CN102402575A - Massive image data quick search method based on shapes - Google Patents
Massive image data quick search method based on shapes Download PDFInfo
- Publication number
- CN102402575A CN102402575A CN2011102704250A CN201110270425A CN102402575A CN 102402575 A CN102402575 A CN 102402575A CN 2011102704250 A CN2011102704250 A CN 2011102704250A CN 201110270425 A CN201110270425 A CN 201110270425A CN 102402575 A CN102402575 A CN 102402575A
- Authority
- CN
- China
- Prior art keywords
- shape
- image
- histogram
- retrieval
- data
- 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
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a massive image data quick search method based on shapes. The method comprises the following steps: firstly preprocessing each image among massive images to obtain image shape histogram data and main triangular elements and utilizing the data to constitute a shape feature database; then carrying out the same processing on target images to be searched currently; traversing all the image data in the database, searching out images which are the same as the main triangular elements of the searched image and adding the images into a subsample set to be searched; selecting the histogram of each image data in the subsample set and comparing the histogram of each image data in the subsample set with the histograms of the target images to be searched, thus obtaining the similarity degree between samples and the target images; and sequencing obtained search results and returning the sequenced result to a search user. The method disclosed by the invention has the advantages of fewer steps, simplicity, practicability, high search speed, good robustness, insensitivity to translation, rotation, stretching and contraction and suitability for shape search among massive image data.
Description
Technical field
The present invention relates to a kind of view data search method, belong to the computer image processing technology field.
Background technology
It is generally, human that an object is experienced the most intuitively is exactly its shape information.In the computer graphic image processing technology field, be a hot topic and important techniques direction based on the image retrieval of shape.
Existing image retrieval means, or speed can meet the demands, and still precision is too low; Perhaps when improving precision, reduced retrieval rate significantly, all be difficult to be applied in the large-scale CBIR and go.
For example; Paired how much histogram methods; This method is divided into the profile of image the set of directed line segment earlier, forms the set of two line segments through choosing any two line segments again, gathers through this and adds up angle and distance relation parameter θ and r between the directed line segment of forming object.Wherein θ represents the angle between these two line segments; R represents two distances (minor increment, ultimate range) between the line segment summit.In the pair set each is set up the geometric properties that a two-dimensional histogram is represented target shape to line segment statistical parameter θ and r.How much histogrammic methods are expressed easy in pairs; Retrieval rate is fast, on a large scale based on the retrieval of shape, use often, but the limitation of this method also clearly: at first θ and r two parameters are independent; θ obviously depends on r, and is not that yardstick is constant; Secondly, it has only considered the global characteristics of image, has but ignored local feature, and the different effects of the various piece of not considering shape during to unitary construction; In addition, this method also has a tangible drawback to be: it can not handle the line segment of equality, and parallel lines are modal relations in the description of culture wire frame.Because these defectives make that how much histogrammic methods scope in the application of reality has received very big restriction in pairs.During practical application, how much histogrammic methods are better for non-regular body effect in pairs, and are unacceptable for the regular body effect.
For another example, the application of the method for invariant moments in present image retrieval is also many.Moment characteristics has mainly characterized the geometric properties of image-region; Be called geometric moment again; Seven square invariants that directly utilize the standardized central moment under the cartesian coordinate system to derive are because it has the invariant features of characteristics such as rotation, translation, yardstick, so be called invariant moments again.In Flame Image Process, geometric invariant moment is represented object as an important characteristic, and characteristic is come image operation such as classify in view of the above.Below be the definition (η wherein of seven invariant moments using always
IjRepresent standardization central moment):
H
1＝η
20+η
02
H
2＝(η
20-η
02)
2+4η
11 2
H
3＝(η
30-3η
12)
2+(3η
21-η
03)
2
H
4＝(η
30+η
12)
2+(η
21+η
03)
2
H
5＝(η
30-3η
12)(η
30+η
12)[(η
30+η
12)
2-3(η
21+η
03)
2]
+(3η
21-η
03)(η
21+η
03)[3(η
30+η
12)
2-(η
21+η
03)
2]
H
6＝(η
20-η
02)[(η
30+η
12)
2-(η
21+η
03)
2]
+4η
11(η
30+η
12)(η
21+η
03)
H
7＝(3η
21-η
03)(η
21+η
03)[3(η
30+η
12)
2-(η
21+η
03)
2]
-(η
30-3η
12)(η
21+η
03)[3(η
30+η
12)
2-(η
21+η
03)
2
But the method retrieval rate of invariant moments is fast inadequately, and the instant calculated amount during retrieval is too big, is difficult in the reality be applied to go in the large-scale view data; Next is that the method for invariant moments can produce bigger influence to last result for local miniature deformation, and these too responsive characteristics are not that we are desirable.In addition, the method for invariant moments is to be absorbed in the global characteristics of image equally, is not well embodied for local feature.
Present massive image retrieval technology is main with global characteristics all, the local feature of less relevance image, but the local feature of image often is again a key factor very important in the image retrieval, this is that a desirable retrieval technique must be considered.Present technology all is difficult in mass data, to resolve the problem of local deformation, is one of major issue that influences the CBIR result and how to handle local deformation.Because the existence of these problems has hindered the widespread use of CBIR.
Summary of the invention
The objective of the invention is for solving the processing of the local deformation problem in the mass data, on the basis of considering local feature information, to propose a kind of novel search method in order to overcome the existing deficiency of search method on efficient and speed.This method not only can be retrieved mass image data fast, and can reach more accurate retrieval effectiveness, also can better handle local deformation problem and the better influence of considering local feature information to result for retrieval in the mass data simultaneously.
The technical scheme that the present invention adopted is following:
A kind of mass image data method for quickly retrieving based on shape.
Step 1, in the large nuber of images each is opened image carry out pre-service, the pre-service result of all images is formed a character shape data storehouse.Wherein, the single image in the large nuber of images being carried out pretreated method is:
At first, the shape information that the method for use image segmentation will be processed image extracts, and the profile result who splits is represented with a bianry image.Then, according to the shape information that the preceding step extracts, calculate the approximate polygon that this is processed image.After obtaining approximate polygon, get all end points on it, and shine upon shape border information through the information of these shape key points as its shape key point.Afterwards, the shape key point that the preceding step is obtained is carried out triangulation and combination, the set of all possible triangular element that is combined into is referred to as the combined triangular shape set of this shape.Subsequently, the square similarity of all triangular elements and equilateral triangle in the calculation combination triangle set, and the square similarity that all triangular elements obtain is polymerized to a histogram.After obtaining the histogram data of picture shape, choose in the histogram peaked row and can determine the main triangular element of this picture shape through this characteristic mark as the characteristic of this picture shape mark.At last, result is deposited in the character shape data storehouse, be used for follow-up retrieval.
Step 2, the target image of current required retrieval is carried out pre-service, obtain its feature histogram and its main triangular element mark, do not go but do not deposit in the character shape data storehouse that step 1 sets up.Pretreated method is identical with the said method of step 1.
All view data in step 3, the traversal character shape data storehouse; If the main triangular element mark of a certain view data is identical with the main triangular element mark of the target image of current required retrieval; Concentrate the subsample that then this view data is joined desire retrieval, otherwise continue the next view data of traversal.By that analogy, all view data also get subsample collection to the end in traversal gestalt shape property data base.
The histogram of the histogram of each view data that concentrate step 4, the subsample that step 3 is selected and the target image of current required retrieval compares, and can obtain the similarity degree of these samples and target image.At last, the result for retrieval that obtains is sorted once more, sorted result is returned to retrieval user.
Beneficial effect
Because the inventive method is implemented on the information of represented as histograms, the characteristic of represented as histograms is to be applied in easily among the retrieval of mass image data.Through carrying out leg-of-mutton fractionation and combination to shape information during pre-processing, can well consider of the influence of local feature information to result for retrieval, can well handle the problem of the local deformation of shape simultaneously.Because this method is the calculating of on the basis of approximate polygon, carrying out, overcome the too responsive character of change small in the Invariant Moment Method to shape.This method contrast prior art, step is few, simple, retrieval rate is fast, robustness is good, and be to translation, rotation, flexiblely have an insensitivity; Since pretreatment stage has been transferred in a lot of calculating, little in the instant calculated amount of real retrieval phase, therefore be well suited for the shape retrieval in the mass image data.
Description of drawings
Fig. 1 is the process flow diagram of the inventive method;
Fig. 2 is the picture shape synoptic diagram that is retrieved in the inventive method;
Fig. 3 is a shape key point synoptic diagram in the inventive method;
Fig. 4 is for carrying out triangle combination synoptic diagram to the shape key point that obtains;
Fig. 5 is for describing picture shape histogram synoptic diagram.
Embodiment
Elaborate below in conjunction with the accompanying drawing specific embodiments of the invention.
A kind of mass image data method for quickly retrieving based on shape; Its implementation procedure is as shown in Figure 1, the description that it is detailed how from piece image, to obtain needed histogram information, how the view data of magnanimity carried out pre-service, how to extract the characteristic information of target image and used it for the process of mass image data retrieval.Specific as follows:
Step 1, in the large nuber of images each is opened image carry out pre-service, the pre-service result of all images is formed a character shape data storehouse.Wherein, it is following the single image in the large nuber of images to be carried out pretreated detailed process:
At first; The shape information that the method for use image segmentation will be processed image extracts (can use some traditional image partitioning algorithms to realize; Such as region growing algorithm, Canny edge detection algorithm, active contour model algorithm, Level Set Method or the like), and the profile result who splits represented with a bianry image.
Then, according to the shape information that the preceding step extracts, calculate the approximate polygon that this is processed image.The shape of image can be simplified and well expressed to this approximate polygon.After obtaining approximate polygon, get all end points on it, and shine upon shape border information through the information of these shape key points as its shape key point.Under the orderly prerequisite of shape key point, quantity of information that the shape key point is entrained and the entrained quantity of information of whole approximate polygon are equal to, like Fig. 2, shown in Figure 3.
Afterwards, the shape key point that the preceding step is obtained is carried out triangulation and combination.At first these shape key points are carried out triangulation; Result according to triangulation is combined into a plurality of triangular elements; The set of all possible triangular element that is combined into is referred to as the combined triangular shape set of this shape, and the triangular element synoptic diagram is as shown in Figure 4.Owing to the combination that the picture shape contour representation has been become triangular element, the influence that the local deformation that therefore can well reduce causes global shape.
Subsequently, the square similarity of all triangular elements and equilateral triangle in the calculation combination triangle set, and the square similarity that all triangular elements obtain is polymerized to a histogram.Specifically can obtain the square similarity of they and equilateral triangle through invariant moments relatively, the method that amount of this use is described the similar program of a diabolo has reduced the redundancy of information to greatest extent, has improved the density of information.The histogram that obtains is as shown in Figure 5, because invariant moments has the invariant features of rotation, translation, convergent-divergent, so the histogram that this method obtains at last equally also satisfies the unchangeability of rotation, translation, convergent-divergent.
Because the histogram of the picture shape that is obtained, its peaked row can mark the characteristic of this picture shape.Therefore, after obtaining the histogram data of picture shape, choose in the histogram peaked row and can determine the main triangular element of this picture shape through this characteristic mark as the characteristic of this picture shape mark.
At last, result is deposited in the character shape data storehouse, be used for follow-up retrieval.
Step 2, the target image of current required retrieval is carried out pre-service, obtain its feature histogram and its main triangular element mark, do not go but do not deposit in the character shape data storehouse that step 1 sets up.Pretreated method is identical with the said method of step 1.
All view data in step 3, the traversal character shape data storehouse; If the main triangular element mark of a certain view data is identical with the main triangular element mark of the target image of current required retrieval; Concentrate the subsample that then this view data is joined desire retrieval, otherwise continue the next view data of traversal.By that analogy, all view data also get subsample collection to the end in traversal gestalt shape property data base.So just dwindle the scope that needs retrieval greatly, reduced instant calculated amount, accelerated the speed of retrieval.
The histogram of the histogram of each view data that concentrate step 4, the subsample that step 3 is selected and the target image of current required retrieval compares, and can obtain the similarity degree of these samples and target image.Histogrammic comparison can use the Bhattacharyya distance of normal distribution comparison to carry out.The histogram that relatively this method is generated of this mode has extraordinary effectiveness.At last, the result for retrieval that obtains is like this carried out a minor sort according to the Bhattacharyya distance again, sorted result is returned to retrieval with producing.
So far, just accomplished the quick retrieval of mass image data.
Claims (1)
1. mass image data method for quickly retrieving based on shape is characterized in that may further comprise the steps:
Step 1, in the large nuber of images each is opened image carry out pre-service, the pre-service result of all images is formed a characteristic information database; Wherein, the single image in the large nuber of images being carried out pretreated method is:
At first, the shape information that the method for use image segmentation will be processed image extracts, and the profile result who splits is represented with a bianry image; Then, according to the shape information that the preceding step extracts, calculate the approximate polygon that this is processed image; After obtaining approximate polygon, get all end points on it, and shine upon shape border information through the information of these shape key points as its shape key point; Afterwards, the shape key point that the preceding step is obtained is carried out triangulation and combination, the set of all possible triangular element that is combined into is referred to as the combined triangular shape set of this shape; Subsequently, the square similarity of all triangular elements and equilateral triangle in the calculation combination triangle set, and the square similarity that all triangular elements obtain is polymerized to a histogram; After obtaining the histogram data of picture shape, choose in the histogram peaked row and can determine the main triangular element of this picture shape through this characteristic mark as the characteristic of this picture shape mark; At last, result is deposited in the character shape data storehouse, be used for follow-up retrieval;
Step 2, the target image of current required retrieval is carried out pre-service, obtain its feature histogram and its main triangular element mark, do not go but do not deposit in the character shape data storehouse that step 1 sets up; Pretreated method is identical with the said method of step 1;
All view data in step 3, the traversal character shape data storehouse; If the main triangular element mark of a certain view data is identical with the main triangular element mark of the target image of current required retrieval; Concentrate the subsample that then this view data is joined desire retrieval, otherwise continue the next view data of traversal; By that analogy, all view data also get subsample collection to the end in traversal gestalt shape property data base;
The histogram of the histogram of each view data that concentrate step 4, the subsample that step 3 is selected and the target image of current required retrieval compares, and can obtain the similarity degree of these samples and target image; At last, the result for retrieval that obtains is sorted once more, sorted result is returned to retrieval user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102704250A CN102402575A (en) | 2011-09-14 | 2011-09-14 | Massive image data quick search method based on shapes |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102704250A CN102402575A (en) | 2011-09-14 | 2011-09-14 | Massive image data quick search method based on shapes |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102402575A true CN102402575A (en) | 2012-04-04 |
Family
ID=45884778
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011102704250A Pending CN102402575A (en) | 2011-09-14 | 2011-09-14 | Massive image data quick search method based on shapes |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102402575A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105046256A (en) * | 2015-07-22 | 2015-11-11 | 福建新大陆自动识别技术有限公司 | Distortion image correction based QR (quick response) code decoding method and system |
| WO2017045443A1 (en) * | 2015-09-16 | 2017-03-23 | 杭州海康威视数字技术股份有限公司 | Image retrieval method and system |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101996245A (en) * | 2010-11-09 | 2011-03-30 | 南京大学 | Form feature describing and indexing method of image object |
-
2011
- 2011-09-14 CN CN2011102704250A patent/CN102402575A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101996245A (en) * | 2010-11-09 | 2011-03-30 | 南京大学 | Form feature describing and indexing method of image object |
Non-Patent Citations (2)
| Title |
|---|
| 刘巍: "基于形状和姿态的WWW图像检索研究", 《中国优秀博硕士学位论文全文数据库（硕士） 信息科技辑》 * |
| 贾元元: "基于形状的图像数据库检索技术研究", 《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 (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105046256A (en) * | 2015-07-22 | 2015-11-11 | 福建新大陆自动识别技术有限公司 | Distortion image correction based QR (quick response) code decoding method and system |
| CN105046256B (en) * | 2015-07-22 | 2018-10-16 | 福建新大陆自动识别技术有限公司 | QR codes coding/decoding method based on distorted image correction and system |
| WO2017045443A1 (en) * | 2015-09-16 | 2017-03-23 | 杭州海康威视数字技术股份有限公司 | Image retrieval method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104573705B (en) | A kind of clustering method of building Point Cloud of Laser Scanner | |
| CN103207879B (en) | The generation method and apparatus of image index | |
| CN104850633B (en) | A kind of three-dimensional model searching system and method based on the segmentation of cartographical sketching component | |
| CN106373118B (en) | The complex curved surface parts point cloud compressing method of border and local feature can be effectively retained | |
| CN103207898B (en) | A kind of similar face method for quickly retrieving based on local sensitivity Hash | |
| CN107369161A (en) | A kind of workpiece point cloud segmentation method at random based on the European cluster of improvement | |
| CN105139379B (en) | Based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface | |
| CN102402621A (en) | Image retrieval method based on image classification | |
| CN105740378B (en) | Digital pathology full-section image retrieval method | |
| CN105205135B (en) | A kind of 3D model retrieval methods and its retrieval device based on topic model | |
| CN106874421A (en) | Image search method based on self adaptation rectangular window | |
| Nunes et al. | Shape based image retrieval and classification | |
| CN110348478B (en) | Method for extracting trees in outdoor point cloud scene based on shape classification and combination | |
| Li et al. | Shrec’16 track: 3D sketch-based 3D shape retrieval | |
| CN110737788B (en) | Rapid three-dimensional model index establishing and retrieving method | |
| CN102708367A (en) | Image identification method based on target contour features | |
| CN103336970B (en) | Distributed images based on many subtrees training and search method | |
| CN104965928B (en) | One kind being based on the matched Chinese character image search method of shape | |
| CN102402575A (en) | Massive image data quick search method based on shapes | |
| Yuan et al. | Automatic cables segmentation from a substation device based on 3D point cloud | |
| CN106127851A (en) | A kind of method of three-dimensional point cloud object detection based on curved surface segmentation | |
| CN106547825B (en) | Method for searching three-dimension model based on global characteristics and block feature | |
| CN113192172B (en) | Airborne LiDAR ground point cloud simplification method | |
| CN106384127B (en) | The method and system of comparison point pair and binary descriptor are determined for image characteristic point | |
| Zhang et al. | Improved SLAM closed-loop detection algorithm based on DBoW2 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20120404 |