CN107506401A - A kind of image retrieval rearrangement method based on drawing method - Google Patents
A kind of image retrieval rearrangement method based on drawing method Download PDFInfo
- Publication number
- CN107506401A CN107506401A CN201710653201.5A CN201710653201A CN107506401A CN 107506401 A CN107506401 A CN 107506401A CN 201710653201 A CN201710653201 A CN 201710653201A CN 107506401 A CN107506401 A CN 107506401A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msup
- image
- msub
- prime
- 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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
Abstract
The invention discloses a kind of image retrieval rearrangement method based on drawing method, by entering rearrangement to the retrieval result of image, optimizes existing ranking results and draws closer to user's desired result;The similitude of neighbour's set between every image is calculated using neighbour's set of original ranking results;Similitude close to set is being regarded as to the similitude of image;Finally using similitude new between image come retrieval result of resequencing.The present invention proposes the image retrieval patterning process based on original ranking results, the similarity that this method is gathered using neighbour constructs the directly proportional Local map of correlation between image, and enter rearrangement with using patterning results, higher retrieval precision is obtained under the conditions of theorem 1 is met.
Description
Technical field
The present invention relates to technical field of image processing, more particularly to a kind of image retrieval rearrangement method based on drawing method.
Background technology
With the continuous development of social network sites, internet and multimedia technology, the multi-medium data such as video, image, audio
Increase with surprising rapidity.For example, Google retrievals are annual to increase about 1,000,000,000,000 new index;
Facebook registered users already exceed 1,000,000,000, monthly upload the picture more than 1,000,000,000.In image
Quantity is ever-increasing simultaneously, and different classes of image is also being enriched constantly.For example, architectural style image, geographic information image, doctor
Learn image, criminal investigation image (footprint, face, fingerprint etc.) and trade mark drm image etc..As the Internet images information is rich
Richness, for mass image data while user is met to image demand, caused noise information is also more and more.In large nuber of images
Found in storehouse and the image retrieval technologies of the useful image information of user are had become with one of priority research areas of retrieval technique.
The content of the invention
The purpose of the present invention is for large nuber of images information, proposes a kind of method for improving retrieval precision.
A kind of image retrieval rearrangement method based on drawing method provided by the invention, by being carried out to the retrieval result of image
Reset, optimize existing ranking results and draw closer to user's desired result;Gathered using the neighbour of original ranking results and calculated
Go out the similitude that neighbour gathers between every image;Similitude close to set is being regarded as to the similitude of image;Finally utilize
New similitude comprises the following steps come retrieval result of resequencing between image:
Step 1:Utilize the similitude of the German number structure image neighbour of outstanding card;
Define an image collection X={ x1,…,xn},xnIt is the numbering of image in image collection;xqIt is image to be retrieved
Or center image,It is xqKNN set;It is by xqPrimitive character preceding k recently
Neighbour's composition;The similitude between image is built using Jie Kade likeness coefficients;Definition set common factor size is big except set union
The small similitude between set:
Step 2:The German number figure of the outstanding card of renewal;
According to the weight map of the German number structure of outstanding card in step 1, the weight after the renewal of side right structure in weight map is updated
Figure is as follows:
Then, by w ' (x, xq) as the weight map after renewal;
Step 3:It is determined that last weight map;
X is determined according to the weight map built in step 2qFinal weight map calculation formula is as follows between x ∈ X:
Finally, by w (x, xq) weight as third layer;The expectation of the final weights built according to formula (3) is as follows:
Beneficial effect:The present invention proposes the image retrieval patterning process based on original ranking results, and this method is using closely
The similarity of neighbour's set constructs the directly proportional Local map of correlation between image, and enters rearrangement with using patterning results,
Meet to obtain higher retrieval precision under the conditions of theorem 1.
Brief description of the drawings
Fig. 1 gathers similitude by the neighbour of node near Jie Kade likeness coefficient node metrics A.
