CN108595596A - A kind of image similarity search method - Google Patents
A kind of image similarity search method Download PDFInfo
- Publication number
- CN108595596A CN108595596A CN201810355429.0A CN201810355429A CN108595596A CN 108595596 A CN108595596 A CN 108595596A CN 201810355429 A CN201810355429 A CN 201810355429A CN 108595596 A CN108595596 A CN 108595596A
- Authority
- CN
- China
- Prior art keywords
- image
- retrieval
- algorithm
- vector
- extraction
- 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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Abstract
The invention discloses a kind of image similarity search methods, including picture identification, image correction process, image encrypting algorithm training, the extraction of image retrieval vector, image similarity calculating and recall precision to realize image similarity retrieval.The beneficial effects of the invention are as follows:The present invention extracts the characteristic area information of image so that retrieval result is more accurate using the method for retrieval vector;First, the problems such as unsupervised learning mode training image retrieval model is to avoid retrieval precision caused by subjective factor difference caused by artificially labelling from reducing is used, second is that retrieval takes at 0.3 second or so every time on treatment effeciency;Third, open private clound Retrieval Interface, user can complete image retrieval by downloading retrieval model in private clound.
Description
Technical field
The present invention relates to a kind of search method, specially a kind of image similarity search method belongs to search method application
Technical field.
Background technology
Presently, there are realization similar technique be a support to scheme to search figure, or the image retrieval retrieved according to label
Product.And its presently, there are deficiency be to need manually to label for training image sample and cause artificial subjectivity when establishing model
The difference of factor leads to that model accuracy is relatively low and retrieval model update is slow, cost is big, can not support private clound or local
Search operaqtion is carried out in net.
1. training pattern needs the peopleware for artificially labelling and labelling to differ, fabric retrieval precision is influenced;
2. the error for solving image rotation, scaling the retrieval result brought;
Above-mentioned similar technique there are the problem of, the demand in exactly current image retrieval market, at present image retrieval be badly in need of
Retrieval precision is high, and model modification is fast, cost is low, the search engine that private clound can be supported to retrieve.
Invention content
The purpose of the present invention is that solve the above-mentioned problems and provides a kind of image similarity search method.
The present invention is achieved through the following technical solutions above-mentioned purpose, a kind of image similarity search method, including picture
Image recognition, image correction process, image encrypting algorithm training, the extraction of image retrieval vector, image similarity calculates and retrieval
Efficiency realizes image similarity retrieval;
Wherein, described image correction process includes the following steps:
Step A, user's uploading pictures analyze the picture of upload;
If step B, there is missing, fuzzy region in the picture that user uploads or there are watermark picture, system roots
Certain adjustment is carried out according to normal region to missing, watermark or fuzzy region to sharpen, if having adjusted or having existed missing,
Allow user's uploading pictures again;
Described image retrieval model is trained:It is carried from training sample image using image information extraction algorithm and watershed algorithm
The data informations such as the RBG of image and Non-surveillance clustering neural network are taken, image encrypting algorithm training is completed;
Described image retrieves vector extraction:The extraction of retrieval vector, retrieval arrow are carried out to image according to image encrypting algorithm
Amount includes the characteristic area information of image;
Described image similarity calculation:Image to be retrieved calculates itself and figure after obtaining retrieval vector by image encrypting algorithm
As a few moral distances calculate similarity value in the normal form distance of the retrieval vector of image pattern in library and Europe;
The recall precision:The retrieval image uploaded to user is not of uniform size, and image rectification, retrieval vector extraction, often
A user images library sample size is not of uniform size, the problem that image procossing, similarity calculation time can be caused to grow, in proportion will figure
As being scaled 256*256 pixels, and pass through distributed type assemblies and the processing in the deconsolidation process procedural mode raising unit interval
Efficiency.
A kind of image similarity search method, described image retrieving include the following steps:
Step A, training retrieval model is broadly divided into and carries out image rectification and scaling processing to training sample before retrieval, solves
Image lacks and fuzzy problem;
Step B, then image information extraction algorithm extraction training sample image etc. data informations;
Step C, the data information of image is handled to solve image rotation and scaling using probability random cropping algorithm
Problem reduces image rotation and scales the error brought;
Step D, training image retrieval model in Non-surveillance clustering neural network is added in processed image data information;
Step E, image retrieval:Load fabric to be retrieved, is corrected processing and scaling processing to figure image to be retrieved, makes
The data information of image is extracted with image information extraction algorithm, watershed algorithm and random chance trimming algorithm;
Step F, retrieval vector is extracted by image encrypting algorithm, by the retrieval for calculating itself and all images in image library
A few moral distances in the normal form distance of vector and Europe obtain and the retrieval highest one group of image data of image similarity value.
Preferably, described image retrieval model training includes the following steps:First, training sample is loaded, then judges figure
As correction process, training sample is reloaded when there is missing after image procossing, there is no missings after image procossing
When enter extraction image data information and anti-rotational scaled error handle;Retrieval model training is carried out later and then is terminated.
The beneficial effects of the invention are as follows:The present invention is extracted the characteristic area information of image, is made using the method for retrieval vector
It is more accurate to obtain retrieval result;First, being made using unsupervised learning mode training image retrieval model to avoid artificially labelling
At subjective factor difference caused by retrieval precision reduce the problems such as, second is that on treatment effeciency every time retrieval take on 0.3 second left side
It is right;Third, open private clound Retrieval Interface, user can complete image retrieval by downloading retrieval model in private clound;Unsupervised
Habit mode training pattern:The similarity model training of the present invention extracts the gray scale of picture without labelling for training sample
The data informations such as figure, after random chance cuts scheduling algorithm processing, training image retrieves vector model, avoids artificial mark
Label lead to the excessively high problem of human cost of the differentia influence similarity calculation precision and more new model of artificial subjective factor, and
Retrieval is carried out in private clound provide possibility for user;The extraction and calculating for retrieving vector avoid two images obviously only
There are a fine differences such as color and the not high situation of similarity value, because retrieval vector contains the feature structure region of image,
More focus on the features such as pattern, the structure of image.
Description of the drawings
Fig. 1 is that retrieval model of the present invention trains flow chart;
Fig. 2 is the retrieval flow figure of the present invention;
Fig. 3 is the embodiment of the present invention retrieval result schematic diagram.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Please refer to Fig.1 shown in -2, a kind of image similarity search method, including picture identification, image correction process,
Image encrypting algorithm training, the extraction of image retrieval vector, image similarity calculates and recall precision realizes image similarity retrieval;
Wherein, described image correction process includes the following steps:
Step A, user's uploading pictures analyze the picture of upload;
If step B, there is missing, fuzzy region in the picture that user uploads or there are watermark picture, system roots
Certain adjustment is carried out according to normal region to missing, watermark or fuzzy region to sharpen, if having adjusted or having existed missing,
Allow user's uploading pictures again;
Described image retrieval model is trained:It is carried from training sample image using image information extraction algorithm and watershed algorithm
The data informations such as the RBG of image and Non-surveillance clustering neural network are taken, image encrypting algorithm training is completed;
Described image retrieves vector extraction:The extraction of retrieval vector, retrieval arrow are carried out to image according to image encrypting algorithm
Amount includes the characteristic area information of image;
Described image similarity calculation:Image to be retrieved calculates itself and figure after obtaining retrieval vector by image encrypting algorithm
As a few moral distances calculate similarity value in the normal form distance of the retrieval vector of image pattern in library and Europe;
The recall precision:The retrieval image uploaded to user is not of uniform size, and image rectification, retrieval vector extraction, often
A user images library sample size is not of uniform size, the problem that image procossing, similarity calculation time can be caused to grow, in proportion will figure
As being scaled 256*256 pixels, and pass through distributed type assemblies and the processing in the deconsolidation process procedural mode raising unit interval
Efficiency.
A kind of image similarity search method, described image retrieving include the following steps:
Step A, training retrieval model is broadly divided into and carries out image rectification and scaling processing to training sample before retrieval, solves
Image lacks and fuzzy problem;
Step B, then image information extraction algorithm extraction training sample image etc. data informations;
Step C, the data information of image is handled to solve image rotation and scaling using probability random cropping algorithm
Problem reduces image rotation and scales the error brought;
Step D, training image retrieval model in Non-surveillance clustering neural network is added in processed image data information;
Step E, image retrieval:Load fabric to be retrieved, is corrected processing and scaling processing to figure image to be retrieved, makes
The data information of image is extracted with image information extraction algorithm, watershed algorithm and random chance trimming algorithm;
Step F, retrieval vector is extracted by image encrypting algorithm, by the retrieval for calculating itself and all images in image library
A few moral distances in the normal form distance of vector and Europe obtain and the retrieval highest one group of image data of image similarity value.
Wherein, described image retrieval model training includes the following steps:First, training sample is loaded, then judges image
Correction process reloads training sample when there is missing after image procossing, and there is no missings after image procossing
When enter extraction image data information and anti-rotational scaled error handle;Retrieval model training is carried out later and then is terminated.
Embodiment
The first step:Excel the or scv tables for including the information such as image library image url are uploaded by front end interface;Second
Step:Picture in image library is handled by image encrypting algorithm;Third walks:User uploads the figure to be retrieved by interface
Piece.4th step:Obtain retrieval result.
Retrieval result is as shown in Figure 3:The first from left is to be intended to retrieving image, remaining is high with retrieving image similarity in image library
Picture.Because being retrieved using the picture in image library, it is same figure preceding two pictures occur.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie
In the case of without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power
Profit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent requirements of the claims
Variation is included within the present invention.Any reference signs in the claims should not be construed as limiting the involved claims.
In addition, it should be understood that although this specification is described in terms of embodiments, but not each embodiment is only wrapped
Containing an independent technical solution, this description of the specification is merely for the sake of clarity, and those skilled in the art should
It considers the specification as a whole, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art
The other embodiment being appreciated that.
Claims (3)
1. a kind of image similarity search method, it is characterised in that:Including picture identification, image correction process, image inspection
Rope model training, the extraction of image retrieval vector, image similarity calculates and recall precision realizes image similarity retrieval;
Wherein, described image correction process includes the following steps:
Step A, user's uploading pictures analyze the picture of upload;
If step B, the picture that user uploads has missing, fuzzy region or there are watermark picture, system is according to just
Normal region carries out certain adjustment to missing, watermark or fuzzy region and sharpens, and is lacked if adjust or having existed, and allows use
Family uploading pictures again;
Described image retrieval model is trained:It is extracted and is schemed from training sample image using image information extraction algorithm and watershed algorithm
The data informations such as the RBG of picture and Non-surveillance clustering neural network complete image encrypting algorithm training;
Described image retrieves vector extraction:The extraction for carrying out retrieval vector to image according to image encrypting algorithm, retrieves vector packet
Characteristic area information containing image;
Described image similarity calculation:Image to be retrieved calculates itself and image library after obtaining retrieval vector by image encrypting algorithm
A few moral distances calculate similarity value in the normal form distance of the retrieval vector of middle image pattern and Europe;
The recall precision:The retrieval image uploaded to user is not of uniform size, and image rectification, retrieval vector extraction, Mei Geyong
Family image library sample size is not of uniform size, and the problem that image procossing, similarity calculation time can be caused to grow in proportion contracts image
It puts as 256*256 pixels, and passes through distributed type assemblies and deconsolidation process procedural mode improves treatment effeciency in the unit interval.
2. a kind of image similarity search method according to claim 1, it is characterised in that:Described image retrieving packet
Include following steps:
Step A, training retrieval model is broadly divided into and carries out image rectification and scaling processing to training sample before retrieval, solves image
Missing and fuzzy problem;
Step B, then image information extraction algorithm extraction training sample image etc. data informations;
Step C, using probability random cropping algorithm come handle the data information of image with solve the problems, such as image rotation and scaling,
It reduces image rotation and scales the error brought;
Step D, training image retrieval model in Non-surveillance clustering neural network is added in processed image data information;
Step E, image retrieval:Load fabric to be retrieved, is corrected processing and scaling processing to figure image to be retrieved, uses figure
As information extraction algorithm, watershed algorithm and random chance trimming algorithm extract the data information of image;
Step F, retrieval vector is extracted by image encrypting algorithm, by the retrieval vector for calculating itself and all images in image library
Normal form distance and Europe in a few moral distances, obtain with retrieval the highest one group of image data of image similarity value.
3. a kind of image similarity search method according to claim 1, it is characterised in that:Described image retrieval model is instructed
White silk includes the following steps:First, training sample is loaded, then judges image correction process, there is missing after image procossing
When reload training sample, there is no enter extraction image data information and anti-rotational when missing after image procossing
Scaled error processing;Retrieval model training is carried out later and then is terminated.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810355429.0A CN108595596A (en) | 2018-04-19 | 2018-04-19 | A kind of image similarity search method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810355429.0A CN108595596A (en) | 2018-04-19 | 2018-04-19 | A kind of image similarity search method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108595596A true CN108595596A (en) | 2018-09-28 |
Family
ID=63611242
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810355429.0A Pending CN108595596A (en) | 2018-04-19 | 2018-04-19 | A kind of image similarity search method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108595596A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110148198A (en) * | 2019-05-10 | 2019-08-20 | 厦门欢乐逛科技股份有限公司 | Poster template recommended method, medium, equipment and device |
CN111709363A (en) * | 2020-06-16 | 2020-09-25 | 湘潭大学 | Chinese painting authenticity identification method based on rice paper grain feature identification |
US11189367B2 (en) * | 2018-05-31 | 2021-11-30 | Canon Medical Systems Corporation | Similarity determining apparatus and method |
CN115861320A (en) * | 2023-02-28 | 2023-03-28 | 天津中德应用技术大学 | Intelligent detection method for automobile part machining information |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104462494A (en) * | 2014-12-22 | 2015-03-25 | 武汉大学 | Remote sensing image retrieval method and system based on non-supervision characteristic learning |
CN104915676A (en) * | 2015-05-19 | 2015-09-16 | 西安电子科技大学 | Deep-level feature learning and watershed-based synthetic aperture radar (SAR) image classification method |
CN105956607A (en) * | 2016-04-22 | 2016-09-21 | 南京师范大学 | Improved hyperspectral image classifying method |
CN106777167A (en) * | 2016-12-21 | 2017-05-31 | 中国科学院上海高等研究院 | Magnanimity Face Image Retrieval System and search method based on Spark frameworks |
CN106980641A (en) * | 2017-02-09 | 2017-07-25 | 上海交通大学 | The quick picture retrieval system of unsupervised Hash and method based on convolutional neural networks |
CN107506800A (en) * | 2017-09-21 | 2017-12-22 | 深圳市唯特视科技有限公司 | It is a kind of based on unsupervised domain adapt to without label video face identification method |
-
2018
- 2018-04-19 CN CN201810355429.0A patent/CN108595596A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104462494A (en) * | 2014-12-22 | 2015-03-25 | 武汉大学 | Remote sensing image retrieval method and system based on non-supervision characteristic learning |
CN104915676A (en) * | 2015-05-19 | 2015-09-16 | 西安电子科技大学 | Deep-level feature learning and watershed-based synthetic aperture radar (SAR) image classification method |
CN105956607A (en) * | 2016-04-22 | 2016-09-21 | 南京师范大学 | Improved hyperspectral image classifying method |
CN106777167A (en) * | 2016-12-21 | 2017-05-31 | 中国科学院上海高等研究院 | Magnanimity Face Image Retrieval System and search method based on Spark frameworks |
CN106980641A (en) * | 2017-02-09 | 2017-07-25 | 上海交通大学 | The quick picture retrieval system of unsupervised Hash and method based on convolutional neural networks |
CN107506800A (en) * | 2017-09-21 | 2017-12-22 | 深圳市唯特视科技有限公司 | It is a kind of based on unsupervised domain adapt to without label video face identification method |
Non-Patent Citations (1)
Title |
---|
郭迎春 等: ""基于概率随机裁剪的图像缩放算法"", 《通信学报》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11189367B2 (en) * | 2018-05-31 | 2021-11-30 | Canon Medical Systems Corporation | Similarity determining apparatus and method |
CN110148198A (en) * | 2019-05-10 | 2019-08-20 | 厦门欢乐逛科技股份有限公司 | Poster template recommended method, medium, equipment and device |
CN111709363A (en) * | 2020-06-16 | 2020-09-25 | 湘潭大学 | Chinese painting authenticity identification method based on rice paper grain feature identification |
CN115861320A (en) * | 2023-02-28 | 2023-03-28 | 天津中德应用技术大学 | Intelligent detection method for automobile part machining information |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108595596A (en) | A kind of image similarity search method | |
CN106709917A (en) | Neural network model training method, device and system | |
CN107067043B (en) | Crop disease and insect pest detection method | |
CN108875821A (en) | The training method and device of disaggregated model, mobile terminal, readable storage medium storing program for executing | |
EP4035168A1 (en) | Predictive personalized three-dimensional body models | |
CN107239514A (en) | A kind of plants identification method and system based on convolutional neural networks | |
CN108287857A (en) | Expression picture recommends method and device | |
CN110163810A (en) | A kind of image processing method, device and terminal | |
CN113920453A (en) | Pig body size weight estimation method based on deep learning | |
CN108416353A (en) | Crop field spike of rice fast partition method based on the full convolutional neural networks of depth | |
CN109344738A (en) | The recognition methods of crop diseases and pest crop smothering and device | |
CN108074236A (en) | Irrigating plant based reminding method, device, equipment and storage medium | |
CN106776675A (en) | A kind of plants identification method and system based on mobile terminal | |
CN111062441A (en) | Scene classification method and device based on self-supervision mechanism and regional suggestion network | |
CN106991449B (en) | Method for identifying blueberry varieties in assistance of living scene reconstruction | |
CN109829071A (en) | Face image searching method, server, computer equipment and storage medium | |
CN112733958A (en) | Greenhouse ozone concentration control method and system | |
CN112132232A (en) | Medical image classification labeling method and system and server | |
Croft et al. | Fossil remains of a new, diminutive Bubalus (Artiodactyla: Bovidae: Bovini) from Cebu island, Philippines | |
CN116884571B (en) | Meal weight intelligent evaluation system based on image processing | |
CN109451292A (en) | Color temp bearing calibration and device | |
CN116156275B (en) | Meteorological information broadcasting method and system | |
CN110555379B (en) | Human face pleasure degree estimation method capable of dynamically adjusting features according to gender | |
CN109872058B (en) | Multimedia crowd sensing excitation method for machine learning system | |
CN109359675A (en) | Image processing method and equipment |
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: 20180928 |