CN109508623A - Item identification method and device based on image procossing - Google Patents
Item identification method and device based on image procossing Download PDFInfo
- Publication number
- CN109508623A CN109508623A CN201811010868.4A CN201811010868A CN109508623A CN 109508623 A CN109508623 A CN 109508623A CN 201811010868 A CN201811010868 A CN 201811010868A CN 109508623 A CN109508623 A CN 109508623A
- Authority
- CN
- China
- Prior art keywords
- article
- identified
- feature value
- image
- items
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention provides a kind of item identification method and device based on image procossing, comprising: obtain article to be identified and compare the image of article;It detects to compare article and article to be identified from image, and extracts the First Eigenvalue for comparing article, the Second Eigenvalue of article to be identified;It is preset first Standard Eigenvalue by one eigenvalue correction of comparison material position of extraction, determines the first correction function of the correction;It corrects to obtain third feature value according to Second Eigenvalue of first correction function to article to be identified;Third feature value is compared with the fourth feature value of images of items in database, article to be identified is identified according to comparison result.The accuracy and speed of article identification can be improved in the method for the present invention.
Description
Technical field
The present embodiments relate to image identification technical field more particularly to a kind of article identification sides based on image procossing
Method and device.
Background technique
In existing image recognition technology, input picture is analyzed in image recognition processes, is usually related only to
Feature extraction and Characteristic Contrast, such as characteristics of image is extracted from input picture, then by the spy of characteristics of image and known object
Sign is compared and is identified.
Text region, recognition of face, object identification are all the typical cases of image recognition technology and are used widely.With object
For identification technology, it can be compared by the characteristic value in the characteristic value and database of extraction object to be identified, to know
Object not to be identified.For example, the similar articles in the picture searching network platform for passing through article.
But erroneous judgement is often resulted in the prior art, cannot obtain accurate result.For example, inventors have found that causing to judge by accident
A main cause be: shoot focal length used by article to be identified and/or shooting distance and there is being not fixed property, will lead to same
One article obtains various sizes of image under different focal length or/and different shooting distances.It is asked of both bringing as a result,
Topic: on the one hand, for identical items, due to shooting focal length or apart from difference, it will cause and judged by accident into not because of picture size difference
Same article;On the other hand, the actual size biggish two pieces of difference or the above article for approximate there are surface characteristics, not
The size that may be shown under same focal length or shooting distance is almost approximate, thus misjudged at same or like article.
Existing article identification technology makes recognition accuracy lower.
Summary of the invention
The embodiment of the present invention is designed to provide a kind of item identification method and device based on image procossing, Neng Gouti
The accuracy of high article identification.
Technical solution used in the embodiment of the present invention is as follows:
One embodiment of the invention provides a kind of item identification method based on image procossing, comprising:
It obtains article to be identified and compares the image of article, wherein the scaling of images of items to be identified and article to be identified material object
Ratio with compare images of items and compare that the scaling of article material object is identical, the size for comparing article can be learned;From image
In detect to compare article and article to be identified, and extract the First Eigenvalue for comparing article, and extract article to be identified
Second Eigenvalue;The First Eigenvalue is corrected to the first Standard Eigenvalue, the first Standard Eigenvalue is that can obtain, and is determined the
One eigenvalue correction is the first correction function of the first Standard Eigenvalue;According to the first correction function to the second of article to be identified
Characteristic value is corrected, and obtains third feature value;By the fourth feature value of each images of items in third feature value and database into
Row compares, and identifies article to be identified according to comparison result, fourth feature value is obtained through correction in advance.
Another embodiment of the present invention provides a kind of item identification methods based on image procossing, wherein obtains the first standard
Characteristic value includes:
Obtain the image for comparing article;Make to compare images of items and compare article material object and contract with the first pre-set zoom ratio
It puts;Image feature value is as the first Standard Eigenvalue for comparing article after obtaining correction.
Another embodiment of the present invention provides a kind of item identification methods based on image procossing, wherein is corrected in advance
Into database, the fourth feature value of images of items includes:
Article is detected from the image of database and compares article, and extracts the fifth feature value for comparing article, extracts article
Sixth feature value;Fifth feature value is corrected, until comparing images of items after correction and comparing article material object is second default
Scaling determines the second correction function that fifth feature is worth to corresponding image rectification to the second pre-set zoom ratio;According to
Second correction function is corrected sixth feature value to obtain fourth feature value.
Another embodiment of the present invention provides a kind of item identification methods based on image procossing, wherein detects from image
Article and article to be identified are compared out, comprising:
Region division is carried out to image, the article characteristics to be identified and comparison article characteristics extracted using convolutional neural networks are distinguished
Classified to each region and is scored;Article to be identified is respectively obtained according to the classification of each region and scoring and compares article
Score, location information and size.
Another embodiment of the present invention provides a kind of item identification methods based on image procossing, by third feature value and data
The fourth feature value of each images of items stored in library is compared, and identifies article to be identified according to comparison result, comprising:
Each article figure stored in article and database to be identified is determined according to third feature value and fourth feature value comparison result
The similarity of picture;
Descending sort is carried out according to the similarity of article to be identified to article in database.
In the present embodiment, it is worth storing in article and database to be identified by comparing third feature value and third feature
Article similarity, and the article in database is subjected to descending arrangement according to the similarity of article to be identified, can will
The item pictures of similarity most are listed in foremost with article to be identified, user can very clear acquisition database neutralize to
Identify the highest article of article similarity.The picture of the higher article of other similarities can be also obtained simultaneously, be convenient for user
Further progress screening and judgement.
Another embodiment of the present invention provides a kind of item identification devices based on image procossing, comprising:
Image collection module, the image for obtaining article to be identified Yu comparing article, wherein image collection module, for obtaining
Take and article to be identified and compare the image of article, wherein the scaling of images of items to be identified and article to be identified material object and
It is identical with the scaling of article material object is compared to compare images of items, the size for comparing article can be learned;Detection module is used
In detecting to compare article and article to be identified from image;Characteristic extracting module, for extracting the fisrt feature for comparing article
Value, and extract the Second Eigenvalue of article to be identified;Correction module, for the First Eigenvalue to be corrected to the first standard feature
Value;And Second Eigenvalue is corrected to third feature value;Matching identification module, for will be in third feature value and database
The fourth feature value of each images of items is compared, and identifies article to be identified according to comparison result, images of items in database
Fourth feature value is that calibrated module corrects obtain in advance.
Another embodiment of the present invention provides a kind of item identification devices based on image procossing, wherein correction module is also used
It is zoomed in and out in the comparison images of items obtained to image collection module with the first preset ratio;Characteristic extracting module is also used to
The characteristic value for obtaining the comparison images of items that correction module is zoomed in and out with the first preset ratio, as the first mark for comparing article
Quasi- characteristic value.
Another embodiment of the present invention provides a kind of item identification devices based on image procossing, wherein detection module is also used
Article is detected in each image from database and compares article;Characteristic extracting module is also used to extract the 5th of comparison article
Characteristic value extracts the sixth feature value of article;Correction module is also used to be corrected fifth feature value, until described after correction
Comparing images of items and the article material object that compares is the second pre-set zoom ratio;And by the second correction function to the 6th
Characteristic value is corrected to obtain fourth feature value.
Another embodiment of the present invention provides a kind of item identification devices based on image procossing, wherein matching identification module,
Include:
Similarity determining unit, for determining article to be identified and data according to third feature value and fourth feature value comparison result
The similarity of each images of items stored in library;
Sequencing unit, for carrying out descending sort according to the similarity of article to be identified to article in database.
The technical solution of the embodiment of the present invention has the advantage that
In the embodiment of the present application, by the way that article image rectification will be compared to Standard Eigenvalue, again with identical bearing calibration school
Article just to be identified, by after correction article characteristics value to be identified and database in article characteristics value be compared, due to wait know
Other article and comparison images of items have identical scaling, and the size for comparing article can be learned, therefore logical
Crossing the embodiment of the present invention can determine the size of article to be identified, when comparing with the characteristic value of article each in database not only
Contoured article, which can be matched, can accurately also match the size of article, so that the precision of identification article be greatly improved.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is that the present invention is based on the one embodiment flow diagrams of item identification method of image procossing;
Fig. 2 is that the present invention is based on another embodiment flow diagrams of the item identification method of image procossing;
Fig. 3 is that the present invention is based on another embodiment flow diagrams of the item identification method of image procossing;
Fig. 4 is that the present invention is based on one example structure schematic diagrams of item identification devices of image procossing;
Fig. 5 is that the present invention is based on another example structure schematic diagrams of the item identification devices of image procossing;
Fig. 6 is that the application is based on one specific example flow chart of image procossing item identification method.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Present inventor has found during research and development, existing to carry out article identification often by shooting or video
Cause to judge by accident, reason is: the size that same article is shown in an image or a video under different focal length or shooting distance is
Different.The prior art can only detect article, but be unable to measure the actual size of object, to cause to judge by accident.
Present inventor uses a variety of methods in order to solve the above problem, for example, being examined by convolutional neural networks model
Article to be identified is surveyed, the object that will test out is zoomed in or out according to a variety of multiples, generates the different object set of several sizes, will
The different object set of size substitutes into computer deep learning model, is translated into condition code, by comparing condition code, calculates
The similarity degree of different objects out.
But these object sets can generate a large amount of calculating when substituting into computer deep learning model, waste huge calculating
Amount;Meanwhile actual size differs the similar article of larger but surface characteristics if it exists, still will be considered that by above-mentioned calculated result
Article similarity is higher, reduces the accuracy of judgement.
Fig. 1 is that the present invention is based on the one embodiment flow diagrams of item identification method of image procossing.According to Fig. 1,
The present invention is based on the item identification methods of image procossing, comprising:
Step 110, it obtains article to be identified and compares the image of article, wherein images of items to be identified and article to be identified are real
The scaling of object with compare images of items and compare that the scaling of article material object is identical, the size for comparing article can be obtained
Know;
In this step, it obtains article to be identified and compares the image of article, wherein images of items to be identified and article to be identified
Scaling in kind is identical as comparing images of items and comparing the scaling of article material object.
It is understood that in this application, the scaling between image and material object refers to that picture size and material object are real
Proportionate relationship between the size of border.Either amplify, or reduces, the ratio which zooms in or out in length and width
It is consistent, that is, only change picture size and keep the ratio of original image on image Aspect Ratio.Wherein, picture size is
The size of image, according to the prior art, the length and width of picture size be can be as unit of pixel, are also possible to length
Spend unit, such as centimetre be unit.When picture size is using pixel as unit, need that resolution ratio is combined to determine picture size, this
It is related to the distance between picture element and point to punish resolution.
It is not necessary it is worth noting that possessing identical scaling, as long as the scaling of the two is can be with
Know, when the scaling of the two is inconsistent, ratio is adjusted to identical by scaling, facilitates and compares.In the application
In embodiment, for convenience of explanation, by taking scaling is identical as an example.
It is understood that for the same object, subject all can lead to imaging scaling ratio not away from, focal length difference
Together.For example, for the same object, in the identical situation of focal length, the size of images of items becomes with the distance of shooting distance
Change can generate variation.The more remote then images of items of shooting distance is smaller, and shooting distance is closer, and images of items is bigger.It is identical in object distance
In the case of, different shooting focal lengths also influence the size of images of items, and the smaller then images of items of focal length is smaller, and the more big then object of focal length
Product image is bigger.
It obtains the identical article to be identified of scaling and compares the image of article, the various of the prior art can be passed through
Mode, the embodiment of the present invention are not particularly limited.For example, it may be imaging is shot under identical object distance and focal length respectively;?
Can be respectively different with object distance and focal length, but it is capable of forming identical scaling;It is also possible to primary shooting imaging;Or
It is shooting imaging twice under identical shooting condition, specific shooting condition can limit by the prior art, may include but not
Be limited to: shooting distance and shooting focal length, furthermore shooting condition can also further include shooting angle, light it is reflective and negative
Shadow etc..Wherein, shooting distance and shooting focal length can influence imaging scaling ratio, and the anti-light and shade of shooting angle, light
The scaling of imaging is had no effect on, different shooting angle can cause different trapezoidal distortions, reflective that light and shade can be caused uneven
Deng.Identical shooting condition herein may include identical, or in the error range of permission.
In this step, obtains article to be identified and can be with the image for comparing article and obtained respectively from different images,
It is also possible to obtain on the same image, the application limits not to this.
Step 120, it detects to compare article and article to be identified from image, and extracts the fisrt feature for comparing article
Value, and extract the Second Eigenvalue of article to be identified;
In this step, detect that comparing article and article to be identified, the detection can adopt in the image obtained in step 110
It is carried out with existing method, for example, being detected by convolutional neural networks or convolutional neural networks in conjunction with full Connection Neural Network
Comparison article and article to be identified in image.That is, detecting two articles with the prior art, and two article figures are extracted respectively
The characteristic value of picture extracts the First Eigenvalue for comparing article, and extracts the Second Eigenvalue of article to be identified.It is understood that
It is that characteristic value herein is specifically as follows condition code.For example, picture is input into convolutional neural networks, obtains characteristic pattern and then obtain
To condition code.The prior art can also be used in the specific method for extracting article characteristic value, and the application is without limitation.
Step 130, the First Eigenvalue is corrected to the first Standard Eigenvalue, the first Standard Eigenvalue can obtain, and determine
The First Eigenvalue is corrected to the first correction function of the first Standard Eigenvalue;
The comparison material product the First Eigenvalue extracted in step 120 is corrected, preset first standard feature is corrected to
Value, which is preset known quantity.Since comparison material product the First Eigenvalue is to mention in the step 120
It obtains, preset first Standard Eigenvalue is known quantity, passes through known the First Eigenvalue and preset known first
Standard Eigenvalue can determine the first correction function.
In the embodiment of the present invention, the first Standard Eigenvalue is pre-set, the comparison article of corresponding pre-set zoom ratio
Image.Specifically, the first Standard Eigenvalue can be the image for comparing that article is shot under specific shooting distance and focal length
Characteristic value;Alternatively, first Standard Eigenvalue is also possible to shoot the comparison article under nonspecific shooting distance and focal length
Image is zoomed in and out later, so that reach pre-set zoom ratio between image and article actual size after scaling, the present invention
To this without limiting.
In the embodiment of the present application, preset scaling can be 1:1, that is, the size and reality of article are compared in image
It is identical that border compares item sizes.It is also possible into other certain ratios, for example, 2:1, the size that article is compared in image is
Practical twice etc. for comparing item sizes.In the embodiment of the present application, only require that scaling is can to know, not
Specific restriction scaling, it is e.g. preset.It is understood that first Standard Eigenvalue can be and be stored in advance
, it is also possible to obtain when needed.
In this step, the First Eigenvalue that article is compared in step 120 is corrected by the first correction function
To preset first Standard Eigenvalue, wherein due to comparing the First Eigenvalue of article and the first mark of the comparison article
Quasi- characteristic value can all know that the first correction function can be by comparing above-mentioned the first Standard Eigenvalue and fisrt feature
It is worth.
Step 140, it is corrected according to Second Eigenvalue of first correction function to article to be identified, obtains third feature
Value;
In this step, the identical correction function correction of images of items images of items to be identified is compared using with correction, treats knowledge
Other images of items carries out the processing of the scaling of same ratio, make images of items to be identified have with compare images of items have it is identical
Scaling.Specifically, being corrected to obtain third spy using Second Eigenvalue of first correction function to article to be identified
Value indicative.For the third feature value is with respect to Second Eigenvalue, the scaling of corresponding image and the first Standard Eigenvalue and
The First Eigenvalue is consistent the scaling of image.
For example, when the First Eigenvalue for comparing article is corrected to the first Standard Eigenvalue by the first correction function, so that school
Picture size is identical with article full size is compared after just, then first correction function is to article Second Eigenvalue to be identified
After being corrected, obtained third feature be worth the size of corresponding images of items to be identified also with the practical ruler of article to be identified
It is very little consistent.
Step 150, third feature value is compared with the fourth feature value of images of items each in database, according to comparison
As a result article to be identified is identified, fourth feature value is obtained through correction in advance.
In this step by each article figure in the third feature value of article to be identified corrected in step 140 and database
The fourth feature value of picture is compared.It is understood that include the image of at least one article in database, this step into
When row compares, it can be and be compared third feature value with the fourth feature value of images of items each in database respectively, directly
To identifying all items in article to be identified, or completeer database in the article of database.For convenience of description, at this
In inventive embodiments, only with the fourth feature value ratio of some images of items in the third feature value and database of article to be identified
Relatively it is illustrated.
In embodiments of the present invention, have between the corresponding images of items to be identified of third feature value and article to be identified material object
Known scaling relationship, and images of items is also image and actual object by correcting in advance, after the correction in database
Between also have known scaling relationship.Pass through above-mentioned two corrected figure for having known scaling relationship between actual object
The comparison of the characteristic value of picture needs not move through scaling or other adjustment, can judge article two in article and database to be identified
Whether person's shape, figure are consistent, and accurately judge whether the two size is consistent, so as to combine shape, pattern and size
Article to be identified is accurately found out in the database.
For example, when corrected third feature is worth item sizes to be identified and practical article to be identified in corresponding image
It is identical, and in database the fourth feature of images of items be worth corresponding image size it is also identical as the article actual size, by
All it is big with respective material object etc. in the corresponding picture size of two characteristic values, therefore reduces because being scaled between image and material object
It is judged by accident caused by ratio difference.
Moreover, not needing in order to which the article in article to be identified and database is carried out relatively respective same ratio in kind yet
Scaling and to characteristic value carry out data processing.For example, it may be desirable to which the First Eigenvalue to article to be identified carries out a variety of times
Several amplifications and diminution, then be compared with the fourth feature value of article in database;Alternatively, respectively to the of article to be identified
One characteristic value carries out the amplification and diminution of a variety of multiples, and, the fourth feature value of article zooms in and out processing in database,
The two is compared again.These modes both increase the complexity of calculation amount and calculating, and reduce the accurate of judgement
Property.
Certainly, in the embodiment of the present application, third feature value and fourth feature value are not limited to so that after correcting in image
Article is identical with full size, is also possible to other preset ratios.For example, pantograph ratio of the third feature value to article to be identified
Example is consistent with scaling of the fourth feature value to article in database, but the scaling is not 1:1, in this case, by
It is consistent in two characteristic value scalings, the size of article to be identified can also be accurately judged in movement images.
Further, third feature value is to the scaling and fourth feature value of article to be identified to article in database
Scaling can also be inconsistent.Since scaling both in this application can be known, when the two scales
When ratio is inconsistent, image can be zoomed in and out in practical applications, until making the corresponding article figure to be identified of third feature value
As the scaling of size and practical item sizes to be identified, in database corresponding with fourth feature value images of items size and
The scaling of actual object size is consistent;Alternatively, before comparison in characteristic value according to pre-set different proportion into
Row correction causes the two scaling to be unanimously compared again.There are many modes for specific value, can reach same effect
Fruit, in this regard, not applying also without limitation.
In the present embodiment, article is compared by introducing, correction compares article image feature value to the first Standard Eigenvalue
So that comparing the image of article after correction and comparing between article actual size has determining scaling relationship, utilization is identical
Correction function is corrected images of items to be identified, so that also there is identical contracting between images of items to be identified and actual object
Relationship is put, the characteristic value after correction is compared with the fourth feature value of article in database, due to the fourth feature value
It is corrected, between article also with determination scaling relationships, therefore, by comparing third feature value and the 4th
Characteristic value can not only identify the pattern form of article to be identified, additionally it is possible to the size for obtaining identification article, in conjunction with more to be identified
Item sizes in article and database picture can more accurately identify article to be identified.
Further, in embodiments of the present invention, comparing article is the article that shape, pattern and size have standard.
In the embodiment of the present application, comparison of the reference substance as article to be identified can be used, due to scaling phase
Together, as long as can know the size of the reference substance in advance, so that it may determine the size of article to be identified by comparing.Work as reference substance
When for standard-sized article, then size, the even characteristics such as shape pattern that can more easily know reference substance.Example
Such as, reference substance can be coin, always be fixed due to the coin dimensions shape and pattern of particular denomination, in practical application
In can accurately know the information of the reference substance.Reference substance can also be bottle cap, credit card, identity card, various models hand
Machine etc..For example, bank card or credit card, size is unified, for another example for the mobile phone of certain model, size and
Shape is also unified and can obtain.In selection criteria object, the comparison with single rule shape, example can choose
Circle, square such as standard, rectangle, triangle etc.;Also it can choose the comparison formed by single rule combination of shapes;
Or it is also possible to the comparison of irregular shape.
It is worth noting that comparing article to have fixed standard for shape, pattern and size is a kind of choosing reality of the application
Example is applied, in practical applications, reference substance, which can be, arbitrarily determines article, as long as can know and determine its size and shape i.e.
Can, reference substance is only to ensure that can know the scaling between third feature value and article full-size(d) to be identified.
In the embodiment of the present application, article to be identified can be with various articles, such as can be flat objects, can also be three-dimensional
Object.When article to be identified is flat objects, and when only needing to identify one side therein, then identifies this face, work as flat objects
Two sides require identification when, then two sides can be identified respectively respectively.It, can be with when article to be identified is three-dimensional object
Each face of the three-dimensional object is identified respectively, for example, six faces of object can be identified respectively, it can also be according to need
It wants, only the face of needs is identified.Any article to be identified can be identified using the application recognition methods.Specifically
Identification method may refer to identification method in the embodiment of the present application.
Since article to be identified may be from each data platform or from each different user, in the present embodiment
In, contoured article, pattern and size have the reference substance of fixed standard to be easier to obtain, and the application method can be made to realize
It is more convenient to come.
Fig. 2 is that the present invention is based on another embodiment flow diagrams of the item identification method of image procossing.Institute according to fig. 2
Show, the present invention is based on the item identification methods of image procossing, comprising:
Step 210, it obtains and compares images of items;
In this step, the image for comparing article can be obtained by existing manner.
Step 220, image is corrected, makes to compare images of items and compares article material object with the first pre-set zoom ratio
It zooms in and out;
In this step, the image of the comparison article obtained in step 210 is zoomed in and out, makes to compare images of items and comparison material
Product material object is zoomed in and out with the first pre-set zoom ratio.Wherein, the first pre-set zoom ratio can arbitrarily be set, such as: 1:1,
Perhaps the ratio of 2:1 perhaps 1:2 etc. or other non-integers, is not defined the embodiment of the present application, can realize
Goal of the invention.It is interest of clarity in embodiment later, will be illustrated as an example using the scaling of 1:1, as
Other scalings then carry out corresponding operation to image feature value, and this will not be repeated here.
Specifically, using 1:1 scaling when, obtained comparison article picture size and compare article after scaling
Full size size is identical.By compare article be side length for the square standard article of 0.5cm for, it is assumed that in step 210
The comparison images of items of acquisition is 100 pixels, and each pixel represents 0.1cm, then the size of image is 1*1cm2, for by image
, as 0.5*0.5 cm2 size big with material object etc. is narrowed down to, two ways: mode one can use, is passed through and is reduced pixel number and contract
The size of small image, for example, article image down will be compared to 25 pixels, each pixel represents 0.1cm, after diminution
Obtained comparison images of items is 0.5*0.5 cm2 size;Mode two, keep the pixel number of image it is constant, still for 100 pictures
Element, each pixel represent 0.05cm, image size reduction to 0.5*0.5 cm2.Two kinds of diminution modes can obtain and reality
The identical picture size of object size, can be different using the resolution ratio of the image of different zoom mode.Specifically using which kind of mode into
Row zooms in or out, and can be determined according to actual demand, the application does not repeat them here.
Step 230, image feature value is as the first Standard Eigenvalue for comparing article after obtaining correction.
The characteristic value of image after correcting in extraction step 220, the specific method for extracting characteristic value can use the prior art
It carries out, the application is defined not to this.In step 220, zooming in and out image to image and full size ratio is 1:1,
The corresponding size for comparing images of items of first Standard Eigenvalue is the actual size of the comparison article.
Step 240, it obtains article to be identified and compares the image of article, wherein images of items to be identified and object to be identified
The scaling of product material object with compare images of items and compare that the scaling of article material object is identical, compare the size energy quilt of article
Know;
Article to be identified and the specific steps for the image for comparing article are obtained in the step may refer to the corresponding embodiment of Fig. 1
Step 110.
Step 250, it detects to compare article and article to be identified from image, and extracts the fisrt feature for comparing article
Value, and extract the Second Eigenvalue of article to be identified;
It detects to compare article and article to be identified from image in the step, and extracts the First Eigenvalue for comparing article, with
And the specific steps of the Second Eigenvalue of extraction article to be identified may refer to the step 120 of the corresponding embodiment of Fig. 1.
Step 260, the First Eigenvalue is corrected to the first Standard Eigenvalue, the first Standard Eigenvalue can be known, determine
The First Eigenvalue is corrected to the first correction function of the first Standard Eigenvalue;
Passing through the first correction function in the step for one eigenvalue correction of comparison material position of extraction is the first mark that can be known
The specific steps of quasi- characteristic value may refer to the step 130 of the corresponding embodiment of Fig. 1.It is understood that in the present embodiment
The corresponding size for comparing images of items of first Standard Eigenvalue is the actual size of the comparison article.
Step 270, it is corrected according to Second Eigenvalue of first correction function to article to be identified, obtains third feature
Value;
It is corrected to obtain third feature value by Second Eigenvalue of first correction function to article to be identified in the step
Specific steps may refer to the step 140 of the corresponding embodiment of Fig. 1.Due to obtained comparison article corrected in step 260
Image is identical with item sizes are actually compared image, therefore, by obtaining in step 240 and comparing article with identical contracting
The image of the article to be identified of ratio is put after the correction of the first correction function, the corresponding image of obtained characteristic value be also with
The identical image of article actual size to be identified.
Step 280, third feature value is compared with the fourth feature value of images of items each in database, according to comparison
As a result article to be identified is identified, fourth feature value is obtained through correction in advance.
Third feature value is compared with the fourth feature value of images of items in database in the step, is tied according to comparing
Fruit identifies article to be identified, and the fourth feature value of images of items is to be corrected in advance in database.Specific steps can be with
Referring to the step 150 of the corresponding embodiment of Fig. 1.It is that corresponding image is and article actual size phase to be identified with third feature value
Together, i.e., for scaling is 1:1, since fourth feature value is also by correcting in advance herein, it is assumed that fourth feature
It is identical as the article actual size in database to be worth images of items size in corresponding database, is also the scaling of 1:1, this
When by comparing third feature value and fourth feature value, then both can accurately judge whether to be same article.
As other implementations, fourth feature is worth corresponding image and may not be 1:1 to the scaling of article,
Image can then be zoomed in and out at this time again and be corrected to 1:1, the image of article to be identified can also zoomed in and out, will be scaled
Ratio is adjusted to consistent with images of items scaling in database, and the application is to this without limiting.To the scaling school of picture
It is just being referred to existing way realization, the embodiment of the present application is to illustrate briefly, to be illustrated with the scaling of 1:1.
In the embodiment of the present application, the first Standard Eigenvalue that article is compared by presetting, determines this feature value pair
Scaling of the image answered to article, it is easier to determine the size of article to be identified.Also, preset the first standard spy
Sign, can enable has image to be compared at once when needing to identify, improves the efficiency of device operation.
Fig. 3 is that the present invention is based on another embodiment flow diagrams of the item identification method of image procossing.According to Fig. 3 institute
Show, the present invention is based on the item identification methods of image procossing, comprising:
Step 310, article is detected from the image of database and compares article, and extracts the fifth feature value for comparing article, is mentioned
Take the sixth feature value of article;
In this step, article is detected from the image of database and compares article, concrete implementation can use existing detection
Mode.The fifth feature value for comparing article and the sixth feature value for extracting article can also be extracted using existing way simultaneously.
In the present embodiment, store same original image in the database comprising a certain images of items in database and
Images of items is compared, alternatively, a certain article and comparison article are also possible to be imaged respectively in database, as long as image and material object
Between possess identical scaling.It is not necessary it is worth noting that possessing identical scaling, as long as the two
Scaling be that can know, when the scaling of the two is inconsistent, ratio is adjusted to identical by scaling, side
Just compare.In the embodiment of the present application, for convenience of explanation, by taking scaling is identical as an example.
Step 320, fifth feature value is corrected, until comparing images of items after correction and with article full size is compared being
Second pre-set zoom ratio determines the second correction letter that fifth feature is worth to corresponding image rectification to the second pre-set zoom ratio
Number;
In this step, fifth feature value is corrected, until comparing images of items after correction and comparing article material object is second
Pre-set zoom ratio.In this step, specific scaling correction is referred to step 220 progress corresponding to Fig. 2.In the present embodiment
In, due to the second pre-set zoom ratio it is known that therefore, the second correction function is confirmable.
Step 330, sixth feature value is corrected according to the second correction function to obtain fourth feature value.
Sixth feature value is corrected to obtain fourth feature value by above-mentioned second correction function, so as to database diagram
Images of items in piece is corrected to scaling identical with comparison material product in database picture.
Step 310 in this embodiment to step 330 can be similar to the corresponding embodiment of Fig. 1 in step 110 to
Step 140, similarly, also compared herein using reference substance, and by a similar method, unlike, it is logarithm herein
It is corrected according to the characteristic value of the article in library.
With step 320 to comparison article corrected scaling 1:1, logarithm in step 330 in database images
According to the corrected scaling 1:1 of article in the image of library.In this case, fourth feature is worth article figure in corresponding database
The size of picture is the size of the actual article.In the subsequent article third feature value to be identified and the 4th for obtaining correction
It is to zoom to images of items to be identified consistent with article actual size to be identified to be with third feature value when characteristic value compares
Example, third feature value and fourth feature value at this time be theoretically identical or the error range in permission in, therefore can essence
True identification article to be identified.
Step 340, it obtains and compares images of items;
The specific steps that comparison images of items is obtained in the step can be found in the step 210 of the corresponding embodiment of Fig. 2.
Step 350, image is corrected, makes to compare images of items and compares article material object with the first pre-set zoom ratio
It zooms in and out;
Image is zoomed in and out in the step, makes to compare images of items and compare article material object to carry out with the first pre-set zoom ratio
Scaling, specific steps can be found in the step 220 of the corresponding embodiment of Fig. 2.
Step 360, image feature value is as the first Standard Eigenvalue for comparing article after obtaining correction.
Image feature value can as the specific steps for the first Standard Eigenvalue for comparing article after obtaining correction in the step
The step 230 of corresponding embodiment referring to fig. 2.
Step 370, it obtains article to be identified and compares the image of article, wherein images of items to be identified and object to be identified
The scaling of product material object with compare images of items and compare that the scaling of article material object is identical, compare the size energy quilt of article
Know;
Article to be identified and the specific steps for the image for comparing article are obtained in the step may refer to the corresponding embodiment of Fig. 1
Step 110.
Step 380, it detects to compare article and article to be identified from image, and extracts the fisrt feature for comparing article
Value, and extract the Second Eigenvalue of article to be identified;
It detects to compare article and article to be identified from image in the step, and extracts the First Eigenvalue for comparing article, with
And the specific steps of the Second Eigenvalue of extraction article to be identified may refer to the step 120 of the corresponding embodiment of Fig. 1.
Step 390, the First Eigenvalue is corrected to the first Standard Eigenvalue, the first Standard Eigenvalue is it is known that determining will
The First Eigenvalue is corrected to the first correction function of the first Standard Eigenvalue;
It is by the first correction function that one eigenvalue correction of comparison material position of extraction is special for preset first standard in the step
The specific steps of value indicative may refer to the step 130 of the corresponding embodiment of Fig. 1.
Step 400, it is corrected according to Second Eigenvalue of first correction function to article to be identified, obtains third feature
Value;
It is corrected to obtain third feature value by Second Eigenvalue of first correction function to article to be identified in the step
Specific steps may refer to the step 140 of the corresponding embodiment of Fig. 1.
Step 410, third feature value is compared with the fourth feature value of images of items each in database, according to comparison
As a result article to be identified is identified, fourth feature value is obtained through correction in advance.
Third feature value is compared with the fourth feature value of images of items in database in the step, is tied according to comparing
Fruit identifies article to be identified, and the fourth feature value of images of items is that the specific steps through correcting in advance can be in database
Referring to the step 150 of the corresponding embodiment of Fig. 1.
In the present embodiment, by images of items in database also by with compare images of items and carry out identical scaling
Correction to reach pre-set dimension, be compared with article to be identified using the identical article that compares, due to article to be identified and
Article is all compared with the identical article that compares in database, and comparison article has certain size, shape and pattern, subtracts
Lack error caused by due to acquiring different comparison article characteristic values, further improves accuracy of identification.
Further, the second pre-set zoom ratio and the first pre-set zoom ratio are identical in the present embodiment.
When the second pre-set zoom ratio and identical the first pre-set zoom ratio, after being corrected twice to comparison article
To image in comparison item sizes be it is identical, passed through respectively due to the images of items in article to be identified and database and
Comparison material condition with scaling correction, therefore, contracting of the images of items Jing Guo same ratio in article and database to be identified
It puts.It can be reduced in this way because different scalings causes with the image identified in article and database article to article to be identified
Error, improve the accuracy of identification.
Further, image is corrected in the present embodiment, comparison article is made to zoom to the first pre-set zoom ratio,
Restoration correction specially is carried out to comparison article.
Fifth feature value is corrected by the second correction function, until comparing images of items to the second pre-set zoom ratio
Example specifically includes and carries out restoration correction to comparison article.
It is understood that restoration correction can be and zoom in and out to picture size into image in the embodiment of the present application
Item sizes are identical with shooting article actual size, eliminate the trapezoidal distortion of image, eliminate anti-light and shade etc..
In embodiments of the present invention, except the scaling processing for carrying out same ratio, other corrections of progress that can also be additional,
For example, eliminating the trapezoidal distortion of image, bright and dark light and reflective etc..
It is understood that the reflective equal shootings factor of trapezoidal distortion and light and shade also will affect into other than scaling
As to influence image feature value.In this application, scaling is necessary correction parameter when being corrected to image, trapezoidal
Skew control and bright and dark light and reflective it can be used as further correction parameter.Different images are decreased because there are different
Error caused by trapezoidal distortion is different with light condition.Specific restoration correction can be carried out using the prior art, the present invention
Embodiment is to this without limiting.
Image is subjected to restoration school, eliminates or reduce the trapezoidal distortion because caused by shooting angle is different, and/or, disappear
The problem of image light and shade unevenness caused by removing or reduce because of light problem, the image after can making correction is to greatest extent close to true
Product in kind improve the accuracy rate that image compares, further increase the speed and order of accuarcy of identification.
Further, it detects to compare article and article to be identified from image in the present embodiment, comprising:
Region division is carried out to image, the article characteristics to be identified and comparison article characteristics extracted using convolutional neural networks are distinguished
Classified to each region and is scored;
For example, the image to input carries out region division, multiple alternative areas are generated, the feature extracted using convolutional neural networks
Value, goes to compare multiple alternative area, each alternative area is classified and scored, the feature vector of data regular length.
Respectively obtain article to be identified according to the classification of each region and scoring and compare the score of article, location information and
Size.It is understood that score herein refers to that general calculated with scoring function finally obtains in convolutional neural networks
Each classification score, defined by the prior art, in this application without limit.
Specifically, region that can be high to score carries out classification judgement and position dimension calculates, finds and compare article and ratio
To the position of article.
Detect that article can be carried out according to the prior art from image, the present invention is defined not to this, for example, can
With the following steps are included:
Multiple candidate frames are determined in the picture;Convolutional neural networks are input into whole picture, obtain characteristic pattern (feature
Map);Mapping weight (patch) of each candidate frame on characteristic pattern is found, which is that training obtains, random initial
Change, adjust in the training process, is input to spatial pyramid pond layer for this mapping weight as the convolution feature of each candidate frame
(SPP layer, full name spatial pyramid pooling layer) and layer later;To the spy extracted in candidate frame
Sign, discriminates whether to belong to a certain kinds using classifier;For belonging to the candidate frame of a certain feature, further adjusted with device is returned
Its whole position.
In the present embodiment, region division is carried out to input picture, it is specific to determine the position for comparing article, it can be to each
Region parallel computation improves computational efficiency.
Further, the scaling and comparison material of images of items to be identified and article to be identified material object in the present embodiment
Product image is identical with the scaling of article material object is compared, the specific can be that:
Two scalings are all 1:1 perhaps 2:1 perhaps 1:2 or other ratios, as long as keeping two scalings identical
?.
In the embodiment of the present application when scaling is identical, does not need to carry out additional correction again, can directly compare
Figure and size.
Further, in the embodiment of the present application, by the of each images of items stored in third feature value and database
Four characteristic values are compared, and identify article to be identified according to comparison result, including
Article to be identified and storage each article figure in the database are determined according to third feature value and fourth feature value comparison result
The similarity of picture;
Wherein, article to be identified and storage each article in the database are determined by comparing third feature value and fourth feature value
The similarity of image can be realized by any prior art.
The similarity of database article and images of items to be identified is ranked up from high in the end.
In the present embodiment, similarity determining unit, which compares third feature value and fourth feature, is worth article sum number to be identified
According to the similarity of the article stored in library, sequencing unit carries out the article in database according to the similarity of article to be identified
Descending arrangement.The item pictures of similarity most can will be listed in foremost with article to be identified by the present embodiment, and user can
In very clear acquisition database and the highest article of article similarity to be identified.It is higher that other similarities can also be obtained simultaneously
Article picture, convenient for user's further progress screen and judge.
Fig. 4 is that the present invention is based on one example structure schematic diagrams of item identification devices of image procossing.According to Fig.4,
The present invention is based on the item identification devices of image procossing, comprising: image collection module 510, detection module 520, feature extraction mould
Block 530, correction module 540, matching identification module 550, wherein
Image collection module 510, the image for obtaining article to be identified Yu comparing article, wherein images of items to be identified and
The scaling of article material object to be identified is identical as comparing images of items and comparing the scaling of article material object, compares article
Size can be learned;
The operation that image collection module 510 is carried out may refer to the step 110 of the corresponding embodiment of the method for Fig. 1, the acquisition mould
Block 510 can be image collection module in the prior art.It is worth noting that acquired images of items to be identified and material object
Scaling identical can also be different with the scaling for comparing images of items and material object.It can lead to when the two ratio difference
Overcorrect module 540 adjusts scaling to identical, is further continued for handling.The embodiment of the present application for convenience of explanation, is contracted with the two
Put ratio it is identical for be illustrated, be not used as limiting.
Detection module 520 compares article and article to be identified for detecting from image;
Detection module 520 detects comparison article and article to be identified respectively from the image that image collection module 510 obtains, should
Device used by detecting or module, can be implemented with the prior art.
Characteristic extracting module 530, for extracting the First Eigenvalue for comparing article, and the second of extraction article to be identified
Characteristic value;
Detection module 520 can detect two articles in image respectively with the prior art, and characteristic extracting module 530 then mentions respectively
The characteristic value of two images of items is taken, extracts the First Eigenvalue for comparing article, and extract the second feature of article to be identified
Value.It is understood that characteristic value herein is specifically as follows condition code.Existing skill can also be used in specific article characteristic value of extracting
Art, the application is without limitation.
Correction module 540, for the First Eigenvalue to be corrected to the first Standard Eigenvalue;And by Second Eigenvalue school
It is just third feature value;
The comparison material product the First Eigenvalue that characteristic extracting module 530 is extracted is corrected by correction module 540, is corrected to pre-
If the first Standard Eigenvalue, which is available, for example, be preset known quantity.It can
With understanding, since preset first Standard Eigenvalue is available, such as known quantity, pass through known fisrt feature
Value and preset known first Standard Eigenvalue can determine the first correction function.
In addition, the Second Eigenvalue that characteristic extracting module 530 is extracted is corrected to third feature value by correction module 540, tool
For body, the Second Eigenvalue of article to be identified is corrected to third feature value with the first correction function by correction module 540.
Matching identification module 550, for carrying out the fourth feature value of each images of items in third feature value and database
It compares, article to be identified is identified according to comparison result, the fourth feature value of images of items is that calibrated module is preparatory in database
What correction obtained.
Correction module 540 is corrected each images of items in the third feature value obtained and database by matching identification module 550
Fourth feature value be compared, judge the corresponding article to be identified of third feature value and fourth feature value according to the result of the comparison
Whether the article in corresponding database is the same article.
It is understood that including the image of at least one article in database, matching identification module 550 is compared
When identification, it can be and be compared third feature value with the fourth feature value of images of items each in database respectively, until
All items in article to be identified, or completeer database are identified in the article of database.For convenience of description, in this hair
In bright embodiment, only compared with the fourth feature value of some images of items in the third feature value and database of article to be identified
It is illustrated.
In the present embodiment, article is compared by introducing, calibration standard object image characteristic value to the first Standard Eigenvalue makes
The image after article corrects, which must be compared, and compared between article actual size determining scaling relationship, utilizes identical school
Positive function is corrected images of items to be identified, so that also there is identical scaling between images of items to be identified and actual object
Characteristic value after correction is compared by relationship with the characteristic value of article in database, since the fourth feature value is also process
Correction, the database images of items after correction and the scaling relationship also between article with determination, therefore, by comparing third spy
Value indicative and fourth feature value, can not only identify the pattern form of article to be identified, additionally it is possible to the size for obtaining identification article, in conjunction with
Item sizes in article and database picture more to be identified can more accurately identify article to be identified.
In another embodiment of item identification devices the present invention is based on image procossing, the present invention is based on the objects of image procossing
Product identification device includes: image collection module 510, detection module 520, characteristic extracting module 530, correction module 540, compares
Identification module 550, wherein other than module corresponding in embodiment corresponding with Fig. 4 executes corresponding operation, correction module
540 and characteristic extracting module 430 also execute following operation:
Correction module 540, be also used to image collection module 510 obtain comparison article graph image with the first preset ratio into
Row scaling;
Correction module 540 zooms in and out the image for the comparison article that image collection module 510 obtains, make compare images of items with
Article material object is compared to zoom in and out with the first pre-set zoom ratio.Wherein, the first pre-set zoom ratio can arbitrarily be set, example
Such as: the ratio of 1:1 perhaps 2:1 perhaps 1:2 etc. or other non-integers is not defined the embodiment of the present application, all may be used
To realize goal of the invention.It is interest of clarity in embodiment later, will be said as an example using the scaling of 1:1
It is bright, as other scalings, then corresponding operation is carried out to image feature value, this will not be repeated here.
Characteristic extracting module 530 is also used to obtain the comparison article that correction module 540 is zoomed in and out with the first preset ratio
The characteristic value of image, as the first Standard Eigenvalue for comparing article.
Characteristic extracting module 530 extracts the characteristic value of image after correction module 540 corrects, the specific method for extracting characteristic value
It can be carried out using the prior art, the application is defined not to this.To zoom in and out image to image and full size ratio
For example is 1:1, the comparison item sizes of the first Standard Eigenvalue correspondence image are the actual size of the comparison article.
In the embodiment of the present application, the first Standard Eigenvalue that article is compared by presetting, determines this feature value pair
Scaling of the image answered to article, it is easier to determine the size of article to be identified.Also, preset the first standard spy
Sign, can enable has image to be compared at once when needing to identify, improves the efficiency of device operation.
Further, in another embodiment of item identification devices the present invention is based on image procossing, the present invention is based on figures
As the item identification devices of processing include: image collection module 510, detection module 520, characteristic extracting module 530, straightening die
Block 540, matching identification module 550, wherein in addition to module corresponding in embodiment corresponding with Fig. 4 executes identical operation
Outside, characteristic extracting module 530 and correction module 540 also execute following operation:
Characteristic extracting module 530 is also used to extract the fifth feature value for comparing article, and, extract the sixth feature value of article;
Correction module 540 is also used to be corrected fifth feature value, until the comparison article picture size after correction
With the article full size that compares in the second pre-set zoom ratio;And by the second correction function to sixth feature value into
Row correction obtains fourth feature value.
In the present embodiment, by images of items in database also by with compare images of items carry out it is identical correction to reach
To pre-set dimension and form, it is compared with article to be identified using the identical article that compares, is had specifically due to comparing article
Size, shape and pattern reduce error caused by due to acquiring different comparison article characteristic values, further improve identification
Precision.
It is understood that the second pre-set zoom ratio and the first pre-set zoom ratio can in above-mentioned apparatus embodiment
It, can not also be identical with identical.
In addition, in embodiments of the present invention, except the scaling processing for carrying out same ratio, progress that can also be additional other
Correction, for example, eliminate the trapezoidal distortion of image, bright and dark light and reflective etc..
For example, image can be carried out restoration school by correction module 540, eliminates or reduction causes because of shooting angle difference
Trapezoidal distortion, and/or, eliminate or reduce the problem because of image light and shade unevenness caused by light problem, make the image after correcting
To greatest extent close to authentic item, the accuracy rate that image compares is improved, the speed and order of accuarcy of identification are further increased.
Further, in another embodiment of item identification devices the present invention is based on image procossing, the present invention is based on figures
As the item identification devices of processing include: image collection module 510, detection module 520, characteristic extracting module 530, straightening die
Block 540, matching identification module 550, wherein matching identification module 550, comprising:
Similarity determining unit 551, for according to third feature value and fourth feature value comparison result determine article to be identified and
The similarity of each images of items stored in database;
Sequencing unit 552, for carrying out descending sort according to the similarity of article to be identified to article in database.
In the present embodiment, similarity determining unit, which compares third feature value and fourth feature, is worth article sum number to be identified
According to the similarity of the article stored in library, sequencing unit carries out the article in database according to the similarity of article to be identified
Descending arrangement.The item pictures of similarity most can will be listed in foremost with article to be identified by the present embodiment, and user can
In very clear acquisition database and the highest article of article similarity to be identified.It is higher that other similarities can also be obtained simultaneously
Article picture, convenient for user's further progress screen and judge.
Fig. 6 is that the application is based on one specific example flow chart of image procossing item identification method, in this specific example, root
According to shown in Fig. 6, be based on image procossing item identification method the following steps are included:
610, article is detected from the image stored in database and compares article, wherein images of items and comparison in database
Scaling between product image and corresponding material object is identical, extracts the fifth feature value for comparing article, extracts the 6th of article
Characteristic value;
620, fifth feature value is corrected, until compare images of items after correction and to compare article material object etc. big, is determined the
The corresponding image rectification of five characteristic values second correction function big to grade;
630, sixth feature value is corrected according to the second correction function to obtain fourth feature value;
Due to the article that stores in database and compare article scaling having the same, will compare article be corrected to
Under the premise of material object etc. is big, the image of the article stored in database is also corrected to and article material object with identical correction function
Deng big.The corresponding characteristic value with the big image such as in kind stored in database of the fourth feature value.
Step 610-630 passes through scaling correction, and the image rectification in kind stored in database is and the big figures such as material object
Picture obtains corresponding characteristic value.
640, it obtains and compares images of items, image is corrected, make to compare images of items and compare article material object etc. greatly;
Size due to comparing article is can to know, can will compare article image rectification to image and pantograph ratio in kind
Example is 1:1.
650, image feature value is as the first Standard Eigenvalue for comparing article after obtaining correction;
Scaling due to comparing images of items and material object after correction is 1:1, and the first Standard Eigenvalue corresponds to comparison
The characteristic value of image when product scaling is 1:1, the i.e. characteristic value of the image big with article etc..
660, it obtains article to be identified and compares the image of article, wherein images of items to be identified and article to be identified are real
The scaling of object is identical as comparing images of items and comparing the scaling of article material object;
Image can be with primary shooting formation, the image of identical two articles of available scaling herein,
It can be what shooting respectively was formed, as long as scaling is identical.
670, it detects to compare article and article to be identified from image, and extract the First Eigenvalue for comparing article, with
And extract the Second Eigenvalue of article to be identified;
680, the First Eigenvalue is corrected to the first Standard Eigenvalue, the first Standard Eigenvalue is it is known that determining fisrt feature
Value is corrected to the first correction function of the first Standard Eigenvalue;
By or the First Eigenvalue of comparison article be corrected to the first Standard Eigenvalue, correspond to and be and material object by image rectification
Scaling be 1:1, and thus obtain first correction function.
690, it is corrected according to Second Eigenvalue of first correction function to article to be identified, obtains third feature value;
First correction function, which will compare article, has same zoom ratio adjustment to big with material object etc. from article to be identified, passes through
First correction function is corrected the Second Eigenvalue of article to be identified, obtained third feature value correspondence and article to be identified
Etc. the characteristic value of big image.
700, third feature value is compared with the fourth feature value of images of items each in database, according to third feature
Value and fourth feature value comparison result determine the similarity of each images of items stored in article and database to be identified;
710, descending sort is carried out according to the similarity of article to be identified to article in database.
It is understood that can only show the article in similarity N number of database in the top after sequence.
It can specifically be realized by the prior art by the method that characteristic value is identified and judgeed.When being compared,
Due to there are multiple images of items in database, article to be identified is compared respectively with it, if it is judged that being that the two is
Same article can then stop comparing, if it is determined that some article is different article in article to be identified and database, then
Continue for other images of items in article to be identified and database to be compared, until finding in the database and object to be identified
Product export result.It is not found and the matched article characteristics of article characteristics value to be identified alternatively, entire database may also be traversed
Value, can also export the result.
It can be seen from the above, it is 1:1 that third feature, which is worth corresponding image and the scaling of article material object to be identified, the 4th is special
The corresponding image of value indicative and article material object scaling are also 1:1, therefore, when to be identified to judge by comparing image feature value
When whether the corresponding article of image is identical in article and database, shape, pattern and size can be compared simultaneously, to reduce
It is judged by accident caused by because of scaling difference, improves the efficiency and accuracy of identification.
The apparatus embodiments described above are merely exemplary, wherein described, unit can as illustrated by the separation member
It is physically separated with being or may not be, component shown as a unit may or may not be physics list
Member, it can it is in one place, or may be distributed over multiple network units.It can be selected according to the actual needs
In some or all of the modules achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying creativeness
Labour in the case where, it can understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be embodied in the form of software products in other words, should
Computer software product may be stored in a computer readable storage medium, such as ROM/RAM, magnetic disk, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Method described in certain parts of example or embodiment.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features;
And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (12)
1. a kind of item identification method based on image procossing characterized by comprising
It obtains article to be identified and compares the image of article, wherein the images of items to be identified and the article to be identified are real
The scaling of object compares images of items and the scaling for comparing article material object is identical with described, the article that compares
Size can be known;
The comparison article and the article to be identified are detected from described image, and extract first spy for comparing article
Value indicative, and extract the Second Eigenvalue of the article to be identified;
The First Eigenvalue is corrected to the first Standard Eigenvalue, first Standard Eigenvalue can obtain, and determine institute
State the first correction function that the First Eigenvalue is corrected to first Standard Eigenvalue;
It is corrected according to Second Eigenvalue of first correction function to the article to be identified, obtains third feature value;
The fourth feature value of each images of items stored in the third feature value and database is compared, according to the ratio
The article to be identified is identified to result, the fourth feature value is obtained through correction in advance.
2. the method according to claim 1, wherein the comparison article is that shape, pattern and size have standard
Article.
3. the method according to claim 1, wherein acquisition first Standard Eigenvalue includes:
Obtain the image for comparing article;
Described image is corrected, makes the comparison images of items and the article material object that compares with the first pre-set zoom ratio
It zooms in and out;
Image feature value is as first Standard Eigenvalue for comparing article after obtaining correction.
4. according to the method described in claim 3, it is characterized in that, described obtain images of items in database through correction in advance
Fourth feature value includes:
The article and the comparison article are detected from the image of the database, and extract the 5th spy for comparing article
Value indicative extracts the sixth feature value of the article, wherein article and the comparison article image scaling ratio in the database
It is identical;
The fifth feature value is corrected, until corresponding the comparisons images of items and the article that compares are in kind after correction
For the second pre-set zoom ratio, determines and the fifth feature is worth corresponding image rectification to the second of the second pre-set zoom ratio
Correction function;
The sixth feature value is corrected according to second correction function to obtain the fourth feature value.
5. according to the method described in claim 4, it is characterized in that, the second pre-set zoom ratio and the first pre-set zoom ratio
Example is identical.
6. according to the method described in claim 4, making the comparison article contract it is characterized in that, be corrected to described image
It puts to the first pre-set zoom ratio, comprising:
Restoration correction is carried out to the comparison images of items;And/or
It is described that fifth feature value is corrected, until corresponding the comparisons images of items and the article that compares are in kind after correction
For the second pre-set zoom ratio, comprising:
Restoration correction is carried out to the comparison images of items.
7. the method according to claim 1, wherein described detect to compare article and object to be identified from image
Product, comprising:
Region division is carried out to described image, the article characteristics to be identified and the comparison extracted using convolutional neural networks
Article characteristics are classified and are scored to described each region respectively;
The article to be identified and the score for comparing article, position are respectively obtained according to the classification of described each region and scoring
Confidence breath and size.
8. method according to any one of claims 1-7, which is characterized in that it is described will be in third feature value and database
The fourth feature value of each images of items of storage is compared, and identifies the article to be identified according to the comparison result, comprising:
It is determined in the article to be identified and database and is deposited according to the third feature value and the fourth feature value comparison result
The similarity of each images of items of storage;
Descending sort is carried out according to the similarity of the article to be identified to article in the database.
9. a kind of item identification devices based on image procossing characterized by comprising
Image collection module, the image for obtaining article to be identified Yu comparing article, wherein the images of items to be identified and
The scaling of the article material object to be identified compares images of items and the scaling phase for comparing article material object with described
Together, the size for comparing article can be learned;
Detection module, for detecting the comparison article and the article to be identified from described image;
Characteristic extracting module for extracting the First Eigenvalue for comparing article, and extracts the of the article to be identified
Two characteristic values;
Correction module, for the First Eigenvalue to be corrected to first Standard Eigenvalue;And by the second feature
Value is corrected to the third feature value;
Matching identification module, for comparing the fourth feature value of the third feature value and images of items each in database
It is right, the article to be identified is identified according to the comparison result, the fourth feature value of images of items is through institute in the database
It states correction module and in advance corrects and obtain.
10. device according to claim 9, which is characterized in that
The correction module is also used to obtain described image the comparison images of items of module acquisition with the first preset ratio
It zooms in and out;
The characteristic extracting module is also used to obtain the comparison article figure that the correction module is zoomed in and out with the first preset ratio
The characteristic value of picture, as first Standard Eigenvalue for comparing article.
11. device according to claim 10, which is characterized in that
The detection module is also used to detect the article and the comparison article from each image of the database;
The characteristic extracting module is also used to extract the fifth feature value for comparing article, and extract the article the 6th is special
Value indicative;
The correction module is also used to be corrected the fifth feature value, until the comparison images of items and institute after correction
Stating and comparing article material object is the second pre-set zoom ratio;And school is carried out to the sixth feature value by the second correction function
Just obtaining the fourth feature value.
12. the device according to any one of claim 9-11, which is characterized in that the matching identification module, comprising:
Similarity determining unit, it is described wait know for being determined according to the third feature value and the fourth feature value comparison result
The similarity of each images of items stored in other article and database;
Sequencing unit, for carrying out descending sort according to the similarity of the article to be identified to article in the database.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811010868.4A CN109508623A (en) | 2018-08-31 | 2018-08-31 | Item identification method and device based on image procossing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811010868.4A CN109508623A (en) | 2018-08-31 | 2018-08-31 | Item identification method and device based on image procossing |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109508623A true CN109508623A (en) | 2019-03-22 |
Family
ID=65745656
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811010868.4A Pending CN109508623A (en) | 2018-08-31 | 2018-08-31 | Item identification method and device based on image procossing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109508623A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112149507A (en) * | 2020-08-25 | 2020-12-29 | 合肥工业大学 | Unmanned aerial vehicle autonomous ground pollutant reasoning and positioning method and system based on images |
CN113012407A (en) * | 2021-02-18 | 2021-06-22 | 上海电机学院 | Eye screen distance prompt myopia prevention system based on machine vision |
CN113569911A (en) * | 2021-06-28 | 2021-10-29 | 北京百度网讯科技有限公司 | Vehicle identification method and device, electronic equipment and storage medium |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050141781A1 (en) * | 2003-12-29 | 2005-06-30 | Chun-Huang Lin | One dimensional feature enhancement |
CN101122950A (en) * | 2007-09-04 | 2008-02-13 | 成都方程式电子有限公司 | Fingerprint elastic deformation correction method and device |
CN102110285A (en) * | 2009-12-28 | 2011-06-29 | 佳能株式会社 | Data correction apparatus, control method thereof and image discriminating apparatus |
US7983486B2 (en) * | 2007-08-29 | 2011-07-19 | Seiko Epson Corporation | Method and apparatus for automatic image categorization using image texture |
CN102136069A (en) * | 2010-01-25 | 2011-07-27 | 华晶科技股份有限公司 | Object image correcting device and method for identification |
CN103426185A (en) * | 2013-08-09 | 2013-12-04 | 北京博思廷科技有限公司 | Method and device for adjusting target scale in pan-tilt-zoom (PTZ) tracking process |
CN103971111A (en) * | 2013-02-04 | 2014-08-06 | 联想(北京)有限公司 | Image processing method and terminal equipment |
CN105512689A (en) * | 2014-09-23 | 2016-04-20 | 苏州宝时得电动工具有限公司 | Lawn identification method based on images, and lawn maintenance robot |
CN105654049A (en) * | 2015-12-29 | 2016-06-08 | 中国科学院深圳先进技术研究院 | Facial expression recognition method and device |
CN106780363A (en) * | 2016-11-21 | 2017-05-31 | 北京金山安全软件有限公司 | Picture processing method and device and electronic equipment |
CN107092914A (en) * | 2017-03-23 | 2017-08-25 | 广东数相智能科技有限公司 | Refuse classification method, device and system based on image recognition |
CN107871137A (en) * | 2016-10-18 | 2018-04-03 | 浙江大学 | A kind of material matching process based on image recognition |
CN108234770A (en) * | 2018-01-03 | 2018-06-29 | 京东方科技集团股份有限公司 | A kind of auxiliary make up system, auxiliary cosmetic method, auxiliary cosmetic device |
-
2018
- 2018-08-31 CN CN201811010868.4A patent/CN109508623A/en active Pending
Patent Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050141781A1 (en) * | 2003-12-29 | 2005-06-30 | Chun-Huang Lin | One dimensional feature enhancement |
US7983486B2 (en) * | 2007-08-29 | 2011-07-19 | Seiko Epson Corporation | Method and apparatus for automatic image categorization using image texture |
CN101122950A (en) * | 2007-09-04 | 2008-02-13 | 成都方程式电子有限公司 | Fingerprint elastic deformation correction method and device |
CN102110285A (en) * | 2009-12-28 | 2011-06-29 | 佳能株式会社 | Data correction apparatus, control method thereof and image discriminating apparatus |
CN102136069A (en) * | 2010-01-25 | 2011-07-27 | 华晶科技股份有限公司 | Object image correcting device and method for identification |
CN103971111A (en) * | 2013-02-04 | 2014-08-06 | 联想(北京)有限公司 | Image processing method and terminal equipment |
CN103426185A (en) * | 2013-08-09 | 2013-12-04 | 北京博思廷科技有限公司 | Method and device for adjusting target scale in pan-tilt-zoom (PTZ) tracking process |
CN105512689A (en) * | 2014-09-23 | 2016-04-20 | 苏州宝时得电动工具有限公司 | Lawn identification method based on images, and lawn maintenance robot |
CN105654049A (en) * | 2015-12-29 | 2016-06-08 | 中国科学院深圳先进技术研究院 | Facial expression recognition method and device |
CN107871137A (en) * | 2016-10-18 | 2018-04-03 | 浙江大学 | A kind of material matching process based on image recognition |
CN106780363A (en) * | 2016-11-21 | 2017-05-31 | 北京金山安全软件有限公司 | Picture processing method and device and electronic equipment |
CN107092914A (en) * | 2017-03-23 | 2017-08-25 | 广东数相智能科技有限公司 | Refuse classification method, device and system based on image recognition |
CN108234770A (en) * | 2018-01-03 | 2018-06-29 | 京东方科技集团股份有限公司 | A kind of auxiliary make up system, auxiliary cosmetic method, auxiliary cosmetic device |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112149507A (en) * | 2020-08-25 | 2020-12-29 | 合肥工业大学 | Unmanned aerial vehicle autonomous ground pollutant reasoning and positioning method and system based on images |
CN112149507B (en) * | 2020-08-25 | 2023-01-13 | 合肥工业大学 | Unmanned aerial vehicle autonomous ground pollutant reasoning and positioning method and system based on images |
CN113012407A (en) * | 2021-02-18 | 2021-06-22 | 上海电机学院 | Eye screen distance prompt myopia prevention system based on machine vision |
CN113569911A (en) * | 2021-06-28 | 2021-10-29 | 北京百度网讯科技有限公司 | Vehicle identification method and device, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wang et al. | MARCH: Multiscale-arch-height description for mobile retrieval of leaf images | |
CN108717531B (en) | Human body posture estimation method based on Faster R-CNN | |
CN104850850B (en) | A kind of binocular stereo vision image characteristic extracting method of combination shape and color | |
KR101303124B1 (en) | Apparatus and method for object tracking using feature descriptor, and apparatus and method for removing garbage feature | |
CN111160249A (en) | Multi-class target detection method of optical remote sensing image based on cross-scale feature fusion | |
CN110223310B (en) | Line structure light center line and box edge detection method based on deep learning | |
CN103077512A (en) | Feature extraction and matching method and device for digital image based on PCA (principal component analysis) | |
CN111126412B (en) | Image key point detection method based on characteristic pyramid network | |
Kiforenko et al. | A performance evaluation of point pair features | |
CN108564092A (en) | Sunflower disease recognition method based on SIFT feature extraction algorithm | |
CN102938065A (en) | Facial feature extraction method and face recognition method based on large-scale image data | |
CN104966081B (en) | Spine image-recognizing method | |
CN109508623A (en) | Item identification method and device based on image procossing | |
CN103778436B (en) | A kind of pedestrian's attitude detecting method based on image procossing | |
CN111709313B (en) | Pedestrian re-identification method based on local and channel combination characteristics | |
CN109871821A (en) | The pedestrian of adaptive network recognition methods, device, equipment and storage medium again | |
CN105654122B (en) | Based on the matched spatial pyramid object identification method of kernel function | |
CN114821102A (en) | Intensive citrus quantity detection method, equipment, storage medium and device | |
Bianco et al. | Robust smile detection using convolutional neural networks | |
Ding et al. | Recognition of hand-gestures using improved local binary pattern | |
EP3311333A1 (en) | Pairing of images of postal articles with descriptors of singularities of the gradient field | |
Yawen et al. | Research on vehicle detection technology based on SIFT feature | |
CN109472279A (en) | Item identification method and device based on image procossing | |
CN105095901B (en) | Method and apparatus for handling the block to be processed of sediment urinalysis image | |
CN107679528A (en) | A kind of pedestrian detection method based on AdaBoost SVM Ensemble Learning Algorithms |
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 |