CN107194424B - A kind of image similar block method for fast searching - Google Patents
A kind of image similar block method for fast searching Download PDFInfo
- Publication number
- CN107194424B CN107194424B CN201710358857.4A CN201710358857A CN107194424B CN 107194424 B CN107194424 B CN 107194424B CN 201710358857 A CN201710358857 A CN 201710358857A CN 107194424 B CN107194424 B CN 107194424B
- Authority
- CN
- China
- Prior art keywords
- image block
- image
- similar
- block
- subgraph
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides a kind of image similar block method for fast searching, using each pixel and its neighborhood territory pixel, determines image block corresponding to each pixel；Using the similarity of image block in neighborhood and the gray value of image block, the feature vector for indicating each image block is determined；Uniform down-sampling is carried out by the pixel to entire image, determines a small-scale thumbnail set of blocks；Using thumbnail set of blocks, the K Wei Shu for indicating the image block set is constructed, entire image is divided into 9 width subgraphs without overlapping；Using image block contained by every width subgraph, the K Wei Shu for indicating the subgraph is constructed, treats search image block, determines image block similar with its in index K dimension tree；Determine the subgraph K Wei Shu most containing similar image block number mesh, image block similar with its is searched out in determining target subgraph K dimension tree, the similar image block is image block similar with image block to be searched, realizes the fast search of similar image block inside image.
Description
Technical field
The present invention relates to image procossing and technical field of computer vision more particularly to a kind of image similar block fast searchs
Method.
Background technique
Image procossing is widely used to the fields such as public safety, medical diagnosis, military and national defense and daily life.So
And limited and the interference of external environment by imaging device, image acquired in imaging device usually contain certain noise or
Person's resolution ratio is lower, to influence the accuracy of subsequent processing.For this reason, it may be necessary to be utilized under the premise of not changing image-forming condition
Image filtering technology handles image, to achieve the purpose that inhibit noise or increase resolution ratio.
Image itself implies a large amount of redundancy, the form of expression are as follows: image memory in many similar image blocks, this
A little similar image blocks are likely located at the different location that the same region may also be distributed in image.Utilize these similar image blocks
Image denoising and super-resolution may be implemented.Common image filtering technology includes part filter and non-local filtering.Part filter
Similitude of the Pohle between image block to be processed image block adjacent thereto, by adding to the image block in these regional areas
Weight average obtains filtered pixel.Non-local filtering is searched out all similar to image block to be processed in entire image
Image block, filtered pixel is then generated by these image blocks weighted average.The treatment effect of part filter is usually poor,
And non-local filtering, although effect is better than part filter, computation complexity is very high, and Calculation bottleneck essentially consists in similar diagram
As the search of block.
Summary of the invention
In order to overcome the deficiencies in the prior art described above, the present invention provides a kind of image similar block method for fast searching, searches
Suo Fangfa includes:
Step 1: utilizing each pixel and its neighborhood territory pixel, determine image block corresponding to each pixel；
Step 2: using the similarity of image block in neighborhood and the gray value of image block, determining the feature for indicating each image block
Vector；
Step 3: uniform down-sampling being carried out by the pixel to entire image, determines a small-scale thumbnail block collection
It closes；
Step 4: utilizing thumbnail set of blocks, construct the K Wei Shu for indicating the image block set, be denoted as index K dimension
Tree；
Step 5: entire image is divided into 9 width subgraphs without overlapping；
Step 6: using image block contained by every width subgraph, constructing the K Wei Shu for indicating the subgraph, be denoted as subgraph K dimension
Tree；
Step 7: treating search image block, determine image block similar with its in index K dimension tree；
Step 8: determining the subgraph K Wei Shu, referred to as target subgraph K Wei Shu most containing similar image block number mesh；
Step 9: searching out image block similar with its, the similar diagram in the dimension tree of the target subgraph K determined by step 7
As block is image block similar with image block to be searched.
Preferably, image block corresponding to each pixel of the step 1 determines that method is as follows: each pixel and its 5 × 5 is adjacent
Pixel definition in domain is an image block.
Preferably, the feature vector of each image block of the step 2 determines that method is as follows:
(1) each image block contains 25 pixels, is denoted as p1,p2,…,p25；
(2) using formula (1) calculate each image block and around it 8 image blocks similarity, be denoted as s1,s2,…,s8；
A is control parameter in formula, and P is image block to be processed, PiFor i-th of image block around P；
(3) 25 pixels of image block and the similarity of image block 8 image blocks adjacent thereto are combined, and carries out normalizing
Change processing forms 33 dimensional feature vectors of an expression image block, is denoted as PF=(p1,p2,...,p25,s1,s2,...,s8)。
Preferably, the step 3 determines that the method for thumbnail set of blocks is as follows by entire image: from the image upper left corner
1st pixel starts in both the horizontal and vertical directions, the right side every 3 pixel samplings, 1 pixel, until sampling image
Until inferior horn；By the image block using centered on sampled pixel as thumbnail block.
Preferably, the building method of the step 4 index K Wei Shu is as follows:
(1) to the feature vector of all image blocks in thumbnail set of blocks, the side of every one-dimensional characteristic is calculated by formula (2)
Difference
V in formulaiValue is characterized,For the mean value of the dimensional feature, m is the number of thumbnail block；
(2) the maximum feature dimensions of variance are chosen, and find out the intermediate value of the dimension；
(3) using image block corresponding to the intermediate value as the root node of K Wei Shu；
(4) using the calculated intermediate value of maximum variance feature dimensions, thumbnail set of blocks is divided into two parts, the dimension
The image block that characteristic value is less than intermediate value is located in the left subtree of root node, and the image block which is greater than intermediate value is then located at root
In the right subtree of node；
(5) to image block contained by the subtree of left and right, above-mentioned steps (2) to (4) are repeated respectively, until thumbnail set of blocks not
Until energy is subdivided.
Preferably, the building method of the step 6 subgraph K Wei Shu is as follows:
(1) to the feature vector of all images block contained by subgraph, the variance of every one-dimensional characteristic is calculated by formula (2)；
(2) the maximum feature dimensions of variance are chosen, and find out the intermediate value of the dimension；
(3) using image block corresponding to the intermediate value as the root node of K Wei Shu；
(4) using the calculated intermediate value of maximum variance feature dimensions, all images block contained by subgraph is divided into two parts,
The image block that the dimensional feature value is less than intermediate value is located in the left subtree of root node, image block then position of the dimensional feature value greater than intermediate value
In the right subtree of root node；
(5) to image block contained by the subtree of left and right, above-mentioned steps (2) to (4) are repeated respectively, until subgraph image block set not
Until energy is subdivided.
Preferably, the step 7 index K ties up the determination method of image block similar with image block to be searched in tree such as
Under:
(1) since the root node of index K Wei Shu, in the feature vector and tree using formula (3) calculating image block to be searched
The similarity of present node feature vector
PF is the feature vector of image block to be searched, PF in formulaiThe feature vector of tree present node is tieed up for K, b is control ginseng
Number；
(2) for preset threshold value λ, if sfi> λ then continues to search in the subtree of the root node of present node
Rope；Otherwise, using present node and its all child nodes as image block similar with image block to be searched.
Preferably, the determination method of the step 8 target subgraph K Wei Shu is as follows: counting every stalk figure K dimension tree is included
Similar image block number mesh, selecting contained maximum that stalk figure K Wei Shu of number is target subgraph K Wei Shu to be determined.
Preferably, the determination method of the step 9 target subgraph K Wei Shu image block similar with image block to be searched is such as
Under:
(1) since the root node of target subgraph K Wei Shu, using formula (3) calculate the feature vector of image block to be searched with
The similarity of present node feature vector in tree；
(2) for preset threshold value λ, if sfi> λ then continues to search in the subtree of the root node of present node
Rope；Otherwise, using present node and its all child nodes as image block similar with image block to be searched.
As can be seen from the above technical solutions, the invention has the following advantages that
Image similar block method for fast searching can be realized the fast search of similar image block inside image, avoid part
The poor processing effect of filtering, the high technical problem of non-local filtering computation complexity.
Detailed description of the invention
In order to illustrate more clearly of technical solution of the present invention, attached drawing needed in description will be made below simple
Ground introduction, it should be apparent that, drawings in the following description are only some embodiments of the invention, for ordinary skill
For personnel, without creative efforts, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is flow chart of the invention；
Fig. 2 is image block and 8 tile location schematic diagrames around it；
Fig. 3 is image uniform down-sampling schematic diagram；
Fig. 4 is K dimension tree construction process schematic diagram；
Fig. 5 be image it is non-overlapping be divided into 9 width subgraph schematic diagrames.
Specific embodiment
It in order to make the invention's purpose, features and advantages of the invention more obvious and easy to understand, below will be with specific
Examples and drawings, the technical solution protected to the present invention are clearly and completely described, it is clear that implementation disclosed below
Example is only a part of the embodiment of the present invention, and not all embodiment.Based on the embodiment in this patent, the common skill in this field
Art personnel all other embodiment obtained without making creative work belongs to the model of this patent protection
It encloses.
The present invention provides a kind of image similar block method for fast searching, describes with reference to Fig. 1, the image phase of the embodiment of the present invention
Like block method for fast searching, comprising the following steps:
Step 1: to the given image I containing n pixel, being by the pixel definition in each pixel and its 5 × 5 neighborhoods
One image block shares n image block, is denoted as Pi, (i=1 ..., n)；
Step 2: to arbitrary image block Pi, with reference to Fig. 2 describe adjacent image block between positional relationship, utilize formula (4) meter
Calculate it and around it 8 image blocks similarity s1,s2,...,s8,
A is control parameter in formula, and P is image block to be processed, PiFor i-th of image block around P, by similarity value s1,
s2,...,s8With image block Pi25 pixel ps for being included1,p2,...,p25It is normalized, obtains an expression image
33 dimensional feature vector PF=(p of block1,p2,...,p25,s1,s2,...,s8)；
Step 3: it is described with reference to Fig. 3, since the 1st pixel in the upper left corner image I in both the horizontal and vertical directions,
Every 3 pixel samplings, 1 pixel, until sampling the lower right corner of image, by the image block centered on sampled pixel
As thumbnail block.
Step 4: to the feature vector of all image blocks in thumbnail set of blocks, calculating every one-dimensional characteristic by formula (5)
Variance, choose the maximum feature dimensions of variance and simultaneously calculate its intermediate value,
V in formulaiValue is characterized,For the mean value of the dimensional feature, m is the number of thumbnail block, is described with reference to Fig. 4 with this
Thumbnail set of blocks is divided into two parts by root node of the image block corresponding to intermediate value as K Wei Shu, and the dimensional feature value is small
It is located in the left subtree of root node in the image block of intermediate value, the image block which is greater than intermediate value is then located at the right side of root node
In subtree, 1 index K Wei Shu is obtained；
Step 5: with reference to Fig. 5 describe, by image I it is non-overlapping be divided into 9 width subgraphs；
Step 6: to the feature vector of image block contained by each subgraph, the variance of every one-dimensional characteristic is calculated by formula (5), is chosen
The maximum feature dimensions of variance simultaneously calculate its intermediate value, describe the root section using image block corresponding to the intermediate value as K Wei Shu with reference to Fig. 4
Thumbnail set of blocks is divided into two parts by point, and the image block which is less than intermediate value is located at the left subtree of root node
In, the image block which is greater than intermediate value is then located in the right subtree of root node, and 9 stalk figure K Wei Shu are obtained；
Step 7: since the root node of index K Wei Shu, feature vector and the tree of image block to be searched are calculated using formula (6)
The similarity of middle present node feature vector,
PF is the feature vector of image block to be searched, PF in formulaiThe feature vector of tree present node is tieed up for K, b is control ginseng
Number, for preset threshold value λ, if sfi> λ is then continued searching in the subtree of the root node of present node, otherwise will
Present node and its all child nodes are as image block similar with image block to be searched；
Step 8: count every stalk figure K dimension and set included similar image block number mesh, select contained number it is maximum that
Subgraph K Wei Shu is as target subgraph K Wei Shu；
Step 9: since the root node of target subgraph K Wei Shu, the feature vector of image block to be searched is calculated using formula (3)
With the similarity of present node feature vector in tree；For preset threshold value λ, if sfi> λ is then with present node
It is continued searching in the subtree of root node；Otherwise, using present node and its all child nodes as similar to image block to be searched
Image block.The similar image block that the step 9 obtains is image block similar with image block to be searched.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, of the invention
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (7)
1. a kind of image similar block method for fast searching, which is characterized in that searching method includes:
Step 1: utilizing each pixel and its neighborhood territory pixel, determine image block corresponding to each pixel；
Step 2: using the similarity of image block in neighborhood and the gray value of image block, determine indicate the feature of each image block to
Amount；
Step 3: uniform down-sampling being carried out by the pixel to entire image, determines a small-scale thumbnail set of blocks；
Step 4: utilizing thumbnail set of blocks, construct the K Wei Shu for indicating the image block set, be denoted as index K Wei Shu；
The building method for indexing K Wei Shu is as follows:
(1) to the feature vector of all image blocks in thumbnail set of blocks, the variance of every one-dimensional characteristic is calculated by formula (2)
V in formulaiValue is characterized,For the mean value of the dimensional feature, m is the number of thumbnail block；
(2) the maximum feature dimensions of variance are chosen, and find out the intermediate value of the dimension；
(3) using image block corresponding to the intermediate value as the root node of K Wei Shu；
(4) using the calculated intermediate value of maximum variance feature dimensions, thumbnail set of blocks is divided into two parts, the dimensional feature
The image block that value is less than intermediate value is located in the left subtree of root node, and the image block which is greater than intermediate value is then located at root node
Right subtree in；
(5) to image block contained by the subtree of left and right, above-mentioned steps (2) to (4) are repeated respectively, until thumbnail set of blocks cannot be again
Until division；
Step 5: entire image is divided into 9 width subgraphs without overlapping；
Step 6: using image block contained by every width subgraph, constructing the K Wei Shu for indicating the subgraph, be denoted as subgraph K Wei Shu；
The building method of the step 6 subgraph K Wei Shu is as follows:
(1) to the feature vector of all images block contained by subgraph, the variance of every one-dimensional characteristic is calculated by formula (2)；
(2) the maximum feature dimensions of variance are chosen, and find out the intermediate value of the dimension；
(3) using image block corresponding to the intermediate value as the root node of K Wei Shu；
(4) using the calculated intermediate value of maximum variance feature dimensions, all images block contained by subgraph is divided into two parts, the dimension
The image block that characteristic value is less than intermediate value is located in the left subtree of root node, and the image block which is greater than intermediate value is then located at root
In the right subtree of node；
(5) to image block contained by the subtree of left and right, above-mentioned steps (2) to (4) are repeated respectively, until subgraph image block set cannot be again
Until division；
Step 7: treating search image block, determine image block similar with its in index K dimension tree；
Step 8: determining the subgraph K Wei Shu, referred to as target subgraph K Wei Shu most containing similar image block number mesh；
Step 9: searching out image block similar with its, the similar image block in the dimension tree of the target subgraph K determined by step 7
Image block as similar with image block to be searched.
2. image similar block method for fast searching according to claim 1, which is characterized in that
Image block corresponding to each pixel of step 1 determines that method is as follows: by the pixel in each pixel and its 5 × 5 neighborhoods
It is defined as an image block.
3. image similar block method for fast searching according to claim 1, which is characterized in that
The feature vector of each image block of step 2 determines that method is as follows:
(1) each image block contains 25 pixels, is denoted as p1,p2,...,p25；
(2) using formula (1) calculate each image block and around it 8 image blocks similarity, be denoted as s1,s2,...,s8；
A is control parameter in formula, and P is image block to be processed, PiFor i-th of image block around P；
(3) 25 pixels of image block and the similarity of image block 8 image blocks adjacent thereto are combined, and place is normalized
Reason forms 33 dimensional feature vectors of an expression image block, is denoted as PF=(p1,p2,...,p25,s1,s2,...,s8)。
4. image similar block method for fast searching according to claim 1, which is characterized in that
The step 3 determines that the method for thumbnail set of blocks is as follows by entire image: opening from the 1st pixel in the image upper left corner
Begin in both the horizontal and vertical directions, every 3 pixel samplings, 1 pixel, until sampling the lower right corner of image；It will
Image block using centered on sampled pixel is as thumbnail block.
5. image similar block method for fast searching according to claim 1, which is characterized in that
The determination method of image block similar with image block to be searched is as follows in the step 7 index K dimension tree:
(1) it since the root node of index K Wei Shu, is calculated using formula (3) current in the feature vector and tree of image block to be searched
The similarity of node diagnostic vector
PF is the feature vector of image block to be searched, PF in formulaiThe feature vector of tree present node is tieed up for K, b is control parameter；
(2) for preset threshold value λ, if sfi> λ is then continued searching in the subtree of the root node of present node；It is no
Then, using present node and its all child nodes as image block similar with image block to be searched.
6. image similar block method for fast searching according to claim 1, which is characterized in that
The determination method of the step 8 target subgraph K Wei Shu is as follows: counting every stalk figure K dimension and sets included similar image
Block number mesh, selecting contained maximum that stalk figure K Wei Shu of number is target subgraph K Wei Shu to be determined.
