CN103198496B - A kind of method of abstracted image being carried out to vector quantization - Google Patents
A kind of method of abstracted image being carried out to vector quantization Download PDFInfo
- Publication number
- CN103198496B CN103198496B CN201310109647.3A CN201310109647A CN103198496B CN 103198496 B CN103198496 B CN 103198496B CN 201310109647 A CN201310109647 A CN 201310109647A CN 103198496 B CN103198496 B CN 103198496B
- Authority
- CN
- China
- Prior art keywords
- image
- abstracted
- carried out
- pixel
- square
- 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.)
- Expired - Fee Related
Links
Abstract
The invention discloses a kind of method of abstracted image being carried out to vector quantization, comprise following implementation step: 1) utilize feature to keep abstraction method to carry out abstract process to pending image and obtain abstracted image; 2) Boundary Extraction is carried out to abstracted image and obtain image boundary; 3) utilize image boundary to sample on described abstracted image, obtain the colouring information matched with each summit of square, namely obtain vectorial images data; 4) utilize Poisson image rebuilding method that described vectorial images data are carried out drafting and obtain grating images.The present invention is by carrying out the abstract process in feature based flow field to image, extract the border of image again, then border is coordinated to carry out color samples, finally utilize Poisson image reconstruction drawing image, the image change that recovers can be made continuous, visual effect is smooth, has similarity with abstracted image.
Description
Technical field
The present invention relates to the image vector field of computer graphics, image procossing, particularly relate to a kind of method of abstracted image being carried out to vector quantization.
Background technology
In a computer, the description of image generally can be divided into two large class---raster image and vector images.Raster image is made up of the point being called pixel (pictorial element), and these are by different built-up pattern composing images.Polar plot describes figure by geometric element and certain method for drafting, and geometric element comprises point, line, matrix, polygon, curve etc.Image vector is a major issue in Digital Image Processing, is the intersection problem of the every subjects such as all round computer vision, Computer Image Processing, computer graphics and an artificial intelligence.Vector image is owing to being describe based on geometric figure, and represent compact, storage organization is more flexible; In addition, vector image is edited, such as, some object in image is rotated, convergent-divergent time only need operate corresponding geometric graphic element, relatively convenient; Moreover vector image is at convergent-divergent, and especially amplification aspect shows very large advantage---information is undistorted, occurs the problem of obscurity boundary unlike raster image amplifies, and namely vector image is that resolution has nothing to do.
Along with the progressively maturation of Chinese entertainment industry and the day by day universal of the Internet, the show business consumption market of China is just at development, and for fields such as cartoon wherein, computer art creation, two-dimensional game, advertising creatives, abstracted image is with its more attractive visual information expression way, easilier leave deep impression to people, have a wide range of applications at these field abstracted image.In addition on the one hand, the terminal device of various outfit different resolution needs to transmit image, process, show, and vectorial images is the effective ways addressed this problem, and the vectorization method for abstracted image has the meaning of this one side.
Abstract (Abstraction) is a kind of method of non-realistic processing, it refers to reduce a concept or the information content of a phenomenon by the process of its vague generalization (Generalization), mainly in order to only preserve customizing messages in a certain respect.Such as, change into a ball by abstract for the football of a leather, only retain the information such as attribute and behavior of general ball.Abstract is mainly in order to make the complexity of handling object reduce.In image processing field, due to image complicacy and be unfavorable for simplifying and effectively expressing of data, so abstract is exactly will seek image more to simplify the expression more outstanding with effective information.The research field of image abstraction is very widely, comprising: the abstract (Shape-SimplifyingAbstraction) etc. simplified based on streamline sense abstract (Flow-Based Abstraction), Shape-based interpolation.
From the sixties in last century, Chinese scholars is just constantly had to propose method for detecting image edge, Laplace edge detection operator and Sobel edge detection operator are all adjacent the change intensity of (four neighborhoods) pixel to judge that whether this point is for marginal position according to image pixel, Sobel operator is then generalized on eight directions by Robinson edge detection operator, Roberts edge detection operator then estimates gradient according to the difference in mutually perpendicular direction, judges whether this position is marginal position.These methods all locate marginal position according to the feature of pixel change, facilitate easy-to-use, but also comparatively responsive to noise, locate also accurate not.
Within 2006, Lecot and Levy develops ARDECO, this system employs triangle gridding to express image, image is carried out trigonometric ratio according to unique point, and utilize Mumford-Shah energy minimization method to carry out Automatic Optimal drawing result, this system can also carry out mesh refinement for the position of specifying to this local, and support SVG form (Scalable Vector Graphics, a kind of a kind of standard format of internet transmission polar plot), but because SVG only supports linear color transition, this method makes the color transition not nature between triangle gridding, another weak point of this method is that the density of triangle gridding is excessive, the triangle gridding of certain density is still needed at the smooth region of image, cause data redundancy relatively serious.
Summary of the invention
For solving the problem such as abstracted image transmission, process, display on the terminal device of different resolution, the invention provides a kind of method of abstracted image being carried out to vector quantization, effectively can avoid the discontinuous problem of color transition, the present invention is also better than grid relatively when expressing irregularly shaped object.
Abstracted image is carried out to a method for vector quantization, comprises following implementation step:
1) utilize feature to keep abstraction method to carry out abstract process to pending image and obtain abstracted image;
2) Boundary Extraction is carried out to abstracted image and obtain image boundary;
A) abstracted image is transformed into Lab color space, for L component, color data scope is divided into n region, and give each region and number accordingly, each pixel number of abstracted image is converted into the numbering in region corresponding to this pixel native color numerical value;
B) for described L component, all four pixels of mutually adjoining are configured to square, utilize division line segment to split each square, make in square, to number identical summit and be in the same area, and the end points of described division line segment is in the mid point of square side;
C) to any two squares that are connected, will the division line segment of intersection point be had to be interconnected to form cut-off rule, the outward flange of this cut-off rule and abstracted image forms described image boundary jointly;
3) utilize image boundary to sample on described abstracted image, obtain the colouring information matched with each summit of square, namely obtain vectorial images data;
4) utilize Poisson image rebuilding method that described vectorial images data are carried out drafting and obtain grating images.
Feature keep abstraction method, the people such as Jan Eric Kyprianidis disclosed in 2008 a kind of feature keep image abstraction method (Jan Eric Kyprianidis,
imageAbstraction by Structure Adaptive Filtering.In Proceedings of EG UK Theoryand Practice of Computer Graphics, European Association for ComputerGraphics, 2008:51-58.).
Feature in described step 1) keeps abstracting process to comprise the flow field using structure tensor computed image, then carries out line integral convolution operation according to flow field to image.
Described structure tensor expression formula is:
Wherein, I
xfor each pixel gradient in the horizontal direction, I
yfor each pixel is in the gradient of vertical direction, the smaller of the eigenwert of this matrix is the characteristic direction of pixel.
The expression formula of described line integral convolution operation is:
In formula, i
0for the two-dimensional coordinate of current pixel, the position of C (i) pointed by the characteristic direction of current pixel, w (i) is weight information, z represents the surrounding pixel coordinate set contiguous with current pixel, i represents the element of z on the characteristic direction of current pixel, I is pixel value, the pixel value that I (C (i)) is C (i) position.
Described weight information w (i
0, expression formula i) is:
w(x,y)＝G(||y-x||,σ
s)G(|I(y)-I(x)|,σ
r)
Wherein, the i described in x, y correspondence
0, i, σ
sand σ
rrepresent the parameter of Gaussian function, I (y) and I (x) represents the pixel value of x and y position respectively.
In step 2) in, described n is the natural number of 2 ~ 256, and each zone number is from 0 to (n-1), and the span in each region is 256/n.Pixel number is 0-255, and meaningless when n is 1, and during n=2, image becomes bianry image, and 256 is the upper limit.Preferably n is 16 ~ 64 further.
As preferably, also comprise and smooth treatment is carried out to cut-off rule, then obtain described image boundary.The advantage of smooth treatment is that the closed boundary entirety making image is more smooth, and the border of general object is all smooth, or even continuous curve shape, can be understood as more near the actual boundary of image object after smooth.
Preferred further, described described smooth treatment is:
Determine the intersection point of cut-off rule and each square side, if the division line segment of these intersection point both sides in respective square all without point of crossing, then this intersection point mobile, makes the angle between the division line segment of these intersection point both sides increase.
First the intersection point of cut-off rule and each square side is determined, judge the division line segment of these intersection point both sides again in respective square with or without point of crossing, this point of crossing is the intersection point dividing line segment in same square, if point of crossing exists, keep the position of intersecting point of current cut-off rule and each square side constant, if without corresponding point of crossing, current intersection point is moved in side, and the angle between the division line segment of these intersection point both sides is increased.
The present invention has following advantage:
1, the present invention is by carrying out the abstract process in feature based flow field to image, recycle the border that a kind of closed boundary extracting method extracts image, then border is coordinated to carry out color samples, finally utilize Poisson image reconstruction drawing image, the image change that recovers can be made continuous, visual effect is smooth, has similarity with abstracted image.
2, the border that the present invention extracts is closed, and the situation of blend of colors mistake not easily appears in the image utilizing vector data of the present invention to draw out other drawing images of comparing.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet that abstracted image of the present invention carries out vectorization method.
Fig. 2 is the schematic flow sheet of image boundary extraction of the present invention.
Fig. 3 is the position view dividing line segment in square of the present invention.
The schematic diagram of intersection point adjustment when Fig. 4 is smooth treatment of the present invention.
Fig. 5 is color samples schematic diagram of the present invention.
Embodiment
As shown in Figure 1, a kind of method of abstracted image being carried out to vector quantization, comprises following implementation step:
(1) utilize feature to keep abstraction method to carry out abstract process to pending image and obtain abstracted image.
Feature keeps abstracting process to comprise the flow field using structure tensor computed image, then carries out line integral convolution operation according to flow field to image,
Structure tensor expression formula is:
Wherein, I
xfor each pixel gradient in the horizontal direction, I
yfor each pixel is in the gradient of vertical direction, the smaller of the eigenwert of this matrix is the characteristic direction of pixel.
The expression formula of line integral convolution operation is:
In formula, i
0for the two-dimensional coordinate of current pixel, the position of C (i) pointed by the characteristic direction of current pixel, w (i) is weight information, Z represents the surrounding pixel coordinate set contiguous with current pixel, i represents the element of z on the characteristic direction of current pixel, I is pixel value, the pixel value that I (C (i)) is C (i) position.
Weight information w (i
0, expression formula i) is:
w(x,y)＝G(||y-x||,σ
s)G(|I(y)-I(x)|,σ
r)
Wherein, the i described in x, y correspondence
0, i, σ
sand σ
rrepresent the parameter of Gaussian function, I (y) and I (x) represents the pixel value of x and y position respectively.
(2) Boundary Extraction is carried out to abstracted image and obtain image boundary, as shown in Figure 2:
A abstracted image is transformed into Lab color space by (), for L component, color data scope is divided into 16 regions, and give each region and number accordingly, each zone number is from 0 to 15, the span in each region is 16, each pixel number of abstracted image is converted into the numbering in region corresponding to this pixel native color numerical value, namely each pixel number of abstracted image is converted into the numerical value in 0 ~ 15.
B all four pixels of mutually adjoining, for L component, are configured to square by (), utilize division line segment to split each square, make to number identical summit in square and be in the same area, and the end points of described division line segment is in the mid point of square side.
Four pixels are configured in square, and when there is different colors in square, wherein the boundary line of two kinds of colors is division line segment.
As shown in Figure 3, all four pixels of mutually adjoining are configured to square, when four pixels are same color (see a figure), in the square that these four pixels are built into, only there is a kind of color, illustrate that this square is not on the border of image, therefore, in this square, there is not division line segment.
As shown in Figure 3 b, have eight squares in b figure, have two colors in each square, dividing line section is positioned on the boundary line of these two colors.
As shown in Figure 3c, have six squares in c figure, have three kinds of different colours in each square, dividing line section is positioned on the boundary line of arbitrary two kinds of colors.
As shown in Figure 3d, in the square in d figure, there are four kinds of colors, divide the boundary line that line segment is four kinds of colors.
Divide line segment in 3 in figure and will number identical vertex partition in square at the same area, and the end points that every bar divides line segment is in the mid point of square side, in part square, divide line segment and there is point of crossing.
C (), to any two squares that are connected, will have the division line segment of intersection point to be interconnected to form cut-off rule, the common composing images border of outward flange of this cut-off rule and abstracted image.
After having the division line segment of intersection point to be interconnected to obtain cut-off rule, also need to carry out smooth treatment to cut-off rule, the method of smooth treatment is: the intersection point first determining cut-off rule and each square side, judge the division line segment of these intersection point both sides again in respective square with or without point of crossing, this point of crossing is the intersection point dividing line segment in same square, if point of crossing exists, keep the position of intersecting point of current cut-off rule and each square side constant, if without corresponding point of crossing, current intersection point is moved along the square side at this intersection point place in side, and the ratio of the displacement of intersection point and the square length of side is 0 ~ 1:2, angle between the division line segment of these intersection point both sides is increased.
As shown in Figure 4, in figure, 4 is the image after nine squares are communicated with, wherein in each square, the line of band arrow is division line segment, be respectively and divide line segment ab, divide line segment bc and divide line segment cd, above-mentioned three division line segments are linked to be cut-off rule abcd, are increase the angle dividing line segment ab and divide between line segment cd to the object of cut-off rule abcd smooth treatment.A dotted line ac is built between a point and c point, dotted line ac is crossing with square side obtains e point, e point is b point along the point after square side transverse shifting, in like manner, between b point and d point, a dotted line bd, dotted line bd are crossing with square side obtains f point, f point for c point vertically move along square side after point, connect the cut-off rule that aefd obtains, be the image boundary after smooth treatment.
(3) utilize image boundary to sample on abstracted image, obtain the colouring information matched with each summit of square, namely obtain vectorial images data;
As shown in Figure 5, the oblique line with arrow in figure represents image boundary, according to the colouring information on each summit of the square at oblique line place, draws 4 colouring informations of oblique line, is respectively L1, L2, R1 and R2, wherein L1=L2=100, R1=R2=160.This schematic diagram being a kind of situation does not comprise the whole circumstances.
(4) utilize Poisson image rebuilding method that vectorial images data are carried out drafting and obtain grating images.
Claims (5)
1. abstracted image is carried out to a method for vector quantization, it is characterized in that, comprise following implementation step:
1) utilize feature to keep abstraction method to carry out abstract process to pending image and obtain abstracted image;
2) Boundary Extraction is carried out to abstracted image and obtain image boundary;
A) abstracted image is transformed into Lab color space, for L component, color data scope is divided into n region, and give each region and number accordingly, each pixel number of abstracted image is converted into the numbering in region corresponding to this pixel native color numerical value;
B) for described L component, all four pixels of mutually adjoining are configured to square, utilize division line segment to split each square, make in square, to number identical summit and be in the same area, and the end points of described division line segment is in the mid point of square side;
C) to any two squares that are connected, by there being the division line segment of intersection point to be interconnected to form cut-off rule, carry out smooth treatment to described cut-off rule, the outward flange of this cut-off rule and abstracted image forms described image boundary jointly;
Described smooth treatment is: the intersection point determining cut-off rule and each square side, if the division line segment of these intersection point both sides in respective square all without point of crossing, then this intersection point mobile, makes the angle between the division line segment of these intersection point both sides increase;
3) utilize image boundary to sample on described abstracted image, obtain the colouring information matched with each summit of square, namely obtain vectorial images data;
4) utilize Poisson image rebuilding method that described vectorial images data are carried out drafting and obtain grating images.
2. method of abstracted image being carried out to vector quantization as claimed in claim 1, it is characterized in that, described step 1) in feature keep abstracting process to comprise the flow field using structure tensor computed image, then according to flow field, line integral convolution operation is carried out to image.
3. method of abstracted image being carried out to vector quantization as claimed in claim 2, it is characterized in that, described structure tensor expression formula is:
Wherein, I
xfor each pixel gradient in the horizontal direction, I
yfor each pixel is in the gradient of vertical direction, the reckling of the eigenwert of this matrix is the characteristic direction of pixel.
4. method of abstracted image being carried out to vector quantization as claimed in claim 2, is characterized in that, the expression formula of described line integral convolution operation is:
In formula, i
0for the two-dimensional coordinate of current pixel, the position of C (i) pointed by the characteristic direction of current pixel, w (i
0, i) be weight information,
zrepresent the surrounding pixel coordinate set contiguous with current pixel, i represents
zelement on the characteristic direction of current pixel, I is pixel value, the pixel value that I (C (i)) is C (i) position.
5. method of abstracted image being carried out to vector quantization as claimed in claim 4, is characterized in that, described weight information w (i
0, expression formula i) is:
w(x,y)＝G(||y-x||,σ
s)G(|I(y)-I(x)|,σ
r)
Wherein, the i described in x, y correspondence
0, i, σ
sand σ
rrepresent the parameter of Gaussian function, I (x) and I (y) represents the pixel value of x and y position respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310109647.3A CN103198496B (en) | 2013-03-29 | 2013-03-29 | A kind of method of abstracted image being carried out to vector quantization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310109647.3A CN103198496B (en) | 2013-03-29 | 2013-03-29 | A kind of method of abstracted image being carried out to vector quantization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103198496A CN103198496A (en) | 2013-07-10 |
| CN103198496B true CN103198496B (en) | 2015-08-26 |
Family
ID=48721001
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310109647.3A Expired - Fee Related CN103198496B (en) | 2013-03-29 | 2013-03-29 | A kind of method of abstracted image being carried out to vector quantization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103198496B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103389953B (en) * | 2013-07-30 | 2016-08-24 | 北京东方惠尔图像技术有限公司 | A kind of image Beamforming Method and device |
| ES2738319T3 (en) * | 2014-09-12 | 2020-01-21 | Microsoft Technology Licensing Llc | Computer system to train neural networks |
| CN105550363B (en) * | 2016-01-06 | 2019-03-22 | 北京联合大学 | The semi-structured representation method of image based on XGML |
| CN114526812A (en) * | 2022-01-05 | 2022-05-24 | 广西防城港核电有限公司 | Abnormal sound monitoring method and system for mechanical equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3718967B2 (en) * | 1997-08-22 | 2005-11-24 | コニカミノルタビジネステクノロジーズ株式会社 | Image feature amount extraction apparatus and method, and recording medium on which image feature amount extraction program is recorded |
| CN102881028A (en) * | 2012-05-22 | 2013-01-16 | 浙江大学 | Method for quickly vectorizing grid digital image |
-
2013
- 2013-03-29 CN CN201310109647.3A patent/CN103198496B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3718967B2 (en) * | 1997-08-22 | 2005-11-24 | コニカミノルタビジネステクノロジーズ株式会社 | Image feature amount extraction apparatus and method, and recording medium on which image feature amount extraction program is recorded |
| CN102881028A (en) * | 2012-05-22 | 2013-01-16 | 浙江大学 | Method for quickly vectorizing grid digital image |
Non-Patent Citations (3)
| Title |
|---|
| Image Abstraction by Structure Adaptive Filtering;Jan Eric Kyprianidis等;《Proceedings of EG UK Theory and Practice of Computer Graphics》;20081231;摘要，第51页左栏第1段-57页左栏第1段 * |
| 一种光栅图像SVG矢量化表示的方法;胡卫军等;《计算机工程与科学》;20100815;第32卷(第8期);摘要，第64页右栏第2段-66页左栏第3段 * |
| 虚拟植物叶片的可视化建模技术研究;王立臣;《中国优秀硕士学位论文全文数据库》;20100115;正文20页第1段及21页4.2.2节，图4-5 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103198496A (en) | 2013-07-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103198496B (en) | A kind of method of abstracted image being carried out to vector quantization | |
| CN102930518B (en) | Improved sparse representation based image super-resolution method | |
| CN108073931A (en) | A kind of complex background image goes down unless the method for character and graphic | |
| CN102567731A (en) | Extraction method for region of interest | |
| Dey et al. | Voronoi-based feature curves extraction for sampled singular surfaces | |
| CN112598793B (en) | Three-dimensional map tile vector grid integrated system and method | |
| Shen et al. | A new approach to simplifying polygonal and linear features using superpixel segmentation | |
| Jung | Imgaug documentation | |
| CN108022285A (en) | Map rendering intent and device | |
| CN107689050A (en) | A kind of depth image top sampling method based on Color Image Edge guiding | |
| CN110232693A (en) | A kind of combination thermodynamic chart channel and the image partition method for improving U-Net | |
| CN103632153A (en) | Region-based image saliency map extracting method | |
| Song et al. | Mosaicking UAV orthoimages using bounded Voronoi diagrams and watersheds | |
| Shen et al. | A progressive method for the collapse of river representation considering geographical characteristics | |
| Yu et al. | An integrated method for DEM simplification with terrain structural features and smooth morphology preserved | |
| CN107038729B (en) | Digital instrument panel drawing method based on OpenGL-ES | |
| US9542724B1 (en) | Systems and methods for stroke rendering on digital maps | |
| CN102136151A (en) | Method for vectorizing raster image | |
| Smith et al. | A method for animating children’s drawings of the human figure | |
| CN104166715A (en) | Vxworks platform electronic chart engine | |
| CN113648655A (en) | Rendering method and device of virtual model, storage medium and electronic equipment | |
| CN107992532A (en) | Based on the method for searching three-dimension model for rendering image angle architectural feature | |
| CN103942819A (en) | Method and device for drawing geologic symbols | |
| CN103971365A (en) | Extraction method for image saliency map | |
| Zheng et al. | Semantic segmentation sample augmentation based on simulated scene generation—case study on dock extraction from high spatial resolution imagery |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150826 Termination date: 20210329 |