储介质
WO2015135208A1 (zh) 对图像进行噪声抑制的方法及装置
KR101085949B1 (ko) 폐 분류 장치 및 그 방법
CN104063864B (zh) 基于四元相位叠合模型的图像模糊度评估方法
CN119919310A (zh) 一种ct图像降噪增强系统及方法
EP2693397B1 (en) Method and apparatus for noise reduction in an imaging system
CN111242853A (zh) 基于光流处理的医学ct图像去噪方法
Storozhilova et al. 5 D Extension of Neighborhood Filters for Noise Reduction in 3D Medical CT Images

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: 14885460

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 07.02.2017)

122 Ep: pct application non-entry in european phase

Ref document number: 14885460

Country of ref document: EP

Kind code of ref document: A1