Side right of the Jie Kade similarity measure values in first layer weight map more than 0 is changed into 1. by Fig. 2 second layers weight map
Side right value of Fig. 3 third layer weight map using the side number summation between second layer weight node of graph as weight map.
Embodiment
For make present invention solves the technical problem that, the technical scheme that uses and the technique effect that reaches it is clearer, below
The present invention is described in further detail in conjunction with the accompanying drawings and embodiments.It is understood that specific implementation described herein
Example is used only for explaining the present invention, rather than limitation of the invention.It also should be noted that for the ease of description, accompanying drawing
In illustrate only part related to the present invention rather than full content.
The invention provides the rearrangement of the image retrieval based on drawing method, by entering rearrangement meter to the retrieval result of image
Calculate, optimize existing ranking results and draw the rearrangement method closer to user's desired result.This method utilizes original ranking results
Neighbour's set calculate the similitude that neighbour gathers between every image;Similitude close to set is being regarded as to the phase of image
Like property;Finally using similitude new between image come retrieval result of resequencing.The key step of rearrangement method is as follows:
Step 1:Utilize the similitude of the German number structure image neighbour of outstanding card
Define an image collection X={ x1,…,xn},xqIt is image to be retrieved (or center image),
It is xqKNN set.It is by xqPrimitive character preceding k arest neighbors composition.We are similar using Jie Kade
Similitude between property coefficient structure image.Wherein, Jie Kade similarity measurements are that a statistics is used for comparative sample collection
Similitude and multifarious metric form.Jie Kade similarity measurements are used to weigh the similitude between finite sample collection, fixed
Similitude of the justice set common factor size except set union size between set:
Step 2:The German number figure of the outstanding card of renewal
According to the weight map of the German number structure of outstanding card in step 1, the weight after the renewal of side right structure in weight map is updated
Figure is as follows:
Then, by w ' (x, xq) as the weight map after renewal.
Step 3:It is determined that last weight map
X is determined according to the weight map built in step 2qFinal weight map calculation formula is as follows between x ∈ X:
Finally, by w (x, xq) weight as third layer.So far, can be according to the set of feature similar neighborhoods and formula (3)
Build the weight map that arbitrary characteristics are reset.Compared to existing characteristic pattern developing algorithm, this paper algorithms can one except intuitively
Determine to eliminate outside the influence that image manifold nearby has boundary point pair retrieval effectiveness in degree.The side right value of weight map is built general
Close to the similarity probabilities between image in rate.The expectation of the final weights built according to formula (3) is as follows:
The overall step of patterning process is as shown in Figure 1 to Figure 3.
Theoretical proof
In order to facilitate description image between similarity relationships, defined function C () (if
Then xiWith xjIt is similar image;IfThen xiWith xjIt is not similar image).Further to prove
TTNG sides weight w (x, xq) and P (C (x, xq)=1) (x and xqSimilar probability) relation, provide following theorem herein:
Theorem 1:If following two condition is satisfied:
(1) P (C (and x, a)=1 | x ∈ Nk(a))=P (C (x, b)=1 | x ∈ Nk(b))=p
(2) | V | > > k
(3)
Then,
Prove:
Before theorem 1 is proved, following two lemma and proof are provided first:
Lemma 1:If following two condition is satisfied:
(1) P (C (and x, a)=1 | x ∈ Nk(a))=P (C (x, b)=1 | x ∈ Nk(b))=p
(2) | V | > > k
Then,
Prove:
By the conclusion of lemma 1, following equilibrium relationships can be obtained according to condition probability formula:
Wherein, k-th of neighbour that Section 1 is node a on the right side of equation meets to combine N when belonging to node b preceding k neighbourk
And N (a)k(b) common factor size is α probability, and similarly Section 2 represents that node a k-th of neighbour is not belonging to save on the right side of equation
Meet to combine N during point b preceding k neighbourkAnd N (a)k(b) common factor size is α probability.The ranking results of different characteristic are mutual
It is independent, therefore:
Further derive and understand:
It was found from above-mentioned derivation, lemma 1 must be demonstrate,proved.
Lemma 2:If following two condition is satisfied:
(1) P (C (and x, a)=1 | x ∈ Nk(a))=P (C (x, b)=1 | x ∈ Nk(b))=p
(2) | V | > > k
Then,
Prove:
By the conclusion of lemma 2, following equilibrium relationships can be obtained according to condition probability formula:
Wherein, k-th of neighbour that Section 1 is node a on the right side of equation meets to combine N when belonging to node b preceding k neighbourk
And N (a)k(b) common factor size is α probability, and similarly Section 2 represents that node a k-th of neighbour is not belonging to save on the right side of equation
Meet to combine N during point b preceding k neighbourkAnd N (a)k(b) common factor size is α probability.The ranking results of different characteristic are mutual
It is independent, therefore:
Further derive and understand:
It was found from above-mentioned derivation, lemma 2 must be demonstrate,proved.
By after the proof of lemma 1 and lemma 2.From the conclusion of theorem 1, can be obtained using condition probability formula
Following derivation:
Next can be derived again according to condition probability formula:
P (J (a, b)=α)
=P (J (a, b)=α | C (a, b)=1) P (C (a, b)=1)+P (J (a, b)=α | C (a, b)=0) P (C (a,
B)=0)
Section 1 side is the probability of J (a, b)=α when image a and image b is associated picture on the right of above-mentioned equation, and equation is right
Side Section 2 is the probability that image a and image b is not associated picture.It follows that equation:
By lemma 1, lemma 2 and above-mentioned formula, can deduce:
As α=0, following equation can be obtained:
Above-mentioned equation P (C (a, b)=1) is unknown number to be solved, and the solution for easily drawing above-mentioned formula is
With 0.Due toHaveAnd P (C (a, b)=1) >=0, therefore P (C (a, b)=1)
=0.It is possible thereby to deduce:
Theorem 1 must be demonstrate,proved.
E (w (x, x are calculated by the conclusion of theorem 1q)) as follows:
Pass through formula (5), w (x, xq) value expectation and P (C (x, xq)=1) it is proportional.So far, formula (3)
This paper patterning process is given, formula (5) gives the theoretical proof of context of methods reliability.
Finally it should be noted that:Various embodiments above is merely illustrative of the technical solution of the present invention, rather than its limitations;To the greatest extent
The present invention is described in detail with reference to foregoing embodiments for pipe, it will be understood by those within the art that:Its is right
Technical scheme described in foregoing embodiments is modified, and either which part or all technical characteristic are equally replaced
Change, the essence of appropriate technical solution is departed from the scope of various embodiments of the present invention technical scheme.
Claims (1)
1. a kind of image retrieval rearrangement method based on drawing method, by entering rearrangement to the retrieval result of image, optimization is existing
Ranking results draw closer to user desired result;Calculated using neighbour's set of original ranking results between every image
The similitude of neighbour's set;Similitude close to set is being regarded as to the similitude of image;Finally utilize phase new between image
Like property come retrieval result of resequencing, comprise the following steps:
Step 1:Utilize the similitude of the German number structure image neighbour of outstanding card;
Define an image collection X={ x1,…,xn},xnIt is the numbering of image in image collection;xqBe image to be retrieved or in
Heart image,It is xqKNN set;It is by xqPrimitive character preceding k arest neighbors composition;
The similitude between image is built using Jie Kade likeness coefficients;Definition set common factor size is set except set union size
Between similitude:
<mrow>
<mi>J</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<msup>
<mi>i</mi>
<mo>&prime;</mo>
</msup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<mo>|</mo>
<msub>
<mi>N</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>&cap;</mo>
<msub>
<mi>N</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<msup>
<mi>i</mi>
<mo>&prime;</mo>
</msup>
<mo>)</mo>
</mrow>
<mo>|</mo>
</mrow>
<mrow>
<mo>|</mo>
<msub>
<mi>N</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>&cup;</mo>
<msub>
<mi>N</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<msup>
<mi>i</mi>
<mo>&prime;</mo>
</msup>
<mo>)</mo>
</mrow>
<mo>|</mo>
</mrow>
</mfrac>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Step 2:The German number figure of the outstanding card of renewal;
According to the outstanding weight maps for blocking German number structure in step 1, the weight map after side right structure updates in renewal weight map is such as
Under:
Then, by w ' (x, xq) as the weight map after renewal;
Step 3:It is determined that last weight map;
X is determined according to the weight map built in step 2qFinal weight map calculation formula is as follows between x ∈ X:
<mrow>
<mi>w</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<msup>
<mi>x</mi>
<mi>q</mi>
</msup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
<mo>&Element;</mo>
<msub>
<mi>N</mi>
<msub>
<mi>k</mi>
<mn>2</mn>
</msub>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
</mrow>
</munder>
<msup>
<mi>w</mi>
<mo>&prime;</mo>
</msup>
<mrow>
<mo>(</mo>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
<mo>,</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
Finally, by w (x, xq) weight as third layer;The expectation of the final weights built according to formula (3) is as follows:
<mrow>
<mi>E</mi>
<mrow>
<mo>(</mo>
<mrow>
<mi>w</mi>
<mrow>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>,</mo>
<msup>
<mi>x</mi>
<mi>q</mi>
</msup>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>E</mi>
<mrow>
<mo>(</mo>
<mrow>
<munder>
<mi>&Sigma;</mi>
<mrow>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
<mo>&Element;</mo>
<msub>
<mi>N</mi>
<msub>
<mi>k</mi>
<mn>2</mn>
</msub>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
</mrow>
</munder>
<msup>
<mi>w</mi>
<mo>&prime;</mo>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
<mo>,</mo>
<mi>x</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>E</mi>
<mrow>
<mo>(</mo>
<mrow>
<munder>
<mi>&Sigma;</mi>
<mrow>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
<mo>&Element;</mo>
<mi>X</mi>
</mrow>
</munder>
<mrow>
<mo>&lsqb;</mo>
<mrow>
<mi>J</mi>
<mrow>
<mo>(</mo>
<mrow>
<msup>
<mi>x</mi>
<mo>&prime;</mo>
</msup>
<mo>,</mo>
<mi>x</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>></mo>
<mn>0</mn>
</mrow>
<mo>&rsqb;</mo>
</mrow>
</mrow>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4</mn>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710653201.5A CN107506401A (en) | 2017-08-02 | 2017-08-02 | A kind of image retrieval rearrangement method based on drawing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710653201.5A CN107506401A (en) | 2017-08-02 | 2017-08-02 | A kind of image retrieval rearrangement method based on drawing method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107506401A true CN107506401A (en) | 2017-12-22 |
Family
ID=60689790
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710653201.5A Pending CN107506401A (en) | 2017-08-02 | 2017-08-02 | A kind of image retrieval rearrangement method based on drawing method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107506401A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108875802A (en) * | 2018-05-30 | 2018-11-23 | 大连理工大学 | Measurement of Similarity between Two Images method based on multilayer neighbour |
CN109684493A (en) * | 2019-01-04 | 2019-04-26 | 重庆大学 | A kind of image rearrangement sequence method based on k neighborhood distribution score |
CN109783691A (en) * | 2018-12-29 | 2019-05-21 | 四川远鉴科技有限公司 | A kind of video retrieval method of deep learning and Hash coding |
CN112784086A (en) * | 2021-01-28 | 2021-05-11 | 北京有竹居网络技术有限公司 | Picture screening method and device, storage medium and electronic equipment |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120221572A1 (en) * | 2011-02-24 | 2012-08-30 | Nec Laboratories America, Inc. | Contextual weighting and efficient re-ranking for vocabulary tree based image retrieval |
CN106776896A (en) * | 2016-11-30 | 2017-05-31 | 董强 | A kind of quick figure fused images search method |
-
2017
- 2017-08-02 CN CN201710653201.5A patent/CN107506401A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120221572A1 (en) * | 2011-02-24 | 2012-08-30 | Nec Laboratories America, Inc. | Contextual weighting and efficient re-ranking for vocabulary tree based image retrieval |
CN106776896A (en) * | 2016-11-30 | 2017-05-31 | 董强 | A kind of quick figure fused images search method |
Non-Patent Citations (1)
Title |
---|
刘胜蓝等: "分组排序多特征融合的图像检索方法", 《计算机研究与发展》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108875802A (en) * | 2018-05-30 | 2018-11-23 | 大连理工大学 | Measurement of Similarity between Two Images method based on multilayer neighbour |
CN109783691A (en) * | 2018-12-29 | 2019-05-21 | 四川远鉴科技有限公司 | A kind of video retrieval method of deep learning and Hash coding |
CN109783691B (en) * | 2018-12-29 | 2022-06-21 | 北京远鉴信息技术有限公司 | Video retrieval method for deep learning and Hash coding |
CN109684493A (en) * | 2019-01-04 | 2019-04-26 | 重庆大学 | A kind of image rearrangement sequence method based on k neighborhood distribution score |
CN112784086A (en) * | 2021-01-28 | 2021-05-11 | 北京有竹居网络技术有限公司 | Picture screening method and device, storage medium and electronic equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111159425B (en) | Temporal knowledge graph representation method based on historical relationship and double-graph convolution network | |
Wu et al. | Contextual bandits in a collaborative environment | |
CN108804633B (en) | Content recommendation method based on behavior semantic knowledge network | |
CN107506401A (en) | A kind of image retrieval rearrangement method based on drawing method | |
CN107133277B (en) | A kind of tourist attractions recommended method based on Dynamic Theme model and matrix decomposition | |
WO2017041541A1 (en) | Method for pushing recommendation information, and server and storage medium | |
CN107301247B (en) | Method and device for establishing click rate estimation model, terminal and storage medium | |
CN108804551B (en) | Spatial interest point recommendation method considering diversity and individuation | |
CN109241405A (en) | A kind of associated education resource collaborative filtering recommending method of knowledge based and system | |
CN107220312B (en) | Interest point recommendation method and system based on co-occurrence graph | |
WO2014107988A1 (en) | Method and system for discovering and analyzing micro-blog user group structure | |
CN109104688A (en) | Wireless network access point model is generated using aggregation technique | |
CN107480694A (en) | Three clustering methods are integrated using the weighting selection evaluated twice based on Spark platforms | |
CN113255895B (en) | Structure diagram alignment method and multi-diagram joint data mining method based on diagram neural network representation learning | |
CN114265986B (en) | Information pushing method and system fusing knowledge graph structure and path semantics | |
CN107368534A (en) | A kind of method for predicting social network user attribute | |
CN110162692B (en) | User label determination method and device, computer equipment and storage medium | |
CN108401565B (en) | Remote sensing image registration method based on improved KAZE features and Pseudo-RANSAC algorithms | |
CN107679239A (en) | Recommend method in a kind of personalized community based on user behavior | |
CN108829804A (en) | Based on the high dimensional data similarity join querying method and device apart from partition tree | |
CN116340646A (en) | Recommendation method for optimizing multi-element user representation based on hypergraph motif | |
CN109948242A (en) | Network representation learning method based on feature Hash | |
Singh et al. | A theoretical approach to discover mutual friendships from social graph networks | |
CN106649731A (en) | Node similarity searching method based on large-scale attribute network | |
Ding et al. | Spatial-temporal distance metric embedding for time-specific POI recommendation |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171222 |