CN111145117A - Spot detection method and system based on soft mathematical morphology operator - Google Patents

Abstract

The invention provides a speckle detection method based on a soft mathematical morphology filter, which comprises the following steps: step 1: inputting a gray level image; step 2: setting parameters of the filter; and step 3: performing SMQ _ A filtering processing on an input image; and 4, step 4: acquiring a candidate spot image; and 5: screening the candidate spots; step 6: performing SMQ _ B filtering processing on the candidate speckle images according to the mask image; and 7: obtaining a filtered image of SMQ _ B; and 8: the spot shape and position are obtained from the SMQ _ B filtered image. The invention verifies that the SMBD has better anti-noise and shape recovery capability than similar spot filters Quoit and N-Quoit based on mathematical morphology from both theory and experiment aspects. In addition, the present invention also compares with different types of the most commonly used LoG and DoH speckle detectors, and SMBD has also obtained significantly better results. The invention provides a speckle detection system based on a soft mathematical morphology filter.

Description

Spot detection method and system based on soft mathematical morphology operator

Technical Field

The invention belongs to the technical field of image processing and computer vision, particularly relates to mathematical morphology filtering, and particularly relates to a spot detection method and system based on a soft mathematical morphology operator.

Background

Speckle generally refers to an area that is differentiated in color or gray from the surrounding area. Since the spot represents a region, compared with a simple corner point, the spot has good stability and strong anti-noise capability. Speckle detection is an important technique of computer vision, and it finds application in many fields: medical image processing [1] [2] [14], video monitoring [15] [16], robot [17] [18] [19], aviation system [20] [21] [22] and the like.

LoG (Laplacian of Gaussian) [4] is a commonly used speckle detector, which is effective for the detection of radially symmetric speckles, but has a relatively high time complexity. In order to reduce the temporal complexity of LoG, document [5] proposes a dog (difference of gaussian) blob detector that replaces the laplacian of gaussian operation of LoG with an approximation of the difference of gaussian operation. In order to improve the performance of LoG for detecting radially asymmetric spots, document [6] proposes a gLoG (generalized laplacian of gaussian) spot detector, which extends the detection to a rotationally asymmetric structure by using different gaussian kernels, thereby being able to detect spots of rotationally symmetric and asymmetric general elliptical structures. In addition, Miao [7] et al propose a blob detection algorithm based on ROLG (rank Order Laplacian of Gaussian), which makes the filter have better robustness to noise by setting rank ordering operation.

Another more commonly used speckle detector is doh (diagnostic of Hessian) [4], which employs a Hessian matrix whose determinants are used to detect speckle on a single scale. In order to extract smaller-scale spots from an image, two spot detectors, HLoG (Hessian-based Laplacian of Gaussian) [8] and HDoG (Hessian-based Difference of Gaussian) [9], are proposed on the basis of DoH, respectively.

The above speckle detection algorithms are not affine invariant, so affine covariance invariant speckle detection algorithms are researched, wherein the research with relatively great influence is that matas [10] proposes a feature detection method of Maximum Stable Extremum Regions (MSERs). The method is used for detecting the local area with the most stable gray level in the image by using the thought of watershed, then the rotation and size normalization are carried out on the detected area, and finally the obtained local features have affine invariance in a strict sense. MSER looks for regions of spots that meet the "minimum rate of area change" measure, rather than looking for spots that meet a specified size.

In addition to the 3 classical speckle detectors LoG, DoH and MSERs and their variants, there are some other types of speckle detection algorithms [11] [12 ].

Document [11] proposes a blob detector SBD (with scale invariance) and a local feature descriptor SLD based on Shearlet (Shearlet is a multi-scale framework with orientation). After transforming the image to the Shearlet domain, the SBD operates by finding local extrema points in the 3 x 3 domain of the adjacent scale space, which have scale invariant properties. That is, although the SBD is named as a blob detector, it is not used to detect blobs that fit a specified size in the image domain, but rather to find local extreme points in the Shearlet transform domain.

Reference [12] proposes a RIFF feature detection algorithm based on radial gradient transformation, which has rotation invariance. The positions of feature points are obtained by DOB (difference of Box), which is to find local extreme points in the 3 × 3 × 3 domain of adjacent scale spaces. That is, RIFF is a feature descriptor, and the feature point detection adopts the DOB algorithm, which is named as a speckle detector, but actually detects the extreme points of the local region, rather than detecting the speckle with a specified size in the image domain.

In addition to the speckle detector described above, there is also a class of speckle detection filters based on mathematical morphology. Document [1] proposes a Quoit filter having a relatively good shape detection capability for radially symmetric spots. Since the time complexity of the quad filter is relatively high, document [2] proposes an N-quad filter, which increases the calculation speed. However, these two algorithms have two major weaknesses: one is weak in noise immunity, and the other is that although spots can be localized, the shape of the spot cannot be detected well.

Disclosure of Invention

The invention firstly popularizes the partition operator of the standard mathematical morphology to the order partition operator of the soft morphology, and designs 3 soft morphology filters by using the partition operator, and further provides a novel spot detection method on the basis, which comprises the following steps:

step 1: inputting a gray level image;

step 2: setting parameters of the filter;

and step 3: performing SMQ _ A filtering processing on an input image;

and 4, step 4: acquiring a candidate spot image;

and 5: screening the candidate spots;

step 6: performing SMQ _ B filtering processing on the candidate speckle images according to the mask image;

and 7: obtaining a filtered image of SMQ _ B;

and 8: the spot shape and position are obtained from the SMQ _ B filtered image.

The filter in step 2 of the invention comprises an SMQ _ A filter and an SMQ _ B filter:

wherein k is_a，k_b＝1，2，...，|D|/2；k_c，k_d＝1，2，...，|R|/2；k_a≤k_b，k_c≤k_d；

Where f is the image to be filtered, k_m，k_n＝1，...，|R|/2，k_m≤k_nR is a ring structural element, D is a disc structural element, (u, v) belongs to D (x, y),

and D is concentric with R.

The step 2 of the invention specifically comprises the following steps:

step 2.1: setting parameters DA, R, k of SMQ _ A filter_a、k_b、k_c、k_d(ii) a DA is the disk structuring element of the SMQ _ a filter;

step 2.2: setting parameters DB, R, k of SMQ _ B filter_m、k_n(ii) a DB is the disk structure element of the SMQ _ B filter;

step 2.3: setting a speckle localization threshold Th_A1And a speckle region threshold Th_A2For controlling the number of blob candidates.

In step 3 of the present invention, an SMQ _ a filter is used to perform speckle filtering on an input image I (x, y) to obtain an anchor point candidate image L (x, y):

L(x，y)＝SMQ_A(D_A，R，k_a，k_b，k_c，k_d，I，x，y)。

in step 5 of the present invention, a mask image M (x, y):

ifL(x，y)＞Th_A1M(x，y)＝1

else M(x，y)＝0。

in step 6 of the present invention, I is set₂(x, y) is the image filtered by SMQ _ B; first, initialize I with 0₂(x, y), then traversing each effective pixel point (x, y) in the image I (x, y), and performing the following filtering processing by using SMQ _ B; v (u, v) is the filtered value of SMQ _ B at (u, v) ∈ DB (x, y);

in step 8 of the present invention, the spot shape is extracted:

if I₂(u，v)≥Th_A2，I₃(u，v)＝I(u，v)

else I₃(u，v)＝0

wherein, I₃(x, y) is the detected speckle image.

Based on the method, the invention also provides a speckle detection system based on the soft mathematical morphology filter, and the system comprises the following steps:

an input module for inputting a grayscale image;

the setting module is used for setting the parameters of the filter;

the device comprises a first processing module, a second processing module and a third processing module, wherein the first processing module is used for carrying out SMQ _ A filtering processing on an input image;

a candidate image acquisition module for acquiring a candidate speckle image;

a screening module for screening candidate blobs;

a second processing module for performing SMQ _ B filtering processing on the candidate speckle images according to the mask image;

a filtered image obtaining module for obtaining a filtered image of SMQ _ B;

an output module for obtaining a blob shape and location from the SMQ _ B filtered image.

The invention provides a soft-form speckle detection method (SMBD for short) for improving the defects of Quoit and N-Quoit speckle filters based on mathematical morphology. The invention firstly populates the basic operator relationship of the standard mathematical morphology to the order relationship of the soft morphology, and then constructs two soft morphology filters SMQ _ A and SMQ _ B by using the order relationship operator, wherein the former is used for positioning the spots, and the latter is used for detecting the shapes of the spots. Finally, a complete speckle detection method SMBD is designed by utilizing SMQ _ A and SMQ _ B filters. In order to verify the performance of SMBD, the method is used for carrying out comparison experiments on a spot image library and 4 methods, namely Quoit, N-Quoit, LoG and DoH, and experiments show that the SMBD is obviously better than other 4 methods in both anti-noise and spot form detection.

Reference to the literature

[1].Yamamoto，Shinji，et al.″Quoit Filter-A new filter based onmathematical morphology to extract the isolated shadow.and its application toautomatic detection of lung cancer in x-ray CT.″Pattern Recognition，1996.，Proceedings of the 13th International Conference on.Vol.2.IEEE，1996.

[2].Miwa，Tomoko，et al.″Automatic Detection of Lung Cancers in ChestCT Images by the Variable N-Quoit Filter.″Systems and Computers in Japan 33.1(2002)：53-63.

[3].Lasse Koskinen，Jaakko T.Astola，and Yrjo A.Neuvo.″Softmorphological filters.″Image Algebra and Morphological Image Processingll.Vol.1568.International Society for Optics and Photonics，1991.

[4].Lindeberg，Tony.″Feature detection with automatic scaleselection.″International journal of computer vision 30.2(1998)：79-116.

[5].Hay，Geoffrey J.，et al.″A scale-space primer for exploring andquantifying complex landscapes.″Ecological Modelling 153.1-2(2002)：27-49.

[6].Kong，Hui，Hatice Cinar Akakin，and Sanjay E.Sarma.″A generalizedLaplacian of Gaussian filter for blob detection and its applications.″IEEEtransactions on cybemetics 43.6(2013)：1719-1733.

[7].Miao，Zhenwei，and Xudong Jiang.″Interest point detection usingrank order LoG filter.″Pattern Recognition 46.11(2013)：2890-2901.

[8].Zhang，Min，Teresa Wu，and Kevin M.Bennett.″Small blobidentification in medical images using regioRal features from optimum scale.″IEEEtransactions on biomedical engineering62.4(2015)：1051-1062.

[9].Zhang，Min，et al.″Efficient small blob detection based on localconvexity，intensity and shape information.″IEEE transactions on medicalimaging 35.4(2016)：1127-1137.

[10].Matas，Jiri，et al.″Robust wide-baseline stereo from maximallystable extremal regions.″Image and vision computing 22.10(2004)：761-767.

[11].Duval-Poo，Miguel A.，et al.″Scale invariant and noise robustinterest points with shearlets.″IEEE Transactions on Image Processing 26.6(2017)：2853-2867.

[12].Takacs，Gabriel，et al.″Fast computation of rotation-invariantimage features by an approximate radial gradient transform.″IEEE Transactionson Image Processing 22.8(2013)：2970-2982.

[13].Kurzendorfer，Tanja，et al.″Cryo-Balloon Catheter LocalizationBased on a Support-Vector-Machine Approach.″IEEE transactions on medicalimaging 35.8(2016)：1892-1902.

[14].Zhang，Min，et al.″Efricient small blob detection based on localconvexity，intensity and shape information.″IEEE transactions onmedicalimaging 35.4(2016)：1127-1137.

[15].Zhang，Guyue，et al.″Physical blob detector and Multi-ChannelColor Shape Descriptor for human detection.″Journal of Visual Communicationand Image Representation52(2018)：13-23.

[16].Afsar，Palwasha，Paulo Cortez，and Henrique Santos.″Automatic humantrajectory destination prediction from video.″Expert Systems withApplications 110(2018)：41-51.

[17].Hu，Yuting，Zhiling Long，and Ghassan AlRegib.″A High-Speed，Real-Time Vision System for Texture Tracking and Thread Counting.″IEEE SignalProcessing Letters 25.6(2018)：758-762.

[18].Durand-Petiteville，A.，S.Vougioukas，and D.C.Slaughter.″Real-timesegmentation of strawberry flesh and calyx from images of singulatedstrawberries during postharvest processing.″Computers and Electronics inAgriculfure 142(2017)：298-313.

[19].Rajasekaran，Keshav，et al.″An accurate perception method for Iowcontrast bright field microscopy in heterogeneous microenvironments.″AppliedSciences 7.12(2017)：1327.

[20].Ewald，Vincentius，Roger M.Groves，and Rinze Benedictus.″TransducerPlacement Option of Lamb Wave SHM System for Hotspot Damage Monitoring.″Aerospace 5.2(2018)：39.

[21].Xin，Long，Delin Luo，and Han Li.″A monocular visual measurementsystem for UAV probe-and-drogue autonomous aerial refueling.″InternationalJournal of Intelligent Computing and Cybernetics just-accepted(2018)：00-00.

[22].Du，Yuhan，Yicheng Jiang，and Wei Zhou.″An Accurate Two-Step ISARCross-Range Scaling Method for Earth-Orbit Target.″IEEE Geoscience and RemoteSensing Letters 14.11(2017)：1893-1897.

Drawings

FIG. 1 is a flow chart of the speckle detection method SMBD of the present invention.

Fig. 2 is an input image.

FIG. 3 is a candidate blob anchor point image obtained after SMQ _ A filtering.

Fig. 4 is an image obtained by performing binarization processing on fig. 3.

FIG. 5 is an image of SMQ _ B morphologically filtering blobs at candidate waypoints.

Fig. 6 is a finally obtained spot shape image.

Detailed Description

The invention is further described in detail with reference to the following specific examples and the accompanying drawings. The procedures, conditions, experimental methods and the like for carrying out the present invention are general knowledge and common general knowledge in the art except for the contents specifically mentioned below, and the present invention is not particularly limited.

The invention firstly populates the partition operator of the standard mathematical morphology to the orderpartition operator of the soft morphology, and designs 3 soft morphology filters by using the partition operator, and further provides a novel speckle detection method SMBD. These are described separately below.

(1) Design of order division operator of soft morphology

Koskinen et al [3] generalize the partition operator of standard mathematical and physical morphology to soft partition, which is mainly characterized by being relatively insensitive to additive noise and relatively insensitive to shape changes requiring filtering. The definition of soft variance is as follows:

the soft morphological operators are defined on the basis of ordering statistics using the structural system B, A, k]Replacing the standard mathematical morphological structural elements. The structural system contains 3 parameters: the set of finite planes a and B,

one satisfies

The natural number k, |, represents the total number of elements in the set. B is called the structure set, A is the hard center of B, B-A gives its soft contour, k is its sequence number. Symbol denotes a repetition of the operation, defined as

For example, {2 × 1, 2, 3 × 3} ═ 1, 1, 2, 3, 3, 3). Obviously, when a is equal to Φ and k is equal to 1, the soft variance degenerates to the standard variance.

In order to design the soft-form Quoit filter, a sequential form partitioning operator is defined on the basis of a soft partitioning operator.

Definition of order diagnosis:

here, B is a structural element, k is a natural number satisfying 1. ltoreq. k.ltoreq. B/2, and ordinal numbers are ordered from large to small. Formula (2) is actually to change the hard center a in the structural system [ B, a, k ] in formula (1) to the empty set Φ, thereby degrading the structural system to [ B, k ]. This means that the morphological dilation of the image f with the soft contours (B-a) and hard centers a originally becomes a shape dilation of the image f with only the soft shapes of the structuring element B.

(2) Design of 3 soft-state Quoit filters

The following 3 soft-form join filters are defined first using the order partitioning operator.

(a) Definition of SMQ filter:

(b) definition of SMQ _ a filter:

here, k_a，k_b＝1，2，...，|D|/2；k_c，k_d＝1，2，...，|R|/2；k_a≤k_b，k_c≤k_d。

(c) Definition of SMQ _ B filter:

where f is the image to be filtered, k_m，k_n＝1，...，|R|/2，k_m≤k_nR is a ring structural element, D is a disc structural element, (u, v) belongs to D (x, y),

and D is concentric with R.

Next, the operation principle of the 3 kinds of soft-state quaot filters will be described.

■ SMQ filter working principle

Intuitively, the Quoit filter actually moves concentric discs and rings to an object point p and then respectively finds out the maximum value V of the disc field_DAnd maximum value V of the annular region_RThrough V_p＝V_D-V_RTo determine whether the target point P belongs to the blob area. The SMQ filter is similar to the quait filter except that it finds its own assigned ordinal value in the disc and ring fields, respectively, instead of finding the maximum value, thus effectively avoiding the effects of noise and morphological changes. K of SMQ filter_DAnd k_RSet to 1, it degenerates to a Quoit filter, and if D is set to an element, it degenerates to an N-Quoit filter. By adjusting the ordinal km and kn, the robustness of the SMQ to noise and morphological changes can be adjusted. The SMQ filter is mainly used for spot localization.

■ SMQ _ A filter operating principle

As can be seen from the above description of the principle of SMQ, SMQ is a morphological filter with assigned order bits, and SMQ _ a is an extension of SMQ. When the spots are morphologically filtered by the structural elements of the disk and the ring, the SMQ _ a is not based on the gray value corresponding to a certain order but on the average value of the gray values corresponding to a certain segment of order. Thus, the SMQ _ a is theoretically more robust to shape than an SMQ based on only a certain ordinal. When the sequence k_a＝k_bAnd k_b＝k_dWhen this happens, SMQ _ A degrades to SMQ. The SMQ _ a filter is mainly used for spot localization.

■ SMQ _ B filter operating principle

Both SMQ and SMQ _ a locate the blob only and cannot detect the actual shape of the blob because the morphological filter of the object point only takes into account the morphological changes of that point and not the morphological changes of other points in the disk field. In order to solve the problem, the SMQ _ A is modified, the original disc structural element D is decomposed into the structural element with only a single element (the image is processed by the dilaton based on the structural element with the single element), and the SQM-A is applied by the structural element to respectively carry out morphological filtering on each point (u, v) in the disc field (but R is always kept to carry out filtering on the object point (x, y)). In other words, when filtering the target point (x, y), the SQM-B filter calculates the difference between each point (u, v) and (x, y) based on the R soft morphological filter response value (right part of equation 12) in the range covered by D with the point (x, y) as the center of the circle, so that the SMQ _ B can detect the shape of the region covered by D at the target point (x, y) while morphological filtering is performed at the target point (x, y). With the ring-structural element R in the order k_m～k_nThe soft-form filtering process (right part of equation 12) is performed on the point (x, y) within the range, which essentially changes the filtering effect of the filter on the spot shape at that point by adjusting the filter response value of that point. Because it adopts in-order bit k_m～k_nThe mean value between them is taken as the morphological filtering response value, so it is theoretically more stable than the morphological filtering based on only a single ordinal.

(3) Design of speckle detection algorithm SMBD

SMQ is a Quoit filter based on soft form partitioning operator designed by the invention, which comprises the Quoit filter and an N-Quoit filter, and SMQ _ A and SMQ _ B are extensions of SMQ. In the following, a bright-speckle detection algorithm SMBD based on SMQ _ a and SMQ _ B filters is given, and if dark speckles are to be detected, it is sufficient to invert the gray scale of the input image.

SMBD algorithm

Step 1: setting filter parameters according to Table 1

(1) Setting parameters DA, R, k of SMQ _ A filter_a、k_b、k_c、k_dWhere DA is the disk structure element of the SMQ _ a filter.

(2) Setting parameters DB, R, k of SMQ _ B filter_m、k_nWhere DB is the disk structure element of the SMQ _ B filter.

(3) Setting a speckle localization threshold Th_A1And a speckle region threshold Th_A2The two thresholds are mainly used to control the number of blob candidates.

Table 1: estimation of SMBD parameters

Step 2: SMQ _ A filtering

Performing speckle filtering on the input image I (x, y) by using an SMQ _ A filter to obtain an anchor point candidate image L (x, y):

L(x，y)＝SMQ_A(D_A，R，k_a，k_b，k_c，k_d，I，x，y)

and step 3: generating mask image M (x, y)

if L(x，y)＞Th_A1M(x，y)＝1

else M(x，y)＝0

And 4, step 4: SMQ _ B filtering

Let I₂(x, y) is the image filtered by SMQ _ B. First, initialize I with 0₂(x, y), then traversing each valid pixel point in the image I (x, y)(x, y), the following filtering process is performed with the SMQ _ B. Here, v (u, v) is a filtered value of SMQ _ B at (u, v) ∈ DB (x, y).

And 5: extracting blob shapes

if I₂(u，v)≥Th_A2，I₃(u，v)＝I(u，v)

elseI₃(u，v)＝0

Here I₃(x, y) is the detected speckle image.

The invention also provides a speckle detection system based on the soft mathematical morphology filter, which comprises the following steps:

an input module for inputting a grayscale image;

the setting module is used for setting the parameters of the filter;

the device comprises a first processing module, a second processing module and a third processing module, wherein the first processing module is used for carrying out SMQ _ A filtering processing on an input image;

a candidate image acquisition module for acquiring a candidate speckle image;

a screening module for screening candidate blobs;

a second processing module for performing SMQ _ B filtering processing on the candidate speckle images according to the mask image;

a filtered image obtaining module for obtaining a filtered image of SMQ _ B;

an output module for obtaining a blob shape and location from the SMQ _ B filtered image.

Examples

The design principles of the present invention are further illustrated in the following figures and specific examples:

step 1: setting all parameters in SMBD method

The image 2 is an image obtained by superimposing salt and pepper noise of 5% on the original image. The object of the present invention is to detect the speckle of radius 26 pixels in the 3 rd to 4 th rows of the image 2, and to provide the following parameters according to the parameter estimation method of Table 1Numerical values: r is_A1＝5，r_A2＝28，r_A3＝9，r_A4＝26，

k_d＝3k_c＝27，k_m＝k_c＝9，k_n＝k_d＝27，Th_A1＝60，Th_A2＝10。

Step 2: the SMQ _ A filter is used for positioning the spots in the input image, wherein the spots are smaller than the inner radius of the designated circle, and the radius of the circle can be smaller in order to improve the processing speed. Since the computation is very large when the SMQ _ B is used for spot filtering directly, the method firstly uses the SMQ _ A to roughly determine the position of the central area of the spot. FIG. 3 shows candidate blob anchor points obtained after SMQ _ A filtering, and it can be seen from the figure that the larger the blob area is, the smaller the corresponding anchor point area is.

And step 3: and screening the spot candidate positioning points to generate a mask image.

And 4, taking a point with the value of 1 in the mask image as the center of the SMQ _ B filter, and then carrying out SMQ _ B filtering processing on the image so as to obtain a filtering response value image of the SMQ _ B, namely the detected speckle shape response value. FIG. 4 shows a threshold Th_A1The image of FIG. 3 is subjected to binarization processing, which can eliminate those images whose filtering values are less than Th_A1The erroneous candidate anchor point of (2). FIG. 5 is an image of SMQ _ B morphologically filtering blobs at candidate waypoints.

And 5: from the filtered image of SMQ _ B, a blob shape whose filter response value is greater than the threshold value ThA2 is extracted from image 2. FIG. 6 is a diagram for extracting a speckle pattern filter value larger than a threshold Th from an original image_A2The true spot shape.

The invention verifies that the SMBD has better anti-noise and shape recovery capability than similar spot filters Quoit and N-Quoit based on mathematical morphology from both theory and experiment aspects. In addition, the present invention also compares with different types of the most commonly used LoG and DoH speckle detectors, and SMBD has also obtained significantly better results.

The embodiments described herein are intended to assist the reader in understanding the principles of the invention and it is to be understood that the scope of the invention is not limited to such specific statements and embodiments. All such possible equivalents and modifications are deemed to fall within the scope of the invention as defined in the claims.

The protection of the present invention is not limited to the above embodiments. Variations and advantages that may occur to those skilled in the art may be incorporated into the invention without departing from the spirit and scope of the inventive concept, and the scope of the appended claims is intended to be protected.

Claims (8)

1. A speckle detection method based on a soft mathematical morphology filter is characterized by comprising the following steps:

step 1: inputting a gray level image;

step 2: setting parameters of the filter;

and step 3: performing SMQ _ A filtering processing on an input image;

and 4, step 4: acquiring a candidate spot image;

and 5: screening the candidate spots;

step 6: performing SMQ _ B filtering processing on the candidate speckle images according to the mask image;

and 7: obtaining a filtered image of SMQ _ B;

and 8: the spot shape and position are obtained from the SMQ _ B filtered image.

2. The speckle detection method based on the soft mathematical morphology filter of claim 1, wherein the filter in step 2 comprises an SMQ _ a filter and an SMQ _ B filter:

wherein k is_a，k_b＝1，2，...，|D|/2；k_c，k_d＝1，2，...，|R|/2；k_a≤k_b，k_c≤k_d；

Where f is the image to be filtered, k_m，k_n＝1，...，|R|/2，k_m≤k_nR is a ring structural element, D is a disc structural element, (u, v) belongs to D (x, y),

and D is concentric with R.

3. The speckle detection method based on the soft mathematical morphology filter as claimed in claim 2, wherein the step 2 specifically comprises:

step 2.1: setting parameters DA, R, k of SMQ _ A filter_a、k_b、k_c、k_d(ii) a DA is the disk structuring element of the SMQ _ a filter;

step 2.2: setting parameters DB, R, k of SMQ _ B filter_m、k_n(ii) a DB is the disk structure element of the SMQ _ B filter;

step 2.3: setting a speckle localization threshold Th_A1And a speckle region threshold Th_A2For controlling the number of blob candidates.

4. The speckle detection method based on the soft mathematical morphology filter as claimed in claim 2, characterized in that in step 3, the SMQ _ a filter is used to perform speckle filtering on the input image I (x, y) to obtain the anchor point candidate image L (x, y):

L(x，y)＝SMQ_A(D_A，R，k_a，k_b，k_c，k_d，I，x，y)。

5. the speckle detection method based on the soft mathematical morphology filter of claim 1, characterized in that in step 5, a mask image M (x, y):

if L(x，y)＞Th_A1M(x，y)＝1

else M(x，y)＝0。

6. the speckle detection method based on the soft mathematical morphology filter of claim 2, wherein in step 6, I is set₂(x, y) is the image filtered by SMQ _ B; first, initialize I with 0₂(x, y), then traversing each effective pixel point (x, y) in the image I (x, y), and performing the following filtering processing by using SMQ _ B; v (u, v) is the filtered value of SMQ _ B at (u, v) ∈ DB (x, y);

If M(x，y)＝1，

do{v(u，v)＝SMQ_B(D_B，R，k_m，k_n，I，x，y，u，v)

if v(u，v)＞I₂(u，v)，I₂(u，v)＝v(u，v)

}。

7. the method for speckle detection based on a soft mathematical morphology filter according to claim 1, characterized in that in step 8, the speckle shape is extracted:

if I₂(u，v)≥Th_A2，I₃(u，v)＝I(u，v)

else I₃(u，v)＝0

wherein, I₃(x, y) is the detected speckle image.

8. A speckle detection system based on a soft mathematical morphological filter, characterized in that it uses a detection method according to any one of claims 1 to 7, said system comprising the following steps:

an input module for inputting a grayscale image;

the setting module is used for setting the parameters of the filter;

the device comprises a first processing module, a second processing module and a third processing module, wherein the first processing module is used for carrying out SMQ _ A filtering processing on an input image;

a candidate image acquisition module for acquiring a candidate speckle image;

a screening module for screening candidate blobs;

a second processing module for performing SMQ _ B filtering processing on the candidate speckle images according to the mask image;

a filtered image obtaining module for obtaining a filtered image of SMQ _ B;

an output module for obtaining a blob shape and location from the SMQ _ B filtered image.

Images