CN106096613A - Image multi-target detection method and device based on corner feature - Google Patents
Image multi-target detection method and device based on corner feature Download PDFInfo
- Publication number
- CN106096613A CN106096613A CN201610380344.9A CN201610380344A CN106096613A CN 106096613 A CN106096613 A CN 106096613A CN 201610380344 A CN201610380344 A CN 201610380344A CN 106096613 A CN106096613 A CN 106096613A
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- target
- angle point
- fiber
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
-
- 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
- G06T2207/20112—Image segmentation details
- G06T2207/20164—Salient point detection; Corner detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The present invention proposes a kind of image multi-target detection method based on Corner Feature and device, by the physical features such as angle point, skeleton and the curvature sequence that fine class image is contained are analyzed with abstract, study such image and can be used for computer understanding and the structured representation method of high compression ratio transmission；Utilize angle point and framework characteristic to describe irregular target area, it is possible to be generalized to relate to other application scenarios of image target area detection, be also beneficial to so that the category identification of blend fibre is towards intelligent, automation direction development.The present invention intends exploring the blind restoration algorithm being suitable for edge line gray scale fuzzy characteristics with light microscope enlarged drawing picture for object, the Corner Detection related with image local feature and be extracted as target, research structure feature hierarchy represents and the description method based on spatial neighborhood correlation, to forming some strategic structural, extract and character representation problem offer effective technical way for solving image texture.
Description
Technical field
The invention belongs to image identification technical field, be specifically related to a kind of image multi-target detection side based on Corner Feature
Method and device.
Background technology
Image multiple target under complicated multifactor interference extracts and effective expression has a wide range of applications background and important section
Learn meaning.In the extraction and expression of multiple image objects, local feature description lays particular emphasis on whole local descriptions connection
Realize the final expression to image altogether.Angle point is exactly the key character describing image local target.Common Corner Detection
Method includes: the method based on edge contour, the method based on gradient of image and gray scale and the method based on parameter model.
Except angle point, the fibre image feature interpretation of non-regular shape can also assist description by means of skeleton, so
The separation that certain overlapping region angle point can be avoided to cause when more is chaotic, this is because single separation entity is marked with by connected domain
Effect, but then usually can judge for overlapping fibers image.Image framework contains the effective digital information of characteristics of image,
Image essential characteristic effectively can be described, for the sign or discrete of each individuality in overlapping fibers, need to do deeper
The algorithm design entering.
With the fibrogram unlike conventional regular figure contours extract, in the such as fine class image of complicated overlapping image object
Shape presents variation, and single fiber has certain flexion torsion, and plurality of fibers then may be overlapping, figure exist multiple crosspoint and
Rise and fall, need to gather more minutia and describe.Main issue is as follows:
(1) edge treated in complicated overlay target region and feature point detection problem in image:
Conventional Edge extraction typically uses morphological method, utilizes image expansion and filling to obtain profile, but
Often result in change and the distortion of image real area, more preferable profile can be obtained in conjunction with Canny algorithm and multi-channel filter
Extraction effect, but still bring certain distortion can to subsequent fiber parameter measurement；For intersection or overlapping fine class image, obtain wheel
Crucial angle point in wide figure is the key of solution problem, uses pixel curvature, oval Support domain method can delete some puppets
Point, but in irregular fiber graphics process, also need to combine some novelty processing schemes.
(2) mark problem when multiple target region is overlapping:
Utilize in the image that optical loupes obtains and usually contain multiple individual fibers, the individuality of these fibers is marked successively
Know and statistics, there is very important practical significance.Current image separation and identification technology typically use " connected domain " algorithm,
This algorithm is individual effective for separate in image, but is not suitable for overlapped individual fibers.
(3) the structuring characterization problems based on angle point correlation:
At present fine class image still being lacked to the structuring expression discussion of system, goal in research is concentrated mainly on fiber
Separate the measurement with diameter.Fine class image is as the special class in numerous image members, due to the flexible nature of fiber itself,
Often lead to multiple fiber overlapping difficult with separation, bring difficulty to the feature extraction of single fiber, it is difficult to use traditional number
Learn modeling to realize accurately describing.In previously studying, discovery uses edges of regions pixels statistics method to there is accuracy deficiency
Problem.
Content of the invention
For solving problems of the prior art, the present invention proposes a kind of target detection side based on Corner Feature
Method, to realize detection automation, replaces the manual operation of current poor efficiency, improves to fiber target identification accurate in image
Rate.
The present invention realizes especially by following technical scheme:
A kind of object detection method based on Corner Feature, said method comprising the steps of:
S1: gather the complicated overlay chart picture of fine class；
S2: carry out to complicated overlay chart picture pre-processing, Corner Detection；
S3: based on feature extraction and the sign of angle point and skeleton knowledge；
The structured features modeling of S4: complicated overlay chart picture；
S5: fiber target identification；Wherein,
Image semantic classification process in described step S2 includes: the binaryzation-> image filtering-> Image Reconstruction-> figure of image
Image intensifying, described image filtering includes that morphological image is processed and edge-smoothing；
Described step S3 is specially the angle point sequence obtaining image and skeleton describes, and angle point includes crosspoint and other angles
Point, utilizes skeletal extraction method to be labeled the single individuality in multiple overlay target, but crosspoint and end points in skeletal extraction
Between there will be multiple line segment, general public line segment occurs in intersection region, needs to be deleted；But overlapping fibers more when
Time there will be " pseudo-public line segment ", now needs to be labeled by means of Intelligent Recognition algorithm；Each image is regarded as one
Set F, the then corresponding subclass F of the edge contour curve of every fiber of overlapping fibers figurei, it comprises a series of phase
Like point；To overlapping fibre image collection F={F1,F2,…,FM, utilize clustering algorithm to classify contour images, find each picture
Vegetarian refreshments PjAffiliated fiber collection Fi, each individual fibers is identified；
Described step S4 is particularly as follows: using the local feature of fine class image as primary image Expressive Features, and then with local
Scale correlations between feature and spatial coherence are foundation, it is achieved the expression to picture structure.
Further, in described step S2, the multi-channel filter bank based on spatial domain convolution kernel masterplate is utilized to come to image
Complex background noise carry out multiple dimensioned multi-direction suppression.
Further, in described step S2, morphological image processes the image outline detection using based on hot spot diffusion model
Algorithm realizes: the hot spot of search target individual, diffusion hot spot to edge contour, memory profile；The essence of hot spot diffusion is logical
Cross search closed loop territory to obtain the edge contour of fibre image, in order to avoid edge transition expand, this algorithm method also often combines
Sobel, Canny or Roberts operator improves edge extracting effect.
Further, in described step S2, morphological image processes the back of the body using B-spline surface matching to obtain fibre image
Then pristine fibre image subtracting background is obtained target image by scape, reaches the mesh removed image background and solve uneven illumination
's；Again target image is carried out the overall situation binaryzation, binary image is filled with process, mark binary image in each right
As the quantity of number of pixels according to contained by object carries out denoising, obtains target fibers image accurately.
Further, in described step S2, the extraction of angle point, based on pixel curvature, counts curvature and changes from small to big, or
Point from large to small, as candidate's flex point, is set up oval supporting zone for candidate's flex point, is utilized supporting zone to carry out this flex point
Judge.
Further, described step S4 is particularly as follows: give local feature set F={fi, fi=(pi,vi,i,si), wherein
pi、vi、iAnd siIt is respectively position, apparent description, principal direction and the scale-value of feature, then mould expressed by stratification picture structure
Type is defined as follows,
M={{vi},{Rspatial},{Rscale}}。
This model is made up of three parts, apparent message part { vi, the space correlation relation { R between featurespatialAnd
Yardstick dependency relation { Rscale}；Then both can use defined level based on space with scale factor to associate
System.
The invention allows for a kind of object detecting device based on Corner Feature, it is characterised in that: described device includes:
For gathering the module of the complicated overlay chart picture of fine class；For carrying out to complicated overlay chart picture pre-processing, the module of Corner Detection；With
In the module based on the feature extraction of angle point and skeleton knowledge and sign；Structured features modeling for complicated overlay chart picture
Module；Module for fiber target identification；Wherein,
Described Image semantic classification process includes: the binaryzation-> image filtering-> Image Reconstruction-> image enhaucament of image, institute
State image filtering and include that morphological image is processed and edge-smoothing；
The described angle point sequence for obtaining image based on the module of the feature extraction of angle point and skeleton knowledge and sign and
Skeleton describes, and angle point includes crosspoint and other angle points, utilizes skeletal extraction method to enter the single individuality in multiple overlay target
Rower is noted, but there will be multiple line segment in skeletal extraction between crosspoint and end points, and general public line segment occurs in intersection region,
Needs are deleted；But there will be " pseudo-public line segment " when overlapping fibers is more, now need to calculate by means of Intelligent Recognition
Method is labeled；Each image is regarded as set F, then an edge contour curve pair for every fiber of overlapping fibers figure
Answer a subclass Fi, it comprises a series of similitude；To overlapping fibre image collection F={F1,F2,…,FM, utilize cluster
Contour images is classified by algorithm, finds each pixel PjAffiliated fiber collection Fi, each individual fibers is identified；
The module of the described structured features modeling for complicated overlay chart picture is using the local feature of fine class image as base
This iamge description feature, and then with the scale correlations between local feature and spatial coherence as foundation, it is achieved image is tied
The expression of structure.
The invention has the beneficial effects as follows: the present invention is by angle point, skeleton and the curvature sequence etc. that are contained fine class image
Physical feature be analyzed with abstract, study such image can be used for computer understanding and high compression ratio transmission structured representation
Method；Utilize angle point and framework characteristic to describe irregular target area, it is possible to be generalized to relate to image target area detection
Other application scenarios, are also beneficial to so that the category identification of blend fibre is towards intelligent, automation direction development.The present invention intend with
Light microscope enlarged drawing picture is that object explores the blind restoration algorithm being suitable for edge line gray scale fuzzy characteristics, with image local feature
Related Corner Detection and be extracted as target, research structure feature hierarchy represents and the structuring based on spatial neighborhood correlation
Description method, to forming some strategic structural, is provided with effect technique way for solving image texture extraction and character representation problem
Footpath.
Brief description
Fig. 1 is the cotton fibriia under light microscope；
Fig. 2 is the image outline detection flow chart based on hot spot diffusion model.
Detailed description of the invention
The present invention is further described for explanation and detailed description of the invention below in conjunction with the accompanying drawings.
1) pretreatment of complicated overlay chart picture and Corner Detection
Linen-cotton or Cashmere and Woolens fiber, due to the physical characteristic of self, there will be fibre image gray scale not in sample making course
Uniformly, the series of problems such as fiber overlap, shape distortion, as shown in Figure 1, these overlapping, profiled filaments of distortion are to accurately
Positioning and parameter extraction have adverse effect on.
Image semantic classification process includes: binaryzation-> image filtering (the smoothing processing)-> Image Reconstruction of image-> image increases
By force.Wherein morphological image is processed and edge-smoothing is the emphasis that the present invention studies.
Background noise suppression and Morphological scale-space
The fine class image that light microscope obtains target area usually occurs and background image contrast is low, without obvious limit
Edge, segmentation and the feature extraction of target area can be caused severe jamming by the therefore existence of background information, and the present invention devises base
Carry out multiple dimensioned multi-direction suppression in the multi-channel filter bank of the spatial domain convolution kernel masterplate complex background noise to image.
Except binary conversion treatment process, in order to obtain the profile of image, mathematical morphology is used for gray level image field, can
With utilize the half-tone information of structural elements go obtain target gray distribution features, major calculations include burn into expand, opening operation and
Closed operation, the problem that so can improve soft edge.Opening operation in mathematical morphology is usually used to remove tiny
Object (such as noise, burr, corner angle etc.), increase the space between target, the basic profile of the object simultaneously keeping big is constant；Close fortune
Calculate and be then used for filling up hole between interior of articles and object or gap, and keep the basic profile of object constant.Said method
Deficiency be that expansion process causes graphics area to increase, cause follow-up Inflexion extracting and parameter measurement to have error.
For the Morphological scale-space of image, the present invention devises two kinds of algorithms and realizes:
One of method is to consider hot spot diffusion model, makes the individual fibers in computer simulation human eye pair cross-section image carry out
Profile detects, and basic step is for as shown in Figure 2.The essence of hot spot diffusion obtains fibre image by searching for closed loop territory
Edge contour, in order to avoid edge transition expands, the method also often combines Sobel, Canny or Roberts operator to improve limit
Edge extraction effect, but the early-stage Study of research and inventor finds both at home and abroad, and this process also results at fiber adhesion sometimes
Easily excessively expanding makes original concave point feature disappear, and false border and dual edge phenomenon occurs in design sketch.Result is in reality
It is applied to profiled filament micro-image, lost efficacy when extracting fibre section profile.This by follow-up fiber geometrical Characteristics Analysis and
Calculating brings very big error.
The two of method are the backgrounds using B-spline surface matching to obtain fibre image, then by pristine fibre figure image subtraction
Background obtains target image, reaches the purpose removed image background and solve uneven illumination.Carry out the overall situation two again to target image
Value, is filled with to binary image processing, each object in mark binary image, number of pixels according to contained by object
Quantity carries out denoising.Obtain target fibers image accurately.The shape that partly overlaps fiber treatment result shows, the method ratio is original
The fibre image profile that Image semantic classification obtains especially intersects, become apparent from adhesion, successfully solve image intersect, adhesion
The problem of place's distortion.Next step will consider the generalization of this innovation algorithm, is allowed to be applied to the overlapping complicated fiber of more Multiple Shape
Process.
The Corner Detection of target area
Image after Morphological scale-space includes multiple flex point (angle point), and they are the key factors of reflection characteristics of image,
But image also likely to be present " burr ", bring interference to follow-up Inflexion extracting, it is therefore desirable to be filtered image
Process, eliminate some unnecessary details.
In image, the extraction of angle point is based on pixel curvature, counts curvature and changes from small to big, or point from large to small is made
For candidate's flex point.But often existing in this contour images " pseudo-flex point ", the present invention is that candidate's flex point sets up oval Support
Territory, utilizes supporting zone to judge this flex point.Experiment shows, when the pixel that supporting zone comprises is more, it is judged that knot
Fruit is just closer to truly.Further research needs to carry out intelligent decision to the flex point of irregular fibre image, needs to existing
Model is perfect further.
2) feature extraction and the characterizing method based on angle point and skeleton knowledge
Obtain the angle point sequence of image and skeleton describes, and then research image object Feature Points Extraction, comprising:
Image intersection angle point grid and common wire segment identification
Fine class figure there will be in overlapping profile diagram more crosspoint, but these crosspoints and other angle points are arranged side by side.As
What is the need except the impact of other angle points, extract these crosspoints, be conducive to carrying out separating to individual fibers and carry out next step
Crucial parameter measurement.
Single individuality in multiple overlay target is labeled by research skeletal extraction method.But crosspoint in skeletal extraction
And between end points, there will be multiple line segment, general public line segment occurs in intersection region, needs to be deleted；But overlapping fibers is relatively
Also there will be " pseudo-public line segment " when many, now need to be labeled by means of Intelligent Recognition algorithm.
Angle point sequence description based on chain code
Each image is regarded as set F, then an edge contour curve correspondence one for every fiber of overlapping fibers figure
Individual subclass Fi, it comprises a series of similitude.
To overlapping fibre image collection: F={F1,F2,…,FM,
The essence of clustering algorithm is that contour images is classified, and finds each pixel PjAffiliated fiber collection Fi, to each fiber
Individuality is identified.
Due to same FiIn each pixel PjCurvature is approximately within the specific limits, and the present invention attempts utilizing skeleton to carry
Take and combine pixel curvature, ownership judgement can be carried out to each point.
It is suitable for detection of characteristic parameters and the characterizing method of irregular fine class image
The identification of blend fibre and classification have numerous research, and extracting characteristic value is still current maximally effective means.From existing
From the point of view of achievement in research, the fibre diameter coefficient of variation, fiber reinforcement direction rate of change, the irregular selection of fiber fragment are all effectively to identify
The feature of cotton fibriia, but linen-cotton overlap may cause figure to intersect and distortion abnormity, brings dry to conventional parameter detecting
Disturb.
By exploring the basic texture primitive of fine class image, use for reference conventional images Similarity of Local Characteristic Structure and represent scheme, select
Suitable mathematical expression mode, can carry out fluction analysis according to their this characteristic, first realizes to fine class image different
The structural description of target.
3) the structured features Modeling Research of complicated overlay chart picture
The present invention uses the local feature of fine class image as primary image Expressive Features, so with local feature between
Scale correlations and spatial coherence are foundation, it is achieved the expression to picture structure.
Formally, local feature set F={f is giveni, fi=(pi,vi,i,si), wherein pi、vi, i and siIt is respectively
The position of feature, apparent description, principal direction and scale-value, then stratification picture structure expression model is defined as follows,
M={{vi},{Rspatial},{Rscale}}。
This model is made up of three parts, apparent message part { vi, the space correlation relation { R between featurespatialAnd
Yardstick dependency relation { Rscale}.Then both can use defined level based on space with scale factor to associate
System.
Based on above-mentioned model, the concrete level incidence relation based on space and scale factor is defined, then permissible
Obtain the image spatial feature expression of stratification, it is achieved expressing at many levels and portraying to image information.
In sum, the present invention designs and in terms of actual application scenarios inspection three from fundamental research, key algorithm
Carry out project key technology research.Wherein, basic theory aspect is mainly special at image understanding, image angle point/framework characteristic, texture
Levy and on the basis of rarefaction representation, study knowledge domain and data field Synergistic method, and image cognition and mathematics meter further
Calculate model；Key algorithm design is main on the basis of existing angle point and skeleton etection theory, optimizes curvature measuring function further,
Explore core algorithm optimization and the design problem of implementation of fine class irregular target area feature detection；The main base of application scenarios inspection
Build demonstration and verification system in the image of actual acquisition, it is achieved the test checking of core algorithm performance.
Above content is to combine concrete preferred embodiment further description made for the present invention, it is impossible to assert
Being embodied as of the present invention is confined to these explanations.For general technical staff of the technical field of the invention,
On the premise of without departing from present inventive concept, some simple deduction or replace can also be made, all should be considered as belonging to the present invention's
Protection domain.
Claims (7)
1. the object detection method based on Corner Feature, it is characterised in that said method comprising the steps of:
S1: gather the complicated overlay chart picture of fine class；
S2: carry out to complicated overlay chart picture pre-processing, Corner Detection；
S3: based on feature extraction and the sign of angle point and skeleton knowledge；
The structured features modeling of S4: complicated overlay chart picture；
S5: fiber target identification；Wherein,
Image semantic classification process in described step S2 includes: the binaryzation-> image filtering-> Image Reconstruction of image-> image increases
By force, described image filtering includes that morphological image is processed and edge-smoothing；
Described step S3 is specially the angle point sequence obtaining image and skeleton describes, and angle point includes crosspoint and other angle points, profit
By skeletal extraction method, the single individuality in multiple overlay target is labeled, but meeting between crosspoint and end points in skeletal extraction
Multiple line segment occur, general public line segment occurs in intersection region, needs to be deleted；But can go out when overlapping fibers is more
Existing " pseudo-public line segment ", now needs to be labeled by means of Intelligent Recognition algorithm；Each image is regarded as a set F,
The then corresponding subclass F of the edge contour curve of every fiber of overlapping fibers figurei, it comprises a series of similitude；
To overlapping fibre image collection F={F1,F2,…,FM, utilize clustering algorithm to classify contour images, find each pixel Pj
Affiliated fiber collection Fi, each individual fibers is identified；
Described step S4 is using the local feature of fine class image as primary image Expressive Features, and then with the chi between local feature
Degree correlation and spatial coherence are foundation, it is achieved the expression to picture structure.
2. method according to claim 1, it is characterised in that: in described step S2, utilize based on spatial domain convolution kernel masterplate
The complex background noise to image for the multi-channel filter bank carry out multiple dimensioned multi-direction suppression.
3. method according to claim 1, it is characterised in that: in described step S2, morphological image processes and uses based on light
The image outline probe algorithm of spot diffusion model realizes: searches for the hot spot of target individual, spread hot spot to edge contour, memory
Profile；The essence of hot spot diffusion is to be obtained the edge contour of fibre image by searching for closed loop territory, in order to avoid edge transition
Expanding, this algorithm method also often combines Sobel, Canny or Roberts operator to improve edge extracting effect.
4. method according to claim 1, it is characterised in that: in described step S2, morphological image processes and uses B-spline
Surface fitting obtains the background of fibre image, then pristine fibre image subtracting background is obtained target image, reaches removal figure
As background and the purpose solving uneven illumination；Carry out overall situation binaryzation again to target image, be filled with place to binary image
Reason, each object in mark binary image, the quantity of number of pixels according to contained by object carries out denoising, obtains target accurately
Fibre image.
5. method according to claim 1, it is characterised in that: in described step S2, the extraction of angle point is bent based on pixel
Rate, counts curvature and changes from small to big, or point from large to small is as candidate's flex point, sets up oval Support for candidate's flex point
Territory, utilizes supporting zone to judge this flex point.
6. method according to claim 1, it is characterised in that: described step S4 is particularly as follows: give local feature set F=
{fi,Wherein pi、vi、And siIt is respectively position, apparent description, principal direction and the yardstick of feature
Value, then stratification picture structure expression model is defined as follows,
M={{vi},{Rspatial},{Rscale}}。
This model is made up of three parts, apparent message part { vi, the space correlation relation { R between featurespatialAnd yardstick
Dependency relation { Rscale}；Then both can use defined level incidence relation based on space and scale factor.
7. the object detecting device based on Corner Feature, it is characterised in that: described device includes:
For gathering the module of the complicated overlay chart picture of fine class；
For carrying out to complicated overlay chart picture pre-processing, the module of Corner Detection；
For the module based on the feature extraction of angle point and skeleton knowledge and sign；
Module for the structured features modeling of complicated overlay chart picture；
Module for fiber target identification；Wherein,
Described Image semantic classification process includes: the binaryzation-> image filtering-> Image Reconstruction-> image enhaucament of image, described figure
As filtering includes that morphological image is processed and edge-smoothing；
Described angle point sequence and skeleton for obtaining image based on the feature extraction of angle point and skeleton knowledge and the module of sign
Describing, angle point includes crosspoint and other angle points, utilizes skeletal extraction method to enter rower to the single individuality in multiple overlay target
Note, but skeletal extraction there will be between crosspoint and end points multiple line segment, general public line segment occurs in intersection region, needs
It is deleted；But there will be " pseudo-public line segment " when overlapping fibers is more, now need to come by means of Intelligent Recognition algorithm
It is labeled；Each image is regarded as set F, then an edge contour curve correspondence one for every fiber of overlapping fibers figure
Individual subclass Fi, it comprises a series of similitude；To overlapping fibre image collection F={F1,F2,…,FM, utilize clustering algorithm
Contour images is classified, finds each pixel PjAffiliated fiber collection Fi, each individual fibers is identified；
The module of the described structured features modeling for complicated overlay chart picture is using the local feature of fine class image as parent map
As Expressive Features, and then with the scale correlations between local feature and spatial coherence as foundation, it is achieved to picture structure
Express.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610380344.9A CN106096613A (en) | 2016-05-31 | 2016-05-31 | Image multi-target detection method and device based on corner feature |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610380344.9A CN106096613A (en) | 2016-05-31 | 2016-05-31 | Image multi-target detection method and device based on corner feature |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106096613A true CN106096613A (en) | 2016-11-09 |
Family
ID=57230002
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610380344.9A Withdrawn CN106096613A (en) | 2016-05-31 | 2016-05-31 | Image multi-target detection method and device based on corner feature |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106096613A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107016394A (en) * | 2017-02-27 | 2017-08-04 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of decussating fibers characteristic point matching method |
| CN107133565A (en) * | 2017-03-31 | 2017-09-05 | 大连理工大学 | Laser incising molded line feature extracting method based on line laser |
| CN107168202A (en) * | 2017-05-25 | 2017-09-15 | 王俊芝 | Emergency control method |
| CN107256406A (en) * | 2017-04-19 | 2017-10-17 | 深圳清华大学研究院 | Overlapping fibers image partition method, device, storage medium and computer equipment |
| CN109063749A (en) * | 2018-07-17 | 2018-12-21 | 山东科技大学 | A kind of robust convolution nuclear volume adaptation method based on angle point radiation field |
| CN109191537A (en) * | 2018-07-19 | 2019-01-11 | 成都智慧海派科技有限公司 | A kind of method of vision positioning mobile device screen |
| CN111460960A (en) * | 2020-03-27 | 2020-07-28 | 重庆电政信息科技有限公司 | Motion classification and counting method |
| CN111461211A (en) * | 2020-03-31 | 2020-07-28 | 中国科学院计算技术研究所 | Feature extraction method for lightweight target detection and corresponding detection method |
| CN111665244A (en) * | 2019-03-05 | 2020-09-15 | 温力力 | Method based on textile fiber identification and component detection system |
| CN111665243A (en) * | 2019-03-05 | 2020-09-15 | 温力力 | Textile fiber identification and component detection system |
| CN111709889A (en) * | 2020-06-09 | 2020-09-25 | 中联云港数据科技股份有限公司 | Image dynamic real-time analysis system and method based on artificial intelligence |
| CN113158791A (en) * | 2021-03-15 | 2021-07-23 | 上海交通大学 | Human-centered image description labeling method, system, terminal and medium |
-
2016
- 2016-05-31 CN CN201610380344.9A patent/CN106096613A/en not_active Withdrawn
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107016394A (en) * | 2017-02-27 | 2017-08-04 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of decussating fibers characteristic point matching method |
| CN107016394B (en) * | 2017-02-27 | 2020-03-24 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Cross fiber feature point matching method |
| CN107133565B (en) * | 2017-03-31 | 2020-04-07 | 大连理工大学 | Line laser-based laser engraving line feature extraction method |
| CN107133565A (en) * | 2017-03-31 | 2017-09-05 | 大连理工大学 | Laser incising molded line feature extracting method based on line laser |
| CN107256406A (en) * | 2017-04-19 | 2017-10-17 | 深圳清华大学研究院 | Overlapping fibers image partition method, device, storage medium and computer equipment |
| CN107256406B (en) * | 2017-04-19 | 2020-05-01 | 深圳清华大学研究院 | Method and device for segmenting overlapped fiber image, storage medium and computer equipment |
| CN107168202A (en) * | 2017-05-25 | 2017-09-15 | 王俊芝 | Emergency control method |
| CN109063749A (en) * | 2018-07-17 | 2018-12-21 | 山东科技大学 | A kind of robust convolution nuclear volume adaptation method based on angle point radiation field |
| CN109063749B (en) * | 2018-07-17 | 2021-06-18 | 山东科技大学 | Robust convolution kernel number adaptation method based on angular point radiation domain |
| CN109191537A (en) * | 2018-07-19 | 2019-01-11 | 成都智慧海派科技有限公司 | A kind of method of vision positioning mobile device screen |
| CN111665244A (en) * | 2019-03-05 | 2020-09-15 | 温力力 | Method based on textile fiber identification and component detection system |
| CN111665243A (en) * | 2019-03-05 | 2020-09-15 | 温力力 | Textile fiber identification and component detection system |
| CN111665243B (en) * | 2019-03-05 | 2024-03-19 | 温力力 | Textile fiber identification and component detection system |
| CN111665244B (en) * | 2019-03-05 | 2024-03-19 | 温力力 | Method based on textile fiber identification and component detection system |
| CN111460960A (en) * | 2020-03-27 | 2020-07-28 | 重庆电政信息科技有限公司 | Motion classification and counting method |
| CN111461211A (en) * | 2020-03-31 | 2020-07-28 | 中国科学院计算技术研究所 | Feature extraction method for lightweight target detection and corresponding detection method |
| CN111709889A (en) * | 2020-06-09 | 2020-09-25 | 中联云港数据科技股份有限公司 | Image dynamic real-time analysis system and method based on artificial intelligence |
| CN111709889B (en) * | 2020-06-09 | 2023-04-07 | 中联云港数据科技股份有限公司 | Image dynamic real-time analysis system and method based on artificial intelligence |
| CN113158791A (en) * | 2021-03-15 | 2021-07-23 | 上海交通大学 | Human-centered image description labeling method, system, terminal and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106096613A (en) | Image multi-target detection method and device based on corner feature | |
| CN110428432B (en) | Deep neural network algorithm for automatically segmenting colon gland image | |
| CN113344849B (en) | Microemulsion head detection system based on YOLOv5 | |
| Zou et al. | CrackTree: Automatic crack detection from pavement images | |
| CN107506761B (en) | Brain image segmentation method and system based on significance learning convolutional neural network | |
| Wan et al. | Robust nuclei segmentation in histopathology using ASPPU-Net and boundary refinement | |
| Bibiloni et al. | A survey on curvilinear object segmentation in multiple applications | |
| CN107945153A (en) | A kind of road surface crack detection method based on deep learning | |
| CN108426994A (en) | Digital holographic microscopy data are analyzed for hematology application | |
| Cao et al. | Ancient mural restoration based on a modified generative adversarial network | |
| Ni et al. | Automatic detection and counting of circular shaped overlapped objects using circular hough transform and contour detection | |
| GB2478593A (en) | Segmentation of cell nuclei in histological sections | |
| Joshua et al. | Segmentation of optic cup and disc for diagnosis of glaucoma on retinal fundus images | |
| Delibasoglu et al. | Improved U-Nets with inception blocks for building detection | |
| Narayanaswamy et al. | Robust adaptive 3-D segmentation of vessel laminae from fluorescence confocal microscope images and parallel GPU implementation | |
| Wen et al. | Review of research on the instance segmentation of cell images | |
| WO2020119624A1 (en) | Class-sensitive edge detection method based on deep learning | |
| Zhang et al. | TUnet-LBF: Retinal fundus image fine segmentation model based on transformer Unet network and LBF | |
| Vijendran et al. | Optimal segmentation and fusion of multi-modal brain images using clustering based deep learning algorithm | |
| CN113096080A (en) | Image analysis method and system | |
| Zhang et al. | Reconnection of interrupted curvilinear structures via cortically inspired completion for ophthalmologic images | |
| Liu et al. | Dunhuang mural line drawing based on bi-dexined network and adaptive weight learning | |
| Sáez et al. | Neuromuscular disease classification system | |
| CN115661703A (en) | Method for extracting shop signboard information based on deep learning | |
| Nguyen et al. | Pavement crack detection and segmentation based on deep neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20161109 |