7. image similar block method for fast searching according to claim 5, which is characterized in that
The determination method of the step 9 target subgraph K Wei Shu image block similar with image block to be searched is as follows:
(1) since the root node of target subgraph K Wei Shu, in the feature vector and tree using formula (3) calculating image block to be searched
The similarity of present node feature vector；
(2) for preset threshold value λ, if sfi> λ is then continued searching in the subtree of the root node of present node；It is no
Then, using present node and its all child nodes as image block similar with image block to be searched.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710358857.4A CN107194424B (en) | 2017-05-19 | 2017-05-19 | A kind of image similar block method for fast searching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710358857.4A CN107194424B (en) | 2017-05-19 | 2017-05-19 | A kind of image similar block method for fast searching |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107194424A CN107194424A (en) | 2017-09-22 |
| CN107194424B true CN107194424B (en) | 2019-08-27 |
Family
ID=59874130
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710358857.4A Active CN107194424B (en) | 2017-05-19 | 2017-05-19 | A kind of image similar block method for fast searching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107194424B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108733780B (en) * | 2018-05-07 | 2020-06-23 | 浙江大华技术股份有限公司 | Picture searching method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101976256A (en) * | 2010-11-01 | 2011-02-16 | 重庆大学 | Double nearest neighbour first searching method in point feature image registration |
| CN102129451A (en) * | 2011-02-17 | 2011-07-20 | 上海交通大学 | Method for clustering data in image retrieval system |
| CN104182974A (en) * | 2014-08-12 | 2014-12-03 | 大连理工大学 | A speeded up method of executing image matching based on feature points |
| CN104778281A (en) * | 2015-05-06 | 2015-07-15 | 苏州搜客信息技术有限公司 | Image index parallel construction method based on community analysis |
| CN106485651A (en) * | 2016-10-11 | 2017-03-08 | 中国人民解放军军械工程学院 | The image matching method of fast robust Scale invariant |
-
2017
- 2017-05-19 CN CN201710358857.4A patent/CN107194424B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101976256A (en) * | 2010-11-01 | 2011-02-16 | 重庆大学 | Double nearest neighbour first searching method in point feature image registration |
| CN102129451A (en) * | 2011-02-17 | 2011-07-20 | 上海交通大学 | Method for clustering data in image retrieval system |
| CN104182974A (en) * | 2014-08-12 | 2014-12-03 | 大连理工大学 | A speeded up method of executing image matching based on feature points |
| CN104778281A (en) * | 2015-05-06 | 2015-07-15 | 苏州搜客信息技术有限公司 | Image index parallel construction method based on community analysis |
| CN106485651A (en) * | 2016-10-11 | 2017-03-08 | 中国人民解放军军械工程学院 | The image matching method of fast robust Scale invariant |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107194424A (en) | 2017-09-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106919909B (en) | The metric learning method and system that a kind of pedestrian identifies again | |
| CN108596975B (en) | Stereo matching algorithm for weak texture region | |
| WO2018083963A1 (en) | System and method for processing input point cloud having points | |
| CN108510504A (en) | Image partition method and device | |
| Long et al. | Neuraludf: Learning unsigned distance fields for multi-view reconstruction of surfaces with arbitrary topologies | |
| JP6111745B2 (en) | Vehicle detection method and apparatus | |
| CN108537844B (en) | Visual SLAM loop detection method fusing geometric information | |
| CN103455991A (en) | Multi-focus image fusion method | |
| Wang et al. | Seggpt: Towards segmenting everything in context | |
| CN108664838A (en) | Based on the monitoring scene pedestrian detection method end to end for improving RPN depth networks | |
| CN108665509A (en) | A kind of ultra-resolution ratio reconstructing method, device, equipment and readable storage medium storing program for executing | |
| CN111105459B (en) | Descriptive sub map generation method, positioning method, device, equipment and storage medium | |
| CN109785246A (en) | A kind of noise-reduction method of non-local mean filtering, device and equipment | |
| CN109118544A (en) | Synthetic aperture imaging method based on perspective transform | |
| CN106980843A (en) | The method and device of target following | |
| CN113222819B (en) | Remote sensing image super-resolution reconstruction method based on deep convolution neural network | |
| CN112766123B (en) | Crowd counting method and system based on criss-cross attention network | |
| CN111105452A (en) | High-low resolution fusion stereo matching method based on binocular vision | |
| CN105608689B (en) | A kind of panoramic mosaic elimination characteristics of image error hiding method and device | |
| CN110348299A (en) | The recognition methods of three-dimension object | |
| CN105678720A (en) | Image matching judging method and image matching judging device for panoramic stitching | |
| CN103065291A (en) | Image fusion method based on promoting wavelet transform and correlation of pixel regions | |
| CN107194424B (en) | A kind of image similar block method for fast searching | |
| Lei et al. | SNLRUX++ for building extraction from high-resolution remote sensing images | |
| Lijun et al. | Geo‐information mapping improves Canny edge detection method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |