CN106874915B - A kind of feature extracting method with target area than invariance - Google Patents
A kind of feature extracting method with target area than invariance Download PDFInfo
- Publication number
- CN106874915B CN106874915B CN201710082876.9A CN201710082876A CN106874915B CN 106874915 B CN106874915 B CN 106874915B CN 201710082876 A CN201710082876 A CN 201710082876A CN 106874915 B CN106874915 B CN 106874915B
- Authority
- CN
- China
- Prior art keywords
- area
- target
- target area
- feature
- extracting method
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
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
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of feature extracting method with target area than invariance.The technical scheme comprises the following steps: being split first with two-dimentional Otsu dividing method to target slice, obtains target area and background area；Then the second-order moment around mean for calculating separately target area and background area eliminates the influence of area ratio；The second-order moment around mean of final integration objective region and background area obtains feature, i.e., by second-order moment around mean multiplied by respective normalization coefficient, then seeks its root sum square, obtain the feature for having target area than invariance.Feature extracting method proposed by the present invention has to target area by the influence of removal target area and background area area than constant property, can more accurately reflect that the gray scale of target slice is floated.Further it is proposed that feature extracting method do not need that additional parameter is arranged, it is succinct easy.
Description
Technical field
The invention belongs to image feature extraction techniques fields, are related to a kind of feature extraction with target area than invariance
The precision of feature extraction can be improved in method.
Background technique
Target classification is one of key areas of image steganalysis, is widely used.And feature extraction is made
For the core content of target classification, have a great impact to classifier design and its performance.Existing feature extracting method obtains
Feature be usually to the target area ratio in slice it is sensitive, in other words do not have invariance.With the standard being widely used
For difference (standard deviation characteristic can refer to document D.E.Kreithen, S.D.Halversen and G.J.Owirka,
“Discriminating targets from clutter,”Lincoln Lab.J.,vol.6,no.1,pp.25–52,
1993.), it describes the gray scale floating degree of pixel in a slice, but this feature can not accurately describe target
The gray scale floating degree of slice.By taking two groups of fixed dimension slices from Same Scene as an example, all contain two in every group of slice
Class sample, such as false-alarm and Ship Target.In first group of slice, target area and the false-alarm class sample of Ship Target class sample
False-alarm area approximation is bigger than false-alarm.In this case, the sample slice standard deviation of false-alarm class will be generally less than target class sample
This slice.In second group of slice, target area is relatively small in Ship Target class slice, and the area in false-alarm region is very big.This
The standard deviation that false-alarm is sliced in the case of kind is possible to be greater than the standard deviation of Ship Target slice.This is because opposite in target area
When smaller, target area is very little for the standard deviation contribution being entirely sliced, although and false-alarm gray scale is generally lower than naval vessel mesh
Gray scale is marked, but since false-alarm area is larger, so being also possible to obtain higher standard deviation.Therefore, standard deviation characteristic is to mesh
Target area ratio is very sensitive in mark slice, can not effectively indicate the gray scale floating degree of target slice.Based on this consideration, this hair
Bright to propose a kind of feature extracting method unrelated with area ratio, it can not be influenced by target area ratio, accurate description target
The gray scale floating degree of slice improves the discrimination of variety classes target slice.
Summary of the invention
The present invention compares the influence of feature, the feature extracted by eliminating target area in target slice.This feature
Extracting method can more precisely reflect that the gray scale of target slice is floated, and effectively improve the precision of feature extraction, improve different
The discrimination of type target slice.
The technical scheme is that
Target slice is split first with two-dimentional Otsu dividing method, obtains target area and background area；So
The second-order moment around mean for calculating separately target area and background area afterwards eliminates the influence of area ratio；Final integration objective region and
The second-order moment around mean of background area obtains feature, i.e., by second-order moment around mean multiplied by respective normalization coefficient, then seeks the flat of its sum
Root obtains the feature for having target area than invariance.
The beneficial effects of the present invention are:
1. feature extracting method proposed by the present invention by removal target area and background area area influence, have pair
Target area can more accurately reflect that the gray scale of target slice is floated than constant property.
2. feature extracting method proposed by the present invention with target area due to that, than constant property, can improve not
The discrimination of class target slice of the same race helps to improve the precision of target slice classification.
3. using feature extracting method proposed by the present invention do not need that additional parameter is arranged, it is succinct easy.
Detailed description of the invention
Fig. 1 is experimental data of the present invention；
Fig. 2 is flow chart of the present invention；
Fig. 3 is experimental result comparison diagram of the present invention.
Specific embodiment
Fig. 1 is experimental data of the present invention, and for 18 TerraSAR-X slice under Same Scene, size is
250 × 250, the wherein bigger slice 6 (big Ship Target, Fig. 1 (a)) of target area, the smaller slice 6 of target area
A (small Ship Target, Fig. 1 (b)), false-alarm targets are sliced 6 (Fig. 1 (c)).
Fig. 2 is flow chart of the present invention, and specific implementation step is as follows:
The first step is split sectioning image using two-dimentional Otsu dividing method, obtains target area and background area.
Two-dimentional Otsu dividing method can be with bibliography J.Liu, W.Li, and Y.Tian, " Automatic
thresholding of gray-level pictures using two-dimensional Otsu method,”China
1991 Int.Conf.on Circuits and Syst.,Shenzhen,China,Jun.1991,pp.325-327.
After target slice is split, a part is target area (Ship Target or false-alarm), and pixel is denoted as xi, as
Plain number is denoted as n1；Another part is background area, and pixel is denoted as xj, number of pixels is denoted as n-n1(wherein n is target slice
Sum of all pixels).
Second step calculates separately the second-order moment around mean of target area and background area, eliminates the influence of area ratio.
The second-order moment around mean of target area and background area is respectively Wherein μ0It is defined as modified mean value, as follows,
Wherein λ is an adjustment target area and the coefficient that influences on feature of background area area ratio, can be taken between (0,1)
Any value, be generally taken as 0.5.
Third step calculates feature proposed by the present invention using the second-order moment around mean of target area and background area, i.e., by two
Rank central moment is multiplied by respective normalization coefficientWithIts root sum square σ is sought again0, σ0It is as of the invention
The feature of proposition, as follows,
By deriving, feature σ can be proved0With target area than unrelated.
Simultaneously in statistical chart 1 standard deviation of 18 target slices and feature extracting method proposed by the present invention as a result, obtaining
As a result such as table 1.
1 standard deviation of table and feature of present invention extracting method contrast table
As can be seen from the table in standard deviation, the standard deviation of small Ship Target and false-alarm all 30 or so, therefore, it is difficult to
Distinguish small Ship Target and false-alarm.And it is calculated using feature extracting method proposed by the present invention, either big Ship Target
Or the result of small Ship Target all concentrates on 100 or so, and the result of false-alarm remains in the value of one smaller level,
Therefore Ship Target and false-alarm can be easily discriminated.
Fig. 3 is experimental result comparison diagram of the present invention, and the abscissa of Fig. 3 (a) and Fig. 3 (b) is the number of test sample, figure
3 (a) ordinates are standard deviation, and Fig. 3 (b) ordinate is that feature extracting method proposed by the present invention obtains as a result, wherein deep
Color triangle solid line represents the value of big Ship Target, and light coloured triangle dotted line represents the value of small Ship Target, dark diamond shape dotted line
Represent the value of false-alarm.As can be seen that the image bigger for target area, standard deviation can accurately reflect mesh from Fig. 3 (a)
The difference of mark slice and false-alarm slice, and can effectively be separated, and the image smaller for target area, it is calculated
The standard deviation come is easy for obscuring with false-alarm targets.From Fig. 3 (b) as can be seen that feature of present invention extracting method result can be quasi-
Really target and false-alarm are separated, have it is good index, help to improve target slice classification precision.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention (such as utilize feature of present invention
Extracting method improves other feature), these modifications and embellishments should also be considered as the scope of protection of the present invention.
Claims (1)
1. a kind of feature extracting method with target area than invariance, it is characterised in that include the following steps: first with
Two-dimentional Otsu dividing method is split target slice, obtains target area and background area；Then target area is calculated separately
The second-order moment around mean in domain and background area；Then, by the second-order moment around mean of target area and background area multiplied by respective normalizing
Change coefficient, then seek the root sum square of above-mentioned two calculated result, obtains the feature that there is target area than invariance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710082876.9A CN106874915B (en) | 2017-02-16 | 2017-02-16 | A kind of feature extracting method with target area than invariance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710082876.9A CN106874915B (en) | 2017-02-16 | 2017-02-16 | A kind of feature extracting method with target area than invariance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106874915A CN106874915A (en) | 2017-06-20 |
| CN106874915B true CN106874915B (en) | 2019-09-20 |
Family
ID=59166285
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710082876.9A Active CN106874915B (en) | 2017-02-16 | 2017-02-16 | A kind of feature extracting method with target area than invariance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106874915B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184538A (en) * | 2011-04-28 | 2011-09-14 | 北京航空航天大学 | Dynamic contour based automatic synthetic aperture radar (SAR) image segmentation method |
| CN103577826A (en) * | 2012-07-25 | 2014-02-12 | 中国科学院声学研究所 | Target characteristic extraction method, identification method, extraction device and identification system for synthetic aperture sonar image |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8538071B2 (en) * | 2009-03-18 | 2013-09-17 | Raytheon Company | System and method for target separation of closely spaced targets in automatic target recognition |
-
2017
- 2017-02-16 CN CN201710082876.9A patent/CN106874915B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184538A (en) * | 2011-04-28 | 2011-09-14 | 北京航空航天大学 | Dynamic contour based automatic synthetic aperture radar (SAR) image segmentation method |
| CN103577826A (en) * | 2012-07-25 | 2014-02-12 | 中国科学院声学研究所 | Target characteristic extraction method, identification method, extraction device and identification system for synthetic aperture sonar image |
Non-Patent Citations (2)
| Title |
|---|
| 《Local Isotropy Indicator for SAR Image Filtering: Application to Envisat/SAR Images of the Donana Wetland》;Belen Marti-Cardona 等;《IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing》;20150430;第8卷(第4期);全文 * |
| 《SAR图像舰船目标识别综述》;陈文婷 等;《现代雷达》;20121130;第34卷(第11期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106874915A (en) | 2017-06-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | SVM-based sea ice classification using textural features and concentration from RADARSAT-2 dual-pol ScanSAR data | |
| CN104217215B (en) | A kind of classifying identification method of water surface foggy image and picture rich in detail | |
| Solberg et al. | A large-scale evaluation of features for automatic detection of oil spills in ERS SAR images | |
| Topouzelis et al. | Detection and discrimination between oil spills and look-alike phenomena through neural networks | |
| EP2889619A1 (en) | Image processing device, program, image processing method, computer-readable medium, and image processing system | |
| Lu et al. | Improving pixel-based change detection accuracy using an object-based approach in multitemporal SAR flood images | |
| Singha et al. | Detection and classification of oil spill and look-alike spots from SAR imagery using an artificial neural network | |
| Mandal et al. | Segmentation of blood smear images using normalized cuts for detection of malarial parasites | |
| CN107103608B (en) | A kind of conspicuousness detection method based on region candidate samples selection | |
| CN103488992B (en) | A kind of oil spilling detection method towards complicated SAR image scene | |
| CN103400156A (en) | CFAR (Constant False Alarm Rate) and sparse representation-based high-resolution SAR (Synthetic Aperture Radar) image ship detection method | |
| Mohammed et al. | Detection and classification of malaria in thin blood slide images | |
| CN106127198A (en) | A kind of image character recognition method based on Multi-classifers integrated | |
| CN105608457B (en) | Gray Histogram square thresholding method | |
| CN105512622B (en) | A kind of visible remote sensing image sea land dividing method based on figure segmentation and supervised learning | |
| D’Onofrio et al. | Surface photometry of WINGS galaxies with GASPHOT | |
| CN104732534B (en) | Well-marked target takes method and system in a kind of image | |
| CN105718924A (en) | High-score satellite image cloud detection method based on multi-feature integration and machine learning | |
| CN111368865B (en) | Remote sensing image oil storage tank detection method and device, readable storage medium and equipment | |
| Sun et al. | Ship detection in SAR images based on multilevel superpixel segmentation and fuzzy fusion | |
| CN106874915B (en) | A kind of feature extracting method with target area than invariance | |
| Liu et al. | Infrared image segmentation method based on 2D histogram shape modification and optimal objective function | |
| Sharma et al. | A new method for word segmentation from arbitrarily-oriented video text lines | |
| CN113505712A (en) | Novel loss function based sea surface oil spill detection method of convolutional neural network | |
| CN111291747B (en) | Depth map-based color small target detection method in complex scene |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |