CN104751459A - Multi-dimensional feature similarity measuring optimizing method and image matching method - Google Patents
Multi-dimensional feature similarity measuring optimizing method and image matching method Download PDFInfo
- Publication number
- CN104751459A CN104751459A CN201510140372.9A CN201510140372A CN104751459A CN 104751459 A CN104751459 A CN 104751459A CN 201510140372 A CN201510140372 A CN 201510140372A CN 104751459 A CN104751459 A CN 104751459A
- Authority
- CN
- China
- Prior art keywords
- multidimensional characteristic
- value
- feature
- descriptor
- coding
- 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.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a multi-dimensional feature similarity measuring optimizing method, and belongs to the technical field of intersection of the mode recognizing technology and the computer technology. The method comprises the steps of performing binary encoding according to the data variation feature of the original multi-dimensional feature describers to obtain binary encoding features; measuring the original multi-dimensional feature describers according to the Hamming distance between the binary encoding features. The invention further discloses an image matching method. Compared with the prior art, the method has the advantages that the feature similarity can be quickly measured, the difficulty at transplanting of feature matching algorithm hardware is reduced, the power consumption of a single-point processing module can be decreased, the utilization rate of a hardware resource of the algorithm can be reduced, and therefore, the performance of the whole feature matching module can be effectively optimized.
Description
Technical field
The present invention relates to a kind of similarity measurement optimization method of multidimensional characteristic, belong to the technical field that mode identification technology and computer technology intersect.
Background technology
Along with the development of computer technology, characteristic matching is used widely at numerous areas such as information retrieval, pattern-recognition, image procossing.Characteristic matching is realized by the similarity between measures characteristic descriptor usually, and for the Image Feature Point Matching in image processing techniques, it is mainly divided into two classes: a class is the linear scanning method compared based on single-point; Another kind of is compare based on division and set up the method that data directory carries out Rapid matching.Conventional images characteristic matching is all unique point and the feature descriptor thereof of first Calculation Basis figure and real-time figure, then Calculation Basis figure mates with the arest neighbors of the key point of real-time figure [David MARSHALL.Nearest Neighbor Searching In High Dimensional MetricSpace [EB/OL] .http: //escience.anu.edu.Au/pro-ject/06S1/DavidMarshall.2006-06-21/2009-12], determines whether matching double points with this.In conventional solution, all Euclidean distance is adopted to the similarity measurement of arest neighbors coupling, namely the SIFT feature point of benchmark image and realtime graphic is first extracted, after setting up Feature Descriptor respectively, according to the size of arest neighbors and time neighbour's ratio, the unique point in judgment standard image whether with the Feature Points Matching in realtime graphic.
Along with scientific-technical progress, various character description method is suggested, and these features often have high dimension (such as, the dimension of SIFT feature is 128), and on the one hand, these features can describe some attribute of sample better more accurately; On the other hand, it is also for the specific implementation of feature similarity measurement brings series of problems.Specifically, traditional matching scheme based on Euclidean distance has following shortcoming:
1) resources occupation rate is high.For single-point coupling, generally, each SIFT feature point has 128 dimensions, and each dimension is carried out floating number by a byte (8) and represented, then a unique point just needs the storage space of 128 bytes;
2) hardware of algorithm is transplanted large.Matching algorithm based on Euclidean distance is carried out in hardware migration process, relate generally to summation operation and product calculation (division also can regard a kind of product calculation as), and the complexity of these two kinds of computings (time complexity and space complexity) with input data space complexity and input data number proportional;
3) arithmetic speed is low.For single-point coupling, the clock frequency of computing affects larger by the dimension size inputting data.Suppose that each unique point has 128 dimensions, each dimension is carried out floating number by a byte (8) and is represented, after the enterprising line algorithm of FPGA (model is Zynq-7000 XC7Z020-1CLG484C) is transplanted, its maximum clock frequency is only 78MHz.
Summary of the invention
Technical matters to be solved by this invention is to overcome the deficiencies in the prior art, a kind of similarity measurement optimization method of multidimensional characteristic is provided, the speed of feature similarity measurement can be accelerated, reduce the difficulty of Feature Correspondence Algorithm hardware transplanting and the power consumption of standalone processes module, reduce the hardware resource occupancy of algorithm, effectively optimize the performance of whole characteristic matching module.
The present invention specifically solves the problems of the technologies described above by the following technical solutions:
A similarity measurement optimization method for multidimensional characteristic, its data variation characteristic first based on original multidimensional characteristic descriptor carries out binary-coding to it, obtains binary-coding feature; Then utilize the Hamming distance between binary-coding feature to measure the similarity between original multidimensional characteristic descriptor.
One of preferred version, specifically in accordance with the following methods binary-coding is carried out to original multidimensional characteristic descriptor:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1/0, otherwise,
value be 0/1, m ∈ m|m=1,2 ..., N-1};
according to
with
difference and a predetermined threshold value between magnitude relationship determine,
with
difference be greater than described threshold value, then
value be 1/0, otherwise,
value be 0/1, m ∈ m|m=1,2 ..., N-1}.
Preferred version two, specifically in accordance with the following methods binary-coding is carried out to original multidimensional characteristic descriptor:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1/0, otherwise,
value be 0/1, m ∈ m|m=1,2 ..., N}.
Preferred version three, specifically in accordance with the following methods binary-coding is carried out to original multidimensional characteristic descriptor:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1/0, otherwise,
value be 0/1;
represent the average of each dimensional feature component in original multidimensional characteristic descriptor.
Following technical scheme can also be obtained according to identical invention thinking:
A kind of image matching method, utilizes the similarity between characteristics of image to carry out images match, it is characterized in that, the similarity between described characteristics of image uses similarity measurement optimization method described in upper arbitrary technical scheme to measure.
Compared to existing technology, difficulty transplanted by the hardware that the present invention significantly can reduce Feature Correspondence Algorithm, improves the arithmetic speed of characteristic matching, and greatly reduces hardware resource occupancy, be with a wide range of applications.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the existing SIFT feature method for measuring similarity based on Euclidean distance;
Fig. 2 is the process flow diagram of similarity measurement optimization method of the present invention;
Fig. 3 is the principle schematic that Feature Descriptor carries out binaryzation coding.
A kind of binary-coding algorithm flow chart that Fig. 4 uses for similarity measurement optimization method of the present invention;
A kind of algorithm flow chart calculating Hamming distance that Fig. 5 uses for similarity measurement optimization method of the present invention.
Embodiment
Below in conjunction with accompanying drawing, technical scheme of the present invention is described in detail:
In the characteristic matching process in the fields such as image procossing, pattern-recognition, information retrieval, because intrinsic dimensionality is higher and the calculating of Euclidean distance is more complicated, feature similarity measurement is caused to need to consume a large amount of hardware resources (comprising storage, computational resource), high to hardware requirement, the difficulty that hardware is transplanted is large.For the image matching algorithm based on SIFT feature, the unique point for carrying out mating generally uses 128 dimensional vector information of place metric space to characterize.And SIFT algorithm outstanding advantages is exactly volume, that is: even if target object is little, also can generate more unique point, this just makes the benchmark image with large-size often have nearly thousands of unique points, and the unique point that the realtime graphic that size is less has is also nearly hundreds of.Image characteristic point institute information too much, becomes to hinder and improves one of images match speed, the principal element promoting business throughput, and the bulky hardware resource needed for it is also for the hardware transplanting of algorithm brings great difficulty.
For solving this problem, the present invention proposes a kind of similarity measurement optimization method of new multidimensional characteristic, its basic thought first carries out binary-coding based on its data variation characteristic of original multidimensional characteristic descriptor to it, obtains binary-coding feature; Then utilize the Hamming distance between binary-coding feature to measure the similarity between original multidimensional characteristic descriptor.For the ease of public understanding, still for the images match based on SIFT feature, technical solution of the present invention is described in detail below.
