CN106056031A - Image segmentation algorithm - Google Patents
Image segmentation algorithm Download PDFInfo
- Publication number
- CN106056031A CN106056031A CN201610108090.5A CN201610108090A CN106056031A CN 106056031 A CN106056031 A CN 106056031A CN 201610108090 A CN201610108090 A CN 201610108090A CN 106056031 A CN106056031 A CN 106056031A
- Authority
- CN
- China
- Prior art keywords
- image
- image segmentation
- threshold
- segmentation
- evaluation function
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
- G06V30/41—Analysis of document content
- G06V30/413—Classification of content, e.g. text, photographs or tables
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
- G06V30/41—Analysis of document content
- G06V30/414—Extracting the geometrical structure, e.g. layout tree; Block segmentation, e.g. bounding boxes for graphics or text
Abstract
The invention relates to an image segmentation algorithm, and belongs to the technical field of electronic image segmentation. The image segmentation algorithm comprises the steps of (1) forming a one-dimensional histogram according to an original image; (2) determining best dual thresholds based on a Fisher evaluation function image segmentation method of the one-dimensional histogram. The beneficial effects are that the image segmentation algorithm provided by the invention has excellent image segmentation effects, and three types of segmentations for images can be realized reasonably. As for three types of grayscale images with frames, characters and backgrounds, the backgrounds, the frames and the characters can be well segmented out independently.
Description
Technical field
The present invention relates to a kind of image segmentation algorithm, belong to electronic image segmentation technology.
Background technology
Digital image processing is the important content of computer vision, accomplishes fluently solid foundation for further image recognition,
Mainly include the contents such as noise filtering, pixel interpolation, image enhaucament and image segmentation.Image segmentation is one of them crucial ring
Joint.The method having occurred in that thousands of kinds of image segmentations at present, does not has a kind of dividing method all to have very for all of image
Good segmentation effect.The quality of segmentation result directly influences the correctness of problem solving in high-rise vision.Especially for three
The class image i.e. image of band text border, background, existing image segmentation algorithm segmentation effect is poor, the picture effect dealt
Tend not to meet the demand of client.
Summary of the invention
In order to solve the problems referred to above, the present invention is directed to band frame, word and the gray level image of background three class, it is provided that Yi Zhongneng
Enough well by image segmentation algorithm independent to background, frame and word three class.
The technical scheme that the present invention provides is: a kind of image segmentation algorithm, the method comprises the steps:
Step one, is generated one dimensional image rectangular histogram by original image；
Step 2, determines optimal dual threshold based on one dimensional image histogrammic Fisher evaluation function image segmentation；
1. the w of three classifications it is provided with0、w1And w2A, B, C tri-part, with Fisher evaluation function J (Y) to three classifications
Between separating degree be quantitatively described, it is defined as:
(formula 2.1)
Wherein, m0, m1And m2It is respectively w0、w1And w2Classification meansigma methods,WithFor w0、w1And w2Class in side
Difference；
2. Lenna image histogram is carried out Double Thresholding Segmentation, select suitable two threshold values s1 and s2, image slices vegetarian refreshments
(m, gray value n) be f (m, n) ∈ [0, L-1], each several part probability of three classifications is:
(formula 2.2)
(formula 2.3)
(formula 2.4)
Its average u0(s), u1(s), u2S () is respectively as follows:
(formula 2.5)
(formula 2.6)
(formula 2.7)
Wherein PiRepresent the probability of gray scale, Pi=fi/ N, fiFor the pixel of gray value i, N is total pixel；
3. three class w0、w1And w2Variance within clusters be respectively as follows:
(formula 2.8)
(formula 2.9)
(formula 3.0)
According to Fisher evaluation function single threshold image partition method, when image carries out three classes segmentations, utilize one-dimensional directly
Side figure on projection meet each cluster between group variable sum two-by-two and intra-class variance and ratio reach very big, then comment based on Fisher
The interpretational criteria of valency function Dual-threshold image segmentation method is:
The threshold value corresponding when J (s1, s2) obtains maximumFor optimal threshold, three classes separated is best, image
Segmentation effect is optimal, and therefore using this Fisher evaluation function as the criterion of image Double Thresholding Segmentation, its threshold value is chosen as:
Step 3, on one dimensional histograms, does Double Thresholding Segmentation, and for pixel, (m n) has
Wherein, fs1,s2(m n) is segmentation result image, threshold valueFor optimal threshold, gray scale be f (m, n).
The beneficial effect that the present invention is reached:
What the present invention provided has good image segmentation, energy based on Fisher evaluation function Double Thresholding Segmentation method
Three classes of image are split by enough reasonably realization.To band frame, word and the gray level image of background three class, can well be the back of the body
Scape, frame and word three class individual segmentation are out.
Accompanying drawing explanation
Fig. 1 is the Two dimensional Distribution schematic diagram of the two class cluster situations of the present invention.
Fig. 2 is the Two dimensional Distribution schematic diagram of the three class cluster situations of the present invention.
Fig. 3 is the schematic diagram of the image histogram Double Thresholding Segmentation of the present invention.
Detailed description of the invention
The invention will be further described below in conjunction with the accompanying drawings.Following example are only used for clearly illustrating the present invention
Technical scheme, and can not limit the scope of the invention with this.
One, the Fisher in pattern recognition is theoretical
In pattern recognition theory, it is possible to use evaluation function carries out feature selection, it is assumed that N-dimensional feature X is in a straight line
Projection, should select a projection straight line, and the distance between formula class is maximum.As it is shown in figure 1, for two classifications w0And w1, it is assumed that all kinds of
Feature be Two dimensional Distribution, such as the A in Fig. 1, part B, they are projected on straight line Y1 and Y2, linear group of feature vector, X
Close y to be expressed as with (formula 1.1):
Y=YTX (formula 1.1)
When | | Y | | is when=1, then y is exactly X projection on Y-direction straight line.
In FIG, A represents that X belongs to w0Distribution；B represents that X belongs to w1Distribution；C representsX belongs to w1Distribution；D
RepresentX belongs to w0Distribution；E representsX belongs to w0Distribution；F representsX belongs to w1Distribution.
Belong to wiMeansigma methods u of feature vector, XiRepresent with (formula 1.2):
(formula 1.2)
Wherein, niRepresent and belong to wiThe number of class X, converts X with (formula 1.1), obtains meansigma methods m of yiAnd variance within clusters
(formula 1.3)
(formula 1.4)
In order to make w0And w1Effectively identifying, what the feature of two classes should separate opens better and better, it is therefore necessary to an evaluation
Function describes the separating degree between two classes, and Fisher evaluation function J (Y) faces and puts forward based on this thought, and it is permissible
Separating degree between two classifications is carried out quantitative description, and it is defined as:
(formula 1.5)
m0For w0Classification meansigma methods, m1For w1Classification meansigma methods, from (formula 1.5) it can be seen that when two classifications are put down
Average spacing is the biggest and all kinds of variance is the least, the meansigma methods spacing of two classifications and variance within clusters and ratio reach very big
Time, J (Y) obtains maximum, and two class separating effects are optimal.
Two, Dual-threshold image segmentation method based on Fisher evaluation function
Be divided into the needs of three classes for some image, the present invention proposes a kind of dual threashold based on Fisher evaluation function
Value image partition method.The half-tone information that this process employs image slices vegetarian refreshments projects, and image carries out three class segmentations.
Fig. 2 is three class cluster situation schematic diagrams, and in the drawings, A represents that X belongs to w0Distribution；B represents that X belongs to w1Distribution；C
Represent that X belongs to w2Distribution；E representsX belongs to w0Distribution；F representsX belongs to w1Distribution；G representsX belongs to
In w2Distribution；L representsX belongs to w2Distribution；M representsX belongs to w1Distribution；N representsX belongs to w0Point
Cloth.
W for three classifications0、w1And w2A, B, C tri-part, in order to make w0、w1And w2Effectively identify, three classes
What feature should separate opens better and better, during three classes segmentations when three classification meansigma methodss two-by-two distance the biggest and each variance within clusters of sum it
The least, the average distance sum of like members two-by-two of each segmentation classification and all kinds of variance within clusters and ratio segmentation effect when reaching very big
Most preferably, what three classes separated is best.Utilize Fisher evaluation function J (Y) that the separating degree between three classifications is quantitatively described, its
It is defined as:
(formula 2.1)
m0For w0Classification meansigma methods, m1For w1Classification meansigma methods, m2For w2Classification meansigma methods,For w0Class in side
Difference,For w1Variance within clusters,For w2Variance within clusters.When three classification meansigma methods the biggest and all kinds of differences of spacing sum and
The least, when between the class two-by-two of each segmentation classification, average distance sum reaches very big with all kinds of variance within clusters sum ratios, segmentation effect is
Good, J (Y) obtains maximum.
It is therefore desirable to utilize image histogram, choose two suitable threshold values, as it is shown on figure 3, image is carried out three classes
Image segmentation.(m, gray value n) is that (m, n) image is divided into three to f by ∈ [0, L-1], threshold value s1 and s2 to image slices vegetarian refreshments
Point, the probability of each several part:
(formula 2.2)
(formula 2.3)
(formula 2.4)
Its average u0(s), u1(s), u2S () is respectively as follows:
(formula 2.5)
(formula 2.6)
(formula 2.7)
Wherein PiRepresent the probability of gray scale, Pi=fi/ N, fiFor the pixel of gray value i, N is total pixel.
The variance within clusters of three classes is respectively as follows:
(formula 2.8)
(formula 2.9)
(formula 3.0)
According to Fisher evaluation function single threshold image partition method, when image carries out three class segmentations, it is considered to all kinds of
Prior probability, utilize the projection on one dimensional histograms meet each cluster between group variable sum two-by-two and intra-class variance and ratio reach
Greatly, then interpretational criteria based on Fisher evaluation function Dual-threshold image segmentation method is:(formula 3.1)
The threshold value corresponding when J (s1, s2) obtains maximumFor optimal threshold, three classes separated is best, image
Segmentation effect is optimal, and therefore using this Fisher evaluation function as the criterion of image Double Thresholding Segmentation, its threshold value is chosen as:
For each pixel (m, n) its gray scale be f (m, n), then Double Thresholding Segmentation is determined as:
(formula 3.3)
According to above-mentioned analysis, can state such as based on one dimensional histograms Fisher evaluation function Dual-threshold image segmentation method
Under:
(1) one dimensional histograms is formed by original image；
(2) Fisher evaluation function image segmentation based on one dimensional histograms determines optimal dual threshold；
(3) on one dimensional histograms, doing Double Thresholding Segmentation, for pixel, (m n) has
Wherein, fs1,s2(m n) is segmentation result image.
By Dual-threshold image segmentation method based on Fisher evaluation function, it is possible to well realize gray level image three
The segmentation of class.
The above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For Yuan, on the premise of without departing from the technology of the present invention principle, it is also possible to make some improvement and deformation, these improve and deformation
Also should be regarded as protection scope of the present invention.
Claims (6)
1. an image segmentation algorithm, is characterized in that, the method comprises the steps:
Step one, is generated one dimensional image rectangular histogram by three original class images；
Step 2, determines optimal dual threshold based on one dimensional image histogrammic Fisher evaluation function image segmentation；
Step 3, on one dimensional histograms, does Double Thresholding Segmentation, and for pixel, (m n) has
Wherein, fS1, s2(m n) is segmentation result image, threshold valueFor optimal threshold, gray scale be f (m, n).
A kind of image segmentation algorithm the most according to claim 1, is characterized in that: described step 2 specifically includes:
1. the w of three classifications it is provided with0、w1And w2A, B, C tri-part, with Fisher evaluation function J (Y) to three classifications between
Separating degree is quantitatively described, and it is defined as:
Wherein, m0, m1And m2It is respectively w0、w1And w2Classification meansigma methods,WithFor w0、w1And w2Variance within clusters；
2. one dimensional image rectangular histogram is carried out Double Thresholding Segmentation, select suitable two threshold values s1 and s2, and image slices vegetarian refreshments (m, n)
Gray value be that (m, n) ∈ [0, L-1], each several part probability of three classifications is w to f0(s), w1(s), w2S (), its average is u0(s),
u1(s), u2(s)；
3. three class w0、w1And w2Variance within clusters be respectively as follows:
Wherein PiRepresent the probability of gray scale, Pi=fi/ N, fiFor the pixel of gray value i, N is total pixel；
According to Fisher evaluation function single threshold image partition method, when image carries out three class segmentations, utilize one dimensional histograms
On projection meet each cluster between group variable sum two-by-two and intra-class variance and ratio reach very big, then based on Fisher evaluate letter
The interpretational criteria of number Dual-threshold image segmentation method is:
The threshold value corresponding when J (s1, s2) obtains maximumFor optimal threshold, three classes separated is best, and image is split
Best results, therefore using this Fisher evaluation function as the criterion of image Double Thresholding Segmentation.
A kind of image segmentation algorithm the most according to claim 2, is characterized in that: the threshold value choosing of described Fisher evaluation function
It is selected as:
Wherein s1 and s2 is threshold value.
A kind of image segmentation algorithm the most according to claim 1, is characterized in that: three described classifications w0、w1And w2Each portion
Point probability calculation formula is:
Wherein PiRepresenting the probability of gray scale, s1 and s2 is threshold value.
A kind of image segmentation algorithm the most according to claim 1, is characterized in that: its average u0(s), u1(s), u2(s)
Computing formula is respectively as follows:
Described w0、w1And w2It it is each several part probability of three classifications.
A kind of image segmentation algorithm the most according to claim 1, is characterized in that: described three class images are band word, background
Image with frame.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610108090.5A CN106056031A (en) | 2016-02-29 | 2016-02-29 | Image segmentation algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610108090.5A CN106056031A (en) | 2016-02-29 | 2016-02-29 | Image segmentation algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106056031A true CN106056031A (en) | 2016-10-26 |
Family
ID=57484339
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610108090.5A Pending CN106056031A (en) | 2016-02-29 | 2016-02-29 | Image segmentation algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106056031A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107966454A (en) * | 2017-12-25 | 2018-04-27 | 陕西科技大学 | A kind of end plug defect detecting device and detection method based on FPGA |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853494A (en) * | 2010-05-24 | 2010-10-06 | 淮阴工学院 | Color image segmentation method based on coring fuzzy Fisher criterion clustering |
| CN102542561A (en) * | 2011-11-23 | 2012-07-04 | 浙江工商大学 | Active contour synthetic aperture radar (SAR) image segmentation method based on Fisher distribution |
| US20140310314A1 (en) * | 2013-04-16 | 2014-10-16 | Samsung Electronics Co., Ltd. | Matching performance and compression efficiency with descriptor code segment collision probability optimization |
-
2016
- 2016-02-29 CN CN201610108090.5A patent/CN106056031A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853494A (en) * | 2010-05-24 | 2010-10-06 | 淮阴工学院 | Color image segmentation method based on coring fuzzy Fisher criterion clustering |
| CN102542561A (en) * | 2011-11-23 | 2012-07-04 | 浙江工商大学 | Active contour synthetic aperture radar (SAR) image segmentation method based on Fisher distribution |
| US20140310314A1 (en) * | 2013-04-16 | 2014-10-16 | Samsung Electronics Co., Ltd. | Matching performance and compression efficiency with descriptor code segment collision probability optimization |
Non-Patent Citations (1)
| Title |
|---|
| 龚淑柯: "基于Fisher评价函数的图像分割方法", 《中国优秀博硕士学位论文全文数据库 (硕士)信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107966454A (en) * | 2017-12-25 | 2018-04-27 | 陕西科技大学 | A kind of end plug defect detecting device and detection method based on FPGA |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110210413B (en) | Multidisciplinary test paper content detection and identification system and method based on deep learning | |
| CN104751142B (en) | A kind of natural scene Method for text detection based on stroke feature | |
| Yi et al. | Text detection in natural scene images by stroke gabor words | |
| WO2018145470A1 (en) | Image detection method and device | |
| CN107491730A (en) | A kind of laboratory test report recognition methods based on image procossing | |
| CN104408449B (en) | Intelligent mobile terminal scene literal processing method | |
| CN110866896B (en) | Image saliency target detection method based on k-means and level set super-pixel segmentation | |
| JP2003515230A (en) | Method and system for separating categorizable symbols of video stream | |
| CN101122953A (en) | Picture words segmentation method | |
| CN101122952A (en) | Picture words detecting method | |
| Jia et al. | Deep convolutional neural network based HEp-2 cell classification | |
| CN110889311A (en) | Financial electronic facsimile document identification system and method | |
| Gattal et al. | Automatic parameter tuning of k-means algorithm for document binarization | |
| Yogarajah et al. | A dynamic threshold approach for skin tone detection in colour images | |
| Shivakumara et al. | A new RGB based fusion for forged IMEI number detection in mobile images | |
| CN110334631B (en) | Sitting posture detection method based on face detection and binary operation | |
| CN109272522B (en) | A kind of image thinning dividing method based on local feature | |
| CN101802844B (en) | Applying a segmentation engine to different mappings of a digital image | |
| JP3923243B2 (en) | Character extraction method from color document image | |
| CN106056031A (en) | Image segmentation algorithm | |
| Liu et al. | Detection and segmentation text from natural scene images based on graph model | |
| CN106446904A (en) | Image recognition method based on global binarization | |
| Sreedevi et al. | Ngfica based digitization of historic inscription images | |
| CN116246298A (en) | Space occupation people counting method, terminal equipment and storage medium | |
| JP2012003358A (en) | Background determination device, method, and program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161026 |