CN110276764A - K-Means underwater picture background segment innovatory algorithm based on the estimation of K value - Google Patents
K-Means underwater picture background segment innovatory algorithm based on the estimation of K value Download PDFInfo
- Publication number
- CN110276764A CN110276764A CN201910457591.8A CN201910457591A CN110276764A CN 110276764 A CN110276764 A CN 110276764A CN 201910457591 A CN201910457591 A CN 201910457591A CN 110276764 A CN110276764 A CN 110276764A
- Authority
- CN
- China
- Prior art keywords
- image
- background
- value
- algorithm
- quantization
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Geometry (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the K-Means underwater picture background segment innovatory algorithms estimated based on K value, belong to underwater picture segmenting Background technology, including following processing step: S1, color adjustment；S2, gray-level quantization: (a) K-Means gray-level quantization；(b) the image gray levels quantization based on K-Means algorithm；(c) the image gray levels quantization based on K-Means algorithm is improved；The judgement of S3 background and segmentation: (a) background based on image background color character is adjudicated；(b) the background judgement based on image background frequency domain character；(c) the background judgement based on image background spatial feature；(d) background segment improves the underwater picture background segment operation based on K-Means algorithm, to improve the accuracy of underwater picture segmenting Background by the K value of estimation K-Means classification, i.e. classification number and center.
Description
Technical field
The present invention relates to underwater picture segmenting Background technologies, more specifically to the K- estimated based on K value
Means underwater picture background segment innovatory algorithm.
Background technique
When light transmits in water, absorption and scattering phenomenon occurs, causing luminous energy, attenuation degree is larger in water, causes water
The viewing distance of lower light imaging reduces, and underwater picture has non-uniform brightness, and details is fuzzy, and picture contrast is poor, and colour cast etc. is asked
Topic, this is also that underwater picture processing increases many difficulty, so need before carrying out the processing of underwater picture background segment to water
Lower image carries out enhancing processing, to improve the adverse effect that colour cast and non-uniform brightness handle background segment.
Background segment is a kind of special Digital Image Segmentation, that is, is extracted from unrelated background interested
Target, background segment can greatly reduce the redundant data in original image, especially in the visual people figure for possessing huge data volume
As in, removing redundant data is a kind of very important operation, and otherwise the image procossing in later period will be that a workload is extremely huge
Big task.
Existing segmenting Background includes the background segment based on histogram thresholding and the back based on edge detection
Scape segmentation.
Threshold Segmentation Algorithm is a kind of simple, the image dividing processing easily implemented, which, which passes through, sets a threshold value,
As the foundation of image segmentation, it is then minimum gray level that the pixel grayscale greater than threshold value is maximum gray scale on the contrary, thus
Several significant regions are divided the image into, the core of the algorithm is how to determine gray threshold, generally according to the ash of image
Histogram feature is spent to select, and Threshold Segmentation Algorithm has simple easily implementation, and calculation amount is small, and the fast feature of splitting speed is being schemed
In the case where gray level contrast height, the available preferable image segmentation of Threshold Segmentation Algorithm, with histogram thresholding
Segmentation is different, and the segmenting Background of edge detection is the dividing processing algorithm based on Image Edge-Detection, marginal information category
In the high frequency section of image, it is reflection image airspace structure information, such as contour of object shape, passes through edge detection algorithm
Object edge information can be obtained, target object region is depicted, computer is allow clearly to identify target object,
The image that edge is obvious and image smoothing noise region is less can be obtained very based on the segmenting Background of edge detection
Good effect, but under water in image background dividing processing, because of underwater picture non-uniform brightness, details is fuzzy, image pair
It is poorer than degree, the features such as colour cast.
Background segment based on histogram thresholding and the background segment based on edge detection cannot obtain ideal effect
Fruit, the segmenting Background based on histogram thresholding often only consider the statistical property of image gray levels and usually ignore figure
The useful informations such as the textural characteristics of the airspace structure of picture, especially image, and it is very sensitive to noise, often over histogram
The image of Threshold segmentation will increase salt-pepper noise, and especially in the case where low-light level low contrast, image segmentation is very
It is undesirable.
When edge blurry in image or existing more high-frequency noise, the partitioning algorithm based on edge detection is often not
Can obtain desired effect, for example, can mistake eliminates foreground image due to the discontinuity at fuzzy region edge, or because of image height
Frequency noise and cause to export image and pseudo- region occur, i.e., do not occur edge but in the place on physical presence boundary, and originally do not having
There is the region at edge to there is edge because of noise, because of edge blurry existing for underwater picture, contrast brightness is low, based on straight
The segmenting Background of square figure Threshold segmentation and the segmenting Background based on edge detection tend not to preferably be tied
Fruit.
Summary of the invention
1. technical problems to be solved
Aiming at the problems existing in the prior art, the purpose of the present invention is to provide the K-Means water estimated based on K value
Lower image background segment innovatory algorithm, it improves base by the K value of estimation K-Means classification, i.e. classification number and center
It is operated in the underwater picture background segment of K-Means algorithm, to improve the accuracy of underwater picture segmenting Background.
2. technical solution
To solve the above problems, the present invention adopts the following technical scheme that:
Fig. 1-3 is please referred to, based on the K-Means underwater picture background segment innovatory algorithm of K value estimation, K-Means algorithm
It is a kind of unsupervised Data Clustering Algorithm, K value is the major parameter image of K-Means algorithm, and image background segmentation is a kind of incites somebody to action
The image operation of image background removal, including following processing step:
S1, color adjustment；
S2, gray-level quantization:
(a) K-Means gray-level quantization；
(b) the image gray levels quantization based on K-Means algorithm；
(c) the image gray levels quantization based on K-Means algorithm is improved；
The judgement of S3 background and segmentation:
(a) the background judgement based on image background color character；
(b) the background judgement based on image background frequency domain character；
(c) the background judgement based on image background spatial feature；
(d) background segment.
Further, color adjusts, and using than histogram equalization algorithm, while Lab color correcting algorithms is used, by RGB
Color image is converted to the image of LAB color space expression, and to A, channel B image is normalized.
Further, K-Means gray-level quantization, K-Means method define K mass center by programmer first, i.e. institute's phase
After the number for the obtained aggregate of data hoped, then determining mass center initial position, the data in data acquisition system can be divided into distance most
In aggregate of data where close mass center, the mass center of the aggregate of data is then calculated in dividing obtained each aggregate of data for the first time,
Then the updated value as each mass center generally calculates mass center using averaging method, and repeated data is divided to update with mass center and be operated,
Until the centroid position of the aggregate of data divided is no longer changed, or variation, apart from when being less than some value, algorithm terminates, thus
Obtain each aggregate of data of K and each aggregate of data centroid position.
Further, the image gray levels quantization based on K-Means algorithm is schemed using less grey scale table up to a pair
As the memory overhead to reduce image.
Further, improving the image gray levels quantization based on K-Means algorithm includes K-Means algorithm in image ash
Spend target gray level estimation of the estimation of K value and K-Means algorithm in grade quantization in image gray levels quantization.
Further, the background judgement based on image background color character, by each gray-level pixels of image after quantization point
From, each grayscale mask is formed, color image adjusted is switched into HSV color space, the space HSV is to utilize coloration (H),
Saturation degree (S) and brightness (V) Lai Dingyi color base are adjudicated in the background of image background frequency domain character.
Further, the background judgement based on image background frequency domain character passes through each pixel ash in the analysis a certain region of image
The grey scale change degree for spending grade and its surrounding pixel gray level, judges whether the region is background.
Further, the background judgement based on image background spatial feature indicates each ash by defining averagely centrifugation degree
It spends grade mask pixels position and deviates central pixel point degree, thus judge whether the mask correspondence image region is background.
3. beneficial effect
Compared with the prior art, the present invention has the advantages that
The present invention describes a kind of new method to underwater picture background segment, and obtains more complete foreground picture
Picture is experimentally confirmed, and improving the underwater segmenting Background of K-Means can be effectively by colour cast, and contrast is low and edge mould
The background segment of the underwater picture of paste, and background segment of the traditional algorithm such as based on histogram thresholding be based on edge detection
Background segment because the characteristics of underwater picture, tend not to obtain satisfactory processing result, and what the present invention described changes
Into the available foreground image with the more consistent underwater picture of artificial treatment of algorithm.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is the functional block diagram at gray-level quantization of the present invention；
Fig. 3 is the functional block diagram of background judgement and segmentation portion of the invention；
Fig. 4 is K-Means algorithm implementation of the invention；
Fig. 5 is K-Means gray-level quantization result of the invention；
Fig. 6 is the histogram of input picture of the invention；
Fig. 7 is histogram mean value fuzzy graph of the invention；
Fig. 8 is estimation K value figure of the invention；
Centroid estimation figure Fig. 9 of the invention；
Image histogram after estimation histogram and quantization Figure 10 of the invention；
Figure 11 is the quantized image for converging on locally optimal solution of the invention；
Figure 12 is the resulting image quantization result of the method for the present invention；
Figure 13 is background mask figure of the invention；
Foreground mask figure Figure 14 of the invention.
Specific embodiment
Technical solution in the embodiment of the present invention that following will be combined with the drawings in the embodiments of the present invention carries out clear, complete
Site preparation description.Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments, is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its
His embodiment, shall fall within the protection scope of the present invention.
Embodiment:
Fig. 1-3 is please referred to, based on the K-Means underwater picture background segment innovatory algorithm of K value estimation, K-Means algorithm
It is a kind of unsupervised Data Clustering Algorithm, K value is the major parameter image of K-Means algorithm, and image background segmentation is a kind of incites somebody to action
The image operation of image background removal, including following processing step:
S1, color adjustment: absorption of the underwater picture because of water body to visible light scatters so that underwater picture is with non-homogeneous
The problems such as brightness, details are fuzzy, and picture contrast is poor, colour cast, this method improve image pair using than histogram equalization algorithm
Than degree and the interference of non-uniform brightness is reduced, while this method improves image color cast problem using Lab color correcting algorithms, it will
Rgb color image is converted to the image of LAB color space expression, and to A, channel B image is normalized, so as to improve
Image color cast problem.
S2, gray-level quantization:
(a) K-Means gray-level quantization: K-Means is a kind of clustering method, and the purpose of clustering algorithm is
Data acquisition system containing different elements is divided into different data subsets, while each data subset contains at least one member
Element, and each element in data set belongs to and only belongs to a subset, clustered obtained subset is also known as aggregate of data, K-
Means algorithm is to carry out clustering operation in known data acquisition system to gather without other training data set auxiliary
Class divides, so K-Means algorithm is a kind of unsupervised clustering algorithm, K-Means method is defined by programmer first
K mass center, i.e., the number of desired obtained aggregate of data, then determine mass center initial position after, the data meeting in data acquisition system
Be divided into the aggregate of data where nearest mass center, the metric form of distance have very much, as Euclidean distance, block away from
The methods of with a distance from, included angle cosine, the mass center of the aggregate of data is then calculated in dividing obtained each aggregate of data for the first time,
Then the updated value as each mass center generally calculates mass center using averaging method, and repeated data is divided to update with mass center and be operated,
Until the centroid position of the aggregate of data divided is no longer changed, or variation, apart from when being less than some value, algorithm terminates, thus
Each aggregate of data of K and each aggregate of data centroid position are obtained, the core of K-Means algorithm is to minimize cost function, K-Means
The cost function of algorithm is as follows:
Wherein dist (xij, mj) indicate to be data point xijWith cluster mass center mjDistance, that is, find point that one group of mass center is constituted
Class mode, make to get the mass centers of data and the group in each group away from and it is minimum, the basic procedure of K-Means algorithm is as follows:
(1) select K position as the initial position of mass center；
(2) by original data set close in each element be divided into nearest mass center where aggregate of data；
(3) mass center of each aggregate of data is recalculated
(4) variation degree for analyzing centroid position, if the centroid position of each aggregate of data no longer changes or variation range is less than
When some value, algorithm terminates, and otherwise, the basic procedure (2) of repeating algorithm and (3), Fig. 4 are one group of data from K mass center of setting
Initial position is to the final process for obtaining K clustering cluster, and it is always convergent to be set as 2, K-means algorithm for K value in figure, but the calculation
For method there is also problems, practical K-Means algorithm may not be able to obtain minimizing the optimal sub-clustering of objective function, because
K-means algorithm is very sensitive to the initial centroid position of K value set by user and K aggregate of data, this leads to K-means
Algorithm objective function may finally converge on locally optimal solution rather than globally optimal solution；
(b) the image gray levels quantization based on K-Means algorithm: image gray levels quantization is to reduce ash in gray scale picture
A process for spending number of stages, i.e., reduce the memory overhead of image using less grey scale table up to a sub-picture, and general
Gray-level quantization is compared, and can carry out non-uniform quantizing to original image gray level using K-Means algorithm, this makes in image
With the more image detail of less grey-preserving after quantization, because the brightness for containing only image in gray level image is special
Pixels tall is H so gray level image is one-dimensional characteristic by sign, and width is the gray level image I of WH*WIt is wide for 1 to be converted to height
Degree is the image I ' of HxW1*(H*W), I ' is ranked up to obtain I ", the K value given according to user, in I " in be randomly provided K each
Starting mass center is denoted as m01, m02..., m0K, define distance function:
Dist (b, a)=(b-a)2Formula (2)
To I " middle element is classified, if j ∈ [1, H*W], the then set that the distance of j to each mass center is constituted are as follows:
Dj={ Dist (I " (j), m02), Dist (I " (j), m01), Dist (I " (j), m0K) formula (3)
Ask T ∈ [1, K] that T is met:
Dist (I " (j), mT)=min (Dj) formula (4)
Then element I " (j) is T class, " has been divided into K subset thus by I and has been denoted as M01, M02..., M0K, according to son
Element is concentrated to update each centroid position, if the initial mass center of T class is m0T, and include NTA element, then gained after updating for the first time
Mass center are as follows:
K mass center is obtained after 1st update, is denoted as: m11, m12..., m1K, to the T mass center, there are a certain δ to make
Obtain Dist (m0T, M1T) < δ, then stop updating, otherwise continue, obtaining the 1st updated K subset by formula (4) is M11,
M12..., M1K, the 2nd updated K mass center: m is obtained by formula (5)21, m22..., m2KIf final updated
Number is n, then final K mass center is mn1, mn2... ..., mnK, the K subset of I " is Mn1, Mn2..., MnK, in each subset
Element value is replaced by corresponding center of mass values, and this effect is mapped to former gray level image IH*WIn, it finally obtains and is quantified as K
The gray level image I " ' of gray level expressing(H*W),
Fig. 5 is K=16, K=8, K=4 respectively, when K=2, the K-Means image gray levels quantization of random starting point
Although the selection of K value has subjectivity as a result, K-Means algorithm simplifies non-homogeneous gray-level quantization, lack certain
Objective basis, this can have an adverse effect to the result of quantization, and the selection of initial mass center will affect K-Means algorithm
Update times and convergence as a result, the K value of i.e. K-Means algorithm and the selection of initial mass center finally will affect gray level image
Storage size and algorithm operational efficiency, therefore it is necessary to which finding a kind of method reduces K value and initial mass center K-
Adverse effect of the Means algorithm in gray-level quantization problem；
(c) improve based on K-Means algorithm image gray levels quantization: for K-means algorithm gray level image ash
The determining and initial centroid position problem of K value in grade quantization is spent, the present invention proposes that one kind is worked as in image gray levels quantization application
One of improved method, to reduce selection of the user to defining K value and initial centroid position to the adverse effect of arithmetic result,
Improve the independence of algorithm；
(1) K value estimation of the K-Means algorithm in image gray levels quantization:
K value has been finally reflected the gray level number after image gray levels quantization, by comparing the histogram of quantization front and back image
Scheme it is not difficult to find that the histogram of K value and gray level image is in the presence of contacting, because image histogram reflection is each gray scale in image
The frequency that grade occurs, and measure grayscaled process really indicates to use using the higher gray level of frequency of use in original image
The lower gray level of frequency, it is possible to which the gray level number for needing to quantify using the histogram estimation of image before quantifying is estimated
The gray level number of meter is K value, counts each gray-level pixels number of former gray level image, histogram Hist is obtained, by histogram
Data progress mean value is fuzzy in Hist obtains Hist ', if i ∈ [0,255], blur radius l, then
Histogram Maximum number after fuzzy is K value, definition:
Diff (i)=Hist ' (i+1)-Hist ' (i) formula (7)
If s ∈ [0,255], a if it existssSo that:
Then asFor the middle maximum of Hist ', the maximum of Hist ' (i) is a1, a2... as, then K=s,
Following figure 6-8 is the histogram of input picture, carries out the histogram after mean value obscures to histogram and K value is estimated,
Blue spike number indicates quantization number, i.e. K is 4；
(2) target gray level estimation of the K-Means algorithm in image gray levels quantization:
Gray-level quantization is carried out using K-Means algorithm, final result is replaced in subset using the mean value of each subset
Element value to realize gray-level quantization, so, peak value section where can use former grayscale image Histogram Maximum is estimated just
Prothyl center value；
If t ∈ (0.255) b if it existstSo that:
Then btFor the middle minimum of Hist ', the minimum of Hist ' (i) is b1, b2... bt, then each peak value section be
[0, b1], [b1+ 1, b2] ..., [bt+ 1,255],
If p ∈ [1, K], then p-th of peak value section is [cp, dp], then p-th of centroid estimation value are as follows:
The centroid estimation value of K mass center is obtained by formula (10), is denoted as m '1, m '2..., m 'K, by centroid estimation value
Initial value as mass center initial value, as K-Means algorithm；
Fig. 9 is to carry out the estimation of K value using the fuzzy rear acquired results of input picture histogram mean value, and carry out centroid estimation
As a result, Figure 10 be estimation quantization after after image histogram and actual quantization image histogram；Figure 11 and Figure 12 are respectively
Gray-level quantization obtained by gray-level quantization result obtained by random mass center and the method for the present invention is as a result, by comparison it is found that the present invention
Method can converge on globally optimal solution rather than locally optimal solution, finally obtain ideal quantized result；
The judgement of S3 background and segmentation: underwater picture background has distinguishing feature, and background is navy blue, and region is more smooth,
In image border, it is based on these features, it can be determined that whether each gray-level pixels region after quantization is background image
Region；
(a) the background judgement based on image background color character: each gray-level pixels of image after quantization are separated, shape
At each grayscale mask Mask1, Mask2......Maskn, color image adjusted is switched into HSV color space, HSV is empty
Between be using coloration (H), saturation degree (S) and brightness (V) Lai Dingyi color image, wherein the value in the channel coloration (H) for [0,
360], color change corresponding to corresponding red light wavelength to blue light wavelength, standard blue 240, blue domain are [180,300], I
Using the chromatic value of pixel to determine whether being background pixel, we define the just too subordinating degree function of background pixel:
X is pixel color angle value, using the chroma channel image of color image under each mask and HSV color space carry out with
Operation obtains the coloration image I of each quantization gray level region1, I2......In, pixel in image obtained by each mask operation is asked
The background degree of membership mean value of pixel, the background degree of membership mean value of K images are as follows:
N is mask MaskkNumber of pixels, H and W are respectively picture altitude width, and it is 0.7 that the present invention, which takes and cuts value λ, if
Meet condition:
B (k) > λ formula (13)
Then MaskkMeet the color judgment condition of background image for region；
(b) the background judgement based on image background frequency domain character: image smoothing degree belongs to the frequency-domain attribute of image, and one
As in the case of, foreground object image is because its marginal information is abundant, and pixel grey scale variation is obvious, and background image is relatively smooth, ash
Degree variation is relatively gentle, the ash that the present invention passes through the analysis a certain region of image each pixel grayscale and its surrounding pixel gray level
Variation degree is spent, judges whether the region is background, average gray variance product (SMD2) function of the present invention is as image area
The Appreciation gist of domain smoothness utilizes each grayscale mask Mask1, Mask2......MasknAsh corresponding with original color image
Degree image carries out the gray level image I that each quantization gray level region is obtained with operation1, I2......In, obtained by each mask and operation
The average gray variance product of image are as follows:
N is mask MaskkNumber of pixels, H and W are respectively picture altitude width, if meeting condition, (σ of the present invention is taken
10):
SMD2 (k) < σ formula (15)
Then MaskkMeet the frequency judgment condition of background image for region；
(c) the background judgement based on image background spatial feature: location of pixels belongs to image airspace structure information, one
As in the case of, background image is always centered around foreground object surrounding, and is in image edge location, is based on this characteristic, this hair
It is bright by defining average centrifugation degree, indicate each grayscale mask location of pixels deviation central pixel point degree, thus judge that this is covered
Whether mould correspondence image region is background, and each grayscale mask is followed successively by Mask1, Mask2......Maskn, then K is respectively covered
Mould MaskkAverage centrifugation degree are as follows:
N is mask MaskkNumber of pixels, H and W are respectively picture altitude width, and (I, J) is picture centre pixel seat
Mark, if meeting condition (δ of the present invention takes 200):
S (k) > δ formula (17)
Then MaskkMeet the airspace judgment condition of background image for region,
(d) it background segment: if meeting the mask of condition formula (13), formula (15) and formula (17) simultaneously, is denoted as respectively
Maskk1, Maskk2..., Maskkn, then background mask Mask0Are as follows:
Acquire background mask Mask0Afterwards, foreground mask Mask ' can be obtained0Are as follows:
Mask′0=Iones-Mask0Formula (19)
IonesFor all 1's matrix, by each channel of color image and foreground mask carries out and operation, after obtaining background segment
Image, Figure 13-14 is the background mask obtained using formula (18) and foreground mask figure.
The foregoing is only a preferred embodiment of the present invention, but protection scope of the present invention be not limited to
This, anyone skilled in the art in the technical scope disclosed by the present invention, according to the technique and scheme of the present invention
And its improve design and be subject to equivalent substitution or change, it should be covered by the scope of protection of the present invention.
Claims (8)
1. K-Means algorithm is a kind of unsupervised number based on the K-Means underwater picture background segment innovatory algorithm of K value estimation
According to clustering algorithm, K value is the major parameter image of K-Means algorithm, and image background segmentation is a kind of to remove image background
Image operation, it is characterised in that: including following processing step:
S1, color adjustment；
S2, gray-level quantization:
(a) K-Means gray-level quantization；
(b) the image gray levels quantization based on K-Means algorithm；
(c) the image gray levels quantization based on K-Means algorithm is improved；
The judgement of S3 background and segmentation:
(a) the background judgement based on image background color character；
(b) the background judgement based on image background frequency domain character；
(c) the background judgement based on image background spatial feature；
(d) background segment.
2. the K-Means underwater picture background segment innovatory algorithm according to claim 1 based on the estimation of K value, feature
Be: color adjustment using than histogram equalization algorithm, while using Lab color correcting algorithms, rgb color image converted
For the image of LAB color space expression, to A, channel B image is normalized.
3. the K-Means underwater picture background segment innovatory algorithm according to claim 1 based on the estimation of K value, feature
Be: K-Means gray-level quantization, K-Means method define K mass center by programmer first, i.e., desired obtained number
According to the number of cluster, then behind determining mass center initial position, the data in data acquisition system can be divided into where nearest mass center
Aggregate of data in, the mass center of the aggregate of data is then calculated in dividing obtained each aggregate of data for the first time, then as each
The updated value of mass center generally calculates mass center using averaging method, and repeated data is divided to update with mass center and be operated, until the data divided
The centroid position of cluster is no longer changed, or variation distance be less than some value when, algorithm terminates, thus to obtain each aggregate of data of K with
And each aggregate of data centroid position.
4. the K-Means underwater picture background segment innovatory algorithm according to claim 1 based on the estimation of K value, feature
It is: the image gray levels quantization based on K-Means algorithm, using less grey scale table up to a sub-picture to reduce image
Memory overhead.
5. the K-Means underwater picture background segment innovatory algorithm according to claim 1 based on the estimation of K value, feature
Be: improving the image gray levels quantization based on K-Means algorithm includes K of the K-Means algorithm in image gray levels quantization
The target gray level estimation of value estimation and K-Means algorithm in image gray levels quantization.
6. the K-Means underwater picture background segment innovatory algorithm according to claim 1 based on the estimation of K value, feature
Be: each gray-level pixels of image after quantization are separated, form each gray scale by the background judgement based on image background color character
Color image adjusted is switched to HSV color space by grade mask, and HSV space is to utilize coloration (H), saturation degree (S) and brightness
(V) Lai Dingyi color base is adjudicated in the background of image background frequency domain character.
7. the K-Means underwater picture background segment innovatory algorithm according to claim 1 based on the estimation of K value, feature
It is: the background judgement based on image background frequency domain character, around each pixel grayscale in the analysis a certain region of image and its
The grey scale change degree of pixel grayscale judges whether the region is background.
8. the K-Means underwater picture background segment innovatory algorithm according to claim 1 based on the estimation of K value, feature
Be: the background judgement based on image background spatial feature indicates each grayscale mask pixel position by defining averagely centrifugation degree
It sets and deviates central pixel point degree, thus judge whether the mask correspondence image region is background.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910457591.8A CN110276764A (en) | 2019-05-29 | 2019-05-29 | K-Means underwater picture background segment innovatory algorithm based on the estimation of K value |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910457591.8A CN110276764A (en) | 2019-05-29 | 2019-05-29 | K-Means underwater picture background segment innovatory algorithm based on the estimation of K value |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110276764A true CN110276764A (en) | 2019-09-24 |
Family
ID=67960460
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910457591.8A Pending CN110276764A (en) | 2019-05-29 | 2019-05-29 | K-Means underwater picture background segment innovatory algorithm based on the estimation of K value |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110276764A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110910398A (en) * | 2019-10-28 | 2020-03-24 | 衢州学院 | Video complex scene region segmentation method and device based on decision layer fusion |
| CN113160238A (en) * | 2021-03-05 | 2021-07-23 | 南京信息工程大学 | Sea surface image segmentation method based on sea wave theory |
| CN116011403A (en) * | 2023-03-27 | 2023-04-25 | 莱芜职业技术学院 | Repeated data identification method for computer data storage |
| CN116385435A (en) * | 2023-06-02 | 2023-07-04 | 济宁市健达医疗器械科技有限公司 | Pharmaceutical capsule counting method based on image segmentation |
| CN116662588A (en) * | 2023-08-01 | 2023-08-29 | 山东省大数据中心 | Intelligent searching method and system for mass data |
| CN116703910A (en) * | 2023-08-07 | 2023-09-05 | 威海丰荟建筑工业科技有限公司 | Intelligent detection method for quality of concrete prefabricated bottom plate |
| CN117392165A (en) * | 2023-12-12 | 2024-01-12 | 南方医科大学南方医院 | Medical sample big data acquisition method based on artificial intelligence |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101488218A (en) * | 2008-12-19 | 2009-07-22 | 四川虹微技术有限公司 | Image self-adapting airspace homographic filtering method |
| CN103218832A (en) * | 2012-10-15 | 2013-07-24 | 上海大学 | Visual saliency algorithm based on overall color contrast ratio and space distribution in image |
-
2019
- 2019-05-29 CN CN201910457591.8A patent/CN110276764A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101488218A (en) * | 2008-12-19 | 2009-07-22 | 四川虹微技术有限公司 | Image self-adapting airspace homographic filtering method |
| CN103218832A (en) * | 2012-10-15 | 2013-07-24 | 上海大学 | Visual saliency algorithm based on overall color contrast ratio and space distribution in image |
Non-Patent Citations (1)
| Title |
|---|
| LOU MARVIN CARAIG: "Color quantization using k-means", 《HTTPS://SE7ENTYSE7EN.DEV/POSTS/COLOR-QUANTIZATION-USING-K-MEANS》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110910398B (en) * | 2019-10-28 | 2021-07-20 | 衢州学院 | Video complex scene region segmentation method and device based on decision layer fusion |
| CN110910398A (en) * | 2019-10-28 | 2020-03-24 | 衢州学院 | Video complex scene region segmentation method and device based on decision layer fusion |
| CN113160238B (en) * | 2021-03-05 | 2023-06-20 | 南京信息工程大学 | Sea surface image segmentation method based on sea wave theory |
| CN113160238A (en) * | 2021-03-05 | 2021-07-23 | 南京信息工程大学 | Sea surface image segmentation method based on sea wave theory |
| CN116011403B (en) * | 2023-03-27 | 2023-10-03 | 莱芜职业技术学院 | Repeated data identification method for computer data storage |
| CN116011403A (en) * | 2023-03-27 | 2023-04-25 | 莱芜职业技术学院 | Repeated data identification method for computer data storage |
| CN116385435A (en) * | 2023-06-02 | 2023-07-04 | 济宁市健达医疗器械科技有限公司 | Pharmaceutical capsule counting method based on image segmentation |
| CN116385435B (en) * | 2023-06-02 | 2023-09-26 | 济宁市健达医疗器械科技有限公司 | Pharmaceutical capsule counting method based on image segmentation |
| CN116662588A (en) * | 2023-08-01 | 2023-08-29 | 山东省大数据中心 | Intelligent searching method and system for mass data |
| CN116662588B (en) * | 2023-08-01 | 2023-10-10 | 山东省大数据中心 | Intelligent searching method and system for mass data |
| CN116703910A (en) * | 2023-08-07 | 2023-09-05 | 威海丰荟建筑工业科技有限公司 | Intelligent detection method for quality of concrete prefabricated bottom plate |
| CN116703910B (en) * | 2023-08-07 | 2023-10-17 | 威海丰荟建筑工业科技有限公司 | Intelligent detection method for quality of concrete prefabricated bottom plate |
| CN117392165A (en) * | 2023-12-12 | 2024-01-12 | 南方医科大学南方医院 | Medical sample big data acquisition method based on artificial intelligence |
| CN117392165B (en) * | 2023-12-12 | 2024-02-23 | 南方医科大学南方医院 | Medical sample big data acquisition method based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110276764A (en) | K-Means underwater picture background segment innovatory algorithm based on the estimation of K value | |
| Tan et al. | Exposure based multi-histogram equalization contrast enhancement for non-uniform illumination images | |
| Jumb et al. | Color image segmentation using K-means clustering and Otsu’s adaptive thresholding | |
| Poletti et al. | A review of thresholding strategies applied to human chromosome segmentation | |
| CN111738064B (en) | Haze concentration identification method for haze image | |
| CN110443778B (en) | Method for detecting irregular defects of industrial products | |
| CN108510499A (en) | A kind of carrying out image threshold segmentation method and device based on fuzzy set and Otsu | |
| CN109166095A (en) | A kind of ophthalmoscopic image cup disk dividing method based on generation confrontation mechanism | |
| WO2007055359A1 (en) | Clustering system and image processing system having same | |
| CN102496023A (en) | Region of interest extraction method of pixel level | |
| CN105513053A (en) | Background modeling method for video analysis | |
| CN109448019B (en) | Adaptive method for smoothing parameters of variable-split optical flow model | |
| CN109377464A (en) | A kind of Double plateaus histogram equalization method and its application system of infrared image | |
| CN108230340A (en) | A kind of SLIC super-pixel extraction Weighting and super-pixel extracting method based on MMTD | |
| CN111199245A (en) | Rape pest identification method | |
| CN110473224B (en) | Automatic RSF level set image segmentation method based on KL entropy | |
| CN111815563A (en) | Retina optic disk segmentation method combining U-Net and region growing PCNN | |
| CN108805826B (en) | Method for improving defogging effect | |
| CN108765316B (en) | Mist concentration self-adaptive judgment method | |
| CN116843581B (en) | Image enhancement method, system, device and storage medium for multi-scene graph | |
| CN116563296B (en) | Identification method for abdomen CT image | |
| Jassim | Hybrid image segmentation using discerner cluster in FCM and histogram thresholding | |
| CN103886574A (en) | Image segmentation device | |
| CN108154188A (en) | Complex Background work Text Extraction based on FCM | |
| Li et al. | Image object detection algorithm based on improved Gaussian mixture model |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190924 |
|
| WD01 | Invention patent application deemed withdrawn after publication |