Fig. 1 shows the basic procedure of the existing SIFT feature method for measuring similarity based on Euclidean distance.For the SIFT feature descriptor of one 128 dimension
(subscript f represents that the data type of current descriptor is floating number, to distinguish the descriptor that two-value is below expressed), generally, every one dimension of descriptor all represents by floating number.Carry out in hardware transplanting at algorithm, at least need to carry out floating-point numerical representation with 8 bits, like this, a unique point descriptor needs 128 × 8=1024 bit.
Fig. 2 shows the basic procedure of similarity measurement optimization method of the present invention.As shown in Figure 2, first read in data to be encoded A sequence and B sequence, wherein A sequence is the one point data of benchmark image, and B sequence is the one point data of realtime graphic.A, B sequence is respectively 128 dimensions, and each dimension byte (8 bits) represents; Then respectively binary-coding is carried out to A sequence and B sequence; Finally XOR is carried out to the A sequence of binary-coding, B sequence, calculate its Hamming distance.Binary-coding method is wherein core of the present invention, it is encoded based on the data variation feature of primitive character descriptor (A sequence, B sequence) self, be equivalent to primitive character descriptor for object, then extract himself data variation feature.The present invention preferably according to current dimension characteristic component and and its dimensional characteristics component separated by a distance between variation relation carry out binary-coding, the specific descriptions of the program are as follows:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1/0, otherwise,
value be 0/1, m ∈ m|m=1,2 ..., N-1};
according to
with
difference and a predetermined threshold value between magnitude relationship determine,
with
difference be greater than described threshold value, then
value be 1/0, otherwise,
value be 0/1, m ∈ m|m=1,2 ..., N-1}.
Such as, following binary-coding method can be adopted when the value of m is 1 time:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1 (or being 0), otherwise,
value be 0 (being 1);
according to
with
difference and a predetermined threshold value Threshold between magnitude relationship determine,
with
difference be greater than described threshold value, then
value be 1 (or being 0), otherwise,
value be 0 (being 1).
Or, if original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1 (or being 0), otherwise,
value be 0 (being 1);
according to
with
difference and a predetermined threshold value Threshold between magnitude relationship determine,
with
difference be greater than described threshold value, then
value be 1 (or being 0), otherwise,
value be 0 (being 1).
Above-mentioned binary-coding side ratio juris as shown in Figure 3.After adopting this encoding scheme, then the SIFT feature descriptor of former needs 1024 bit storage space only needs 128 × 2=256 bit, significantly reduces required storage space.In addition, in above-mentioned binary-coding scheme, d
i2 2introducing be uniqueness in order to improve data, also can not have in practical application, namely every one dimension component of primitive character only uses a two-value number to encode, thus reduces storage space further.The value of Threshold is lower, then the uniqueness of two valued description obtained is better, but can reduce the recall ratio of algorithm simultaneously, the concrete value of Threshold can according to for primitive character determined by test.Through verification experimental verification, for SIFT feature, its value is 0.05 better.
Certainly, concrete binary-coding also can adopt alternate manner, such as, using the component of a certain specific dimension of primitive character (such as the first dimension component) as benchmark, corresponding two-value number is determined with each dimensional feature component and the magnitude relationship between benchmark, specific as follows:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1 (or being 0), otherwise,
value be 0 (being 1), m ∈ m|m=1,2 ..., N}.
Or, respectively tie up the average of component as benchmark using primitive character, determine corresponding two-value number with each dimensional feature component and the magnitude relationship between benchmark, specific as follows:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1/0, otherwise,
value be 0/1;
represent the average of each dimensional feature component in original multidimensional characteristic descriptor.
Above-mentioned two schemes also can introduce d further as the first preferred version
i2 2, thus improve the uniqueness of data.
Fig. 4, Fig. 5 show the algorithm examples adopting the inventive method to carry out SIFT feature similarity measurement, and wherein Fig. 4 is the algorithm realization flow process of binary-coding, and Fig. 5 is the algorithm flow calculating Hamming distance.
The concrete steps of binary-coding as shown in Figure 4, comprising:
Step one: first, 128 of A sequence floating datas read in successively in FIFO (first-in first-out), by this FIFO called after F_A, F_A is made up of 128 eight bit registers; Then, 128 of B sequence floating datas read in successively in FIFO (first-in first-out), by this FIFO called after F_B, F_B is made up of 128 eight bit registers.
Step 2: carry out binary-coding to A sequence, the deposit data after coding is in register C, and C is 256 bit registers.Wherein, C [0] determines with A [127] with the coding A [0] of C [1]:
If 0<A [0]-A [127] <threshold, then C [0]=1, C [1]=0;
If A [0]-A [127] >threshold, then C [0]=1, C [1]=1;
If 0<A [127]-A [0] <threshold, then C [0]=0, C [1]=0;
If A [0]-A [127] >threshold, then C [0]=0, C [1]=1.
And the coding A [k] of C [2 × k] and C [2 × k+1] and A [k-1] determine, wherein 1<k<127:
If 0<A [k]-A [k-1] <threshold, then C [2 × k]=1, C [2 × k+1]=0;
If A [k]-A [k-1] >threshold, then C [2 × k]=1, C [2 × k+1]=1;
If 0<A [k-1]-A [k] <threshold, then C [2 × k]=0, C [2 × k+1]=0;
If A [k-1]-A [k] >threshold, then C [2 × k]=0, C [2 × k+1]=1.
Step 3: carry out binary-coding to B sequence, the deposit data after coding is in register D, and D is 256 bit registers.Wherein, D [0] determines with B [127] with the coding B [0] of D [1]:
If 0<B [0]-B [127] <threshold, then D [0]=1, D [1]=0;
If B [0]-B [127] >threshold, then D [0]=1, D [1]=1;
If 0<B [127]-B [0] <threshold, then D [0]=0, D [1]=0;
If B [0]-B [127] >threshold, then D [0]=0, D [1]=1.
And the coding B [k] of D [2 × k] and D [2 × k+1] and B [k-1] determine, wherein 1<k<127:
If 0<B [k]-B [k-1] <threshold, then D [2 × k]=1, D [2 × k+1]=0;
If B [k]-B [k-1] >threshold, then D [2 × k]=1, D [2 × k+1]=1;
If 0<B [k-1]-B [k] <threshold, then D [2 × k]=0, D [2 × k+1]=0;
If B [k-1]-B [k] >threshold, then D [2 × k]=0, D [2 × k+1]=1.
The calculation procedure of Hamming distance is as shown in Figure 5, specific as follows:
Step one: coding step two and step 3 are obtained, deposit in the two-value data step-by-step XOR of C register and D register respectively, and be kept in extension register by result, wherein E is 256 bit registers.
Step 2: the Hamming distance obtaining binary sequence in register E: first, judges whether E [0] equals 1, equals 1, then counter sum increases 1; Otherwise it is constant.Next clock, E sequence moves right 1, repeats to judge that E [0] is to determine whether sum increases 1, till E sequence 256 bit has all judged simultaneously.The then net result of counter sum, is sequence A, B converts the Hamming distance after binary sequence to.
Adopt the SIFT feature Point matching method of feature similarity measurement optimization method of the present invention, because carried out binary-coding, and adopt Hamming distance to substitute the metric form of Euclidean distance as characteristic matching, so only need a step XOR just can realize the tolerance of similarity, difficulty transplanted by the hardware greatly reducing matching algorithm, improve the arithmetic speed of images match, and greatly reduce hardware resource occupancy, be with a wide range of applications.
In order to verify effect of the present invention, same hardware platform (FPGA) compares and adopts tradition based on the similarity measurement of Euclidean distance and adopt in similarity measurement optimization method situation of the present invention, the hardware resource shared by SIFT feature matching algorithm and can flank speed be reached.Table 1 shows comparative result, and wherein Euclidean distance represents the similarity measurement adopting tradition based on Euclidean distance, and Hamming distance represents employing method for measuring similarity of the present invention.
Hardware resource under the different similarity measurement mode of table 1 and matching speed contrast
As can be seen from Table 1, under same platform (FPGA), adopt the matching way of method for measuring similarity of the present invention to take less resource (the former is about the latter's 27.4%) than classic method, but have processing speed (speed promotes about 304%) faster.
Claims (8)
1. a similarity measurement optimization method for multidimensional characteristic, is characterized in that, its data variation characteristic first based on original multidimensional characteristic descriptor carries out binary-coding to it, obtains binary-coding feature; Then utilize the Hamming distance between binary-coding feature to measure the similarity between original multidimensional characteristic descriptor.
2. the similarity measurement optimization method of multidimensional characteristic as claimed in claim 1, is characterized in that, specifically carry out binary-coding to original multidimensional characteristic descriptor in accordance with the following methods:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1/0, otherwise,
value be 0/1, m ∈ m|m=1,2 ..., N-1};
according to
with
difference and a predetermined threshold value between magnitude relationship determine,
with
difference be greater than described threshold value, then
value be 1/0, otherwise,
value be 0/1, m ∈ m|m=1,2 ..., N-1}.
3. the similarity measurement optimization method of multidimensional characteristic as claimed in claim 2, it is characterized in that, the value of m is 1.
4. the similarity measurement optimization method of multidimensional characteristic as described in Claims 2 or 3, it is characterized in that, described original multidimensional characteristic descriptor is image SIFT feature descriptor.
5. the similarity measurement optimization method of multidimensional characteristic as claimed in claim 4, it is characterized in that, described threshold value is 0.05.
6. the similarity measurement optimization method of multidimensional characteristic as claimed in claim 1, is characterized in that, specifically carry out binary-coding to original multidimensional characteristic descriptor in accordance with the following methods:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1/0, otherwise,
value be 0/1, m ∈ m|m=1,2 ..., N}.
7. the similarity measurement optimization method of multidimensional characteristic as claimed in claim 1, is characterized in that, specifically carry out binary-coding to original multidimensional characteristic descriptor in accordance with the following methods:
If original multidimensional characteristic descriptor
n is intrinsic dimensionality, and subscript f represents that its data type is floating number; Order
subscript 2 represents that its data type is two-value number, i=1,2 ..., N; Namely the binary-coding feature of original multidimensional characteristic descriptor is obtained; Wherein,
according to
with
magnitude relationship determine,
be greater than
then
value be 1/0, otherwise,
value be 0/1;
represent the average of each dimensional feature component in original multidimensional characteristic descriptor.
8. an image matching method, utilizes the similarity between characteristics of image to carry out images match, it is characterized in that, the similarity between described characteristics of image uses similarity measurement optimization method described in any one of claim 1 ~ 7 to measure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510140372.9A CN104751459B (en) | 2015-03-27 | 2015-03-27 | Multi-dimensional feature similarity measuring optimizing method and image matching method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510140372.9A CN104751459B (en) | 2015-03-27 | 2015-03-27 | Multi-dimensional feature similarity measuring optimizing method and image matching method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104751459A true CN104751459A (en) | 2015-07-01 |
| CN104751459B CN104751459B (en) | 2017-05-17 |
Family
ID=53591078
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510140372.9A Expired - Fee Related CN104751459B (en) | 2015-03-27 | 2015-03-27 | Multi-dimensional feature similarity measuring optimizing method and image matching method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104751459B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127243A (en) * | 2016-06-22 | 2016-11-16 | 上海师范大学 | A kind of image matching method describing son based on binaryzation SIFT |
| CN106778771A (en) * | 2016-11-22 | 2017-05-31 | 上海师范大学 | A kind of new two-value SIFT descriptions and its image matching method |
| CN108615006A (en) * | 2018-04-23 | 2018-10-02 | 百度在线网络技术（北京）有限公司 | Method and apparatus for output information |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120045130A1 (en) * | 2010-08-17 | 2012-02-23 | Hon Hai Precision Industry Co., Ltd. | Electronic device and method for matching images |
| CN103778619A (en) * | 2012-10-17 | 2014-05-07 | 华中科技大学 | Image matching method based on Zernike matrix |
| CN104081435A (en) * | 2014-04-29 | 2014-10-01 | 中国科学院自动化研究所 | Image matching method based on cascading binary encoding |
| CN104166977A (en) * | 2013-05-17 | 2014-11-26 | 中国航空工业集团公司洛阳电光设备研究所 | Image matching similarity measuring method and image matching method thereof |
| CN104392431A (en) * | 2014-10-27 | 2015-03-04 | 华东师范大学 | Image matching method based on image variable length coding |
-
2015
- 2015-03-27 CN CN201510140372.9A patent/CN104751459B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120045130A1 (en) * | 2010-08-17 | 2012-02-23 | Hon Hai Precision Industry Co., Ltd. | Electronic device and method for matching images |
| CN103778619A (en) * | 2012-10-17 | 2014-05-07 | 华中科技大学 | Image matching method based on Zernike matrix |
| CN104166977A (en) * | 2013-05-17 | 2014-11-26 | 中国航空工业集团公司洛阳电光设备研究所 | Image matching similarity measuring method and image matching method thereof |
| CN104081435A (en) * | 2014-04-29 | 2014-10-01 | 中国科学院自动化研究所 | Image matching method based on cascading binary encoding |
| CN104392431A (en) * | 2014-10-27 | 2015-03-04 | 华东师范大学 | Image matching method based on image variable length coding |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127243A (en) * | 2016-06-22 | 2016-11-16 | 上海师范大学 | A kind of image matching method describing son based on binaryzation SIFT |
| CN106778771A (en) * | 2016-11-22 | 2017-05-31 | 上海师范大学 | A kind of new two-value SIFT descriptions and its image matching method |
| CN108615006A (en) * | 2018-04-23 | 2018-10-02 | 百度在线网络技术（北京）有限公司 | Method and apparatus for output information |
| CN108615006B (en) * | 2018-04-23 | 2020-04-17 | 百度在线网络技术（北京）有限公司 | Method and apparatus for outputting information |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104751459B (en) | 2017-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2017092183A1 (en) | Image retrieval method based on variable-length deep hash learning | |
| CN107682319B (en) | Enhanced angle anomaly factor-based data flow anomaly detection and multi-verification method | |
| CN103218427B (en) | The extracting method of local description, image search method and image matching method | |
| CN103425639A (en) | Similar information identifying method based on information fingerprints | |
| CN102682024B (en) | Method for recombining incomplete JPEG file fragmentation | |
| CN104050247A (en) | Method for realizing quick retrieval of mass videos | |
| CN110110113A (en) | Image search method, system and electronic device | |
| CN103955952A (en) | Extraction and description method for garment image color features | |
| CN109325510A (en) | A kind of image characteristic point matching method based on lattice statistical | |
| CN103473307A (en) | Cross-media sparse Hash indexing method | |
| US20170017717A1 (en) | Sequence Data Analyzer, DNA Analysis System and Sequence Data Analysis Method | |
| CN103020321B (en) | Neighbor search method and system | |
| CN104881449A (en) | Image retrieval method based on manifold learning data compression hash | |
| TW202217597A (en) | Image incremental clustering method, electronic equipment, computer storage medium thereof | |
| CN104751459A (en) | Multi-dimensional feature similarity measuring optimizing method and image matching method | |
| CN110097581B (en) | Method for constructing K-D tree based on point cloud registration ICP algorithm | |
| CN104361135A (en) | Image search method | |
| CN103839074A (en) | Image classification method based on matching of sketch line segment information and space pyramid | |
| CN110110120B (en) | Image retrieval method and device based on deep learning | |
| CN109389172B (en) | Radio signal data clustering method based on non-parameter grid | |
| Li et al. | Coverless Video Steganography Based on Frame Sequence Perceptual Distance Mapping. | |
| EP3528133A1 (en) | Data stream join method and device | |
| Hu et al. | Feature reduction of multi-scale LBP for texture classification | |
| CN108536772B (en) | Image retrieval method based on multi-feature fusion and diffusion process reordering | |
| CN109409407A (en) | A kind of industry monitoring data clustering method based on LE algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170517 Termination date: 20210327 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |