CN105868676A - Improved two-dimension code region positioning system and positioning method thereof - Google Patents

Abstract

The invention discloses an improved two-dimension code region positioning system, which comprises an image collecting module, a training module, a feature extraction module, a feature classification module and a recognition module, wherein the image collecting module is used for collecting image layer information; the training module uses a sample image layer collected by the image collection module as input information to obtain a plurality of pieces of transmission image layer information relevant to the sample image through calculation; the feature extraction module uses the transmission image layer information to extract the feature vectors of image layers collected by the image collecting module; the feature classification module is used for performing classified storage on the feature vectors of the image layers; the recognition module uses the classification result of the feature vectors of the image layers to be recognized to be compared with the classification result of the feature vectors of the sample image layer, and judges whether the image layer to be recognized is a two-dimension code region or not. The invention also discloses a positioning method of the improved two-dimension code region positioning system. The defects in the prior art can be solved; the problems of the existing method that a great number of training samples are needed, but the conflict of trapping into the locally optimal solution can easily occur are solved.

Description

The two-dimension code area alignment system of a kind of improvement and localization method thereof

Technical field

The present invention relates to field of information security technology, the two-dimension code area alignment system of a kind of improvement and location thereof Method.

Background technology

In recent years, along with the development of the Internet and popularizing of smart mobile phone, and Quick Response Code is in quantity of information and safety side The advantage in face, the Quick Response Code application in fields such as information transmission, authentication and mobile payments is more and more extensive.Quick Response Code is applied Time, first the reading equipment (such as scanner or smart mobile phone) with photographic head is directed at Quick Response Code；Photographic head is adopted by reading equipment Collect to image carry out two-dimension code area location, i.e. finding the Quick Response Code in image；Intercept the successful image in location and carry out follow-up Video procession.

The influence factor of success or not of two-dimension code area location mainly has the following aspects:

1, reading the distance of equipment and Quick Response Code, when distance is for 25cm～30cm, position success rate is higher；

2, read the angle between equipment and Quick Response Code, read the angle gathered between face and Quick Response Code of equipment closer to 0 degree, position success rate is the highest；

3, the space deformation of Quick Response Code carrier, space deformation is the least, and position success rate is the highest；

4, Quick Response Code position in the picture, during closer to middle position, position success rate is the highest；

5, Quick Response Code occupies the ratio of image, and ratio is the biggest, and position success rate is the highest；

6, the background of Quick Response Code place image, background is the simplest, and time the highest with Quick Response Code contrast, position success rate is more High.

Existing two-dimension code area localization method cannot carry out feature extraction to target on deeper level, it is difficult in complexity Environment positions accurately.Existing two-dimension code area location technology is broadly divided into following a few class:

1, area positioning technology based on Quick Response Code designator, i.e. carries out region by the designator of searching Quick Response Code fixed Position.Although location technology based on designator is the most intuitively, but when image information is impaired, when particularly designator is impaired, hold Easily cause and position unsuccessfully.

2, two-dimension code area location technology based on textural characteristics, i.e. carries out two dimension by the textural characteristics of extraction Quick Response Code Code zone location.This kind of technology make use of the global characteristics of Quick Response Code, has certain robustness for image information is impaired, but very Difficulty extracts the Quick Response Code textural characteristics that adaptability is stronger.

3, two-dimension code area location technology based on frequency domain character, i.e. by image is transformed into frequency domain, and extracts two dimension The frequency domain character of code carries out zone location.This kind of technology is similar with technology based on textural characteristics, simply feature is turned from spatial domain Arrive frequency domain.

4, two-dimension code area location technology based on machine learning, i.e. finds out two-dimension code area by the method for machine learning With the grader of non-two-dimension code area, carry out two-dimension code area location.This kind of technology needs substantial amounts of sample to be trained, and works as When sample size is bigger, tend to be absorbed in locally optimal solution.

Summary of the invention

The technical problem to be solved in the present invention is to provide two-dimension code area alignment system and the localization method thereof of a kind of improvement, Can solve the problem that the deficiencies in the prior art, but a large amount of training sample of needs solving the existence of existing method is easily absorbed in local optimum The contradiction solved.

For solving above-mentioned technical problem, the technical solution used in the present invention is as follows.

A kind of two-dimension code area alignment system of improvement, including,

Image capture module, is used for gathering map data mining platform；

Training module, the sample graph layer that use image capture module collects is as input information, many by being calculated The individual transmission map data mining platform relevant to sample image；

Characteristic extracting module, uses transmission map data mining platform to put forward the characteristic vector of image capture module collection figure layer Take；

Tagsort module, for carrying out classification storage to the characteristic vector of figure layer；

Identification module, uses the classification results of figure layer characteristic vector to be identified and the classification results of sample graph layer characteristic vector Contrast, it is determined that whether figure layer to be identified is two-dimension code area.

The localization method of the two-dimension code area alignment system of a kind of above-mentioned improvement, comprises the following steps:

A, use image capture module 1 gather the sample map data mining platform of training, form sample set；

B, training module 2 calculate the probability that transmission figure layer jth node is 0 or 1,

$P(y_j=1|x)={(1+e^{-{\Σ}_i{k}_{ij}{x}_i-a_j})}^{-1},$

Wherein, x is sample graph layer, and y is transmission figure layer, and k is the degree of association between sample graph layer and transmission figure layer；

C, training module 2 calculate, according to the information of transmission figure layer, the probability that sample graph layer jth node is 0 or 1,

$P\left(x\right|y^0)=\underset{i}{\Π}P\left(x_i\right|y),$

$P(x_i=1|y^0)={(1+e^{-{\Σ}_i{k}_{ij}{x}_i-b_i})}^{-1};$

D, training module 2 calculate the δ making L (δ) must be worth maximum,

$\mathrm{\δ}=\{k,a,b\},L\left(\mathrm{\δ}\right)=\frac{1}{N}{\mathrm{\Σ}}_{n=1}^N\log P_{\mathrm{\θ}}\left(x^{\left(n\right)}\right)-\frac{\mathrm{\λ}}{N}\left|\right|k|{|}_F^2,$

The information of transmission figure layer is calculated by δ；

E, using known transmission figure layer as transmission figure layer, repeat step B～D, show that several transmit map data mining platform；

F, characteristic extracting module 3 use transmission map data mining platform to carry out the characteristic vector of image capture module 1 collection figure layer Extract；

G, tagsort module 4 carry out classification storage to the characteristic vector of figure layer；

H, figure layer to be identified is brought in step F and G, obtain the classification results of figure layer characteristic vector to be identified, use sample The classification results of this figure layer characteristic vector contrasts with the classification results of figure layer characteristic vector to be identified, it is determined that figure layer to be identified Whether it is two-dimension code area；If figure layer to be identified is two-dimension code area, realize the location of two-dimension code area, if figure layer to be identified It not that two-dimension code area then repeats step H, till two-dimension code area being detected.

As preferably, in step E, transmission figure layer and sample graph layer are adjusted.

As preferably, in step F, extract characteristic vector and comprise the following steps,

F1, each transmission figure layer is used to obtain the transformation matrix between transmission figure layer and target figure layer respectively；

F2, obtain the characteristic vector of each transformation matrix；

F3, the most linear incoherent characteristic vector of extraction are as the characteristic vector of target figure layer.

As preferably, in step G, characteristic vector classification is comprised the following steps,

G1, obtain the eigenmatrix of target figure layer gray scale, obtain the characteristic vector of eigenmatrix；

It is right that each characteristic vector of the target figure layer obtained in G1, use step F and the characteristic vector of eigenmatrix are carried out Ratio, asks for its similarity；

The characteristic vector of target figure layer is divided into 10 groups～20 groups by G3, height according to similarity.

What employing technique scheme was brought has the beneficial effects that: the present invention can automatically carry out feature by multilamellar Extracting, whole process need not any priori.Feature not only includes color, edge and the texture etc. that existing method extracts Information, further comprises the Quick Response Code feature of the high abstraction that existing method cannot be extracted.These features substantially increase Quick Response Code Some existing methods cannot be positioned the image of Quick Response Code, also can accurately position by the efficiency of zone location and precision.

Accompanying drawing explanation

Fig. 1 is the schematic diagram of one detailed description of the invention of the present invention.

In figure: 1, image capture module；2, training module；3, characteristic extracting module；4, tagsort module；5, mould is identified Block.

Detailed description of the invention

With reference to Fig. 1, one detailed description of the invention of the present invention includes,

Image capture module 1, is used for gathering map data mining platform；

Training module 2, the sample graph layer that use image capture module 1 collects is as input information, by being calculated Multiple transmission map data mining platform relevant to sample image；

Characteristic extracting module 3, uses transmission map data mining platform to put forward the characteristic vector of image capture module 1 collection figure layer Take；

Tagsort module 4, for carrying out classification storage to the characteristic vector of figure layer；

Identification module 5, uses the classification results of figure layer characteristic vector to be identified and the classification knot of sample graph layer characteristic vector Fruit contrasts, it is determined that whether figure layer to be identified is two-dimension code area.

The localization method of the two-dimension code area alignment system of one above-mentioned improvement, comprises the following steps,

A, use image capture module 1 gather the sample map data mining platform of training, form sample set；

B, training module 2 calculate the probability that transmission figure layer jth node is 0 or 1,

$P(y_j=1|x)={(1+e^{-{\Σ}_i{k}_{ij}{x}_i-a_j})}^{-1},$

Wherein, x is sample graph layer, and y is transmission figure layer, and k is the degree of association between sample graph layer and transmission figure layer；

C, training module 2 calculate, according to the information of transmission figure layer, the probability that sample graph layer jth node is 0 or 1,

$P\left(x\right|y^0)=\underset{i}{\Π}P\left(x_i\right|y),$

$P(x_i=1|y^0)={(1+e^{-{\Σ}_i{k}_{ij}{x}_i-b_i})}^{-1};$

D, training module 2 calculate the δ making L (δ) must be worth maximum,

$\mathrm{\δ}=\{k,a,b\},L\left(\mathrm{\δ}\right)=\frac{1}{N}{\mathrm{\Σ}}_{n=1}^N\log P_{\mathrm{\θ}}\left(x^{\left(n\right)}\right)-\frac{\mathrm{\λ}}{N}\left|\right|k|{|}_F^2,$

The information of transmission figure layer is calculated by δ；

E, using known transmission figure layer as transmission figure layer, repeat step B～D, show that several transmit map data mining platform；

F, characteristic extracting module 3 use transmission map data mining platform to carry out the characteristic vector of image capture module 1 collection figure layer Extract；

G, tagsort module 4 carry out classification storage to the characteristic vector of figure layer；

H, figure layer to be identified is brought in step F and G, obtain the classification results of figure layer characteristic vector to be identified, use sample The classification results of this figure layer characteristic vector contrasts with the classification results of figure layer characteristic vector to be identified, it is determined that figure layer to be identified Whether it is two-dimension code area；If figure layer to be identified is two-dimension code area, realize the location of two-dimension code area, if figure layer to be identified It not that two-dimension code area then repeats step H, till two-dimension code area being detected.

In step E, transmission figure layer and sample graph layer are adjusted.During adjustment, use transmission figure layer and sample graph The characteristic point that in Ceng, texture image is identical is as a reference point, is adjusted the edge of figure layer, and figure tomographic image follows layer edges Conscientious synchronization control, makes the similarity of whole figure layer remain maximum.This can effectively reduce each transmission figure layer when asking for The picture distortion produced.

It addition, degree of association k between sample graph layer and transmission figure layer is before every use, the joint once calculated before all using Point probability and degree of association k are modified,

k′_j=k_j+P(k_j-1-k_j),

This can effectively reduce the Acquisition Error impact for whole figure layer characteristic vector of single sample, and it is whole fixed to improve The serious forgiveness of position process.

The present invention can realize quick two-dimension code area location, and method is easy, and recognition accuracy is high.

Foregoing description is only used as the enforceable technical scheme of the present invention and proposes, single not as to its technical scheme itself Restrictive condition.

Claims (5)

1. the two-dimension code area alignment system improved, it is characterised in that: include,

Image capture module (1), is used for gathering map data mining platform；

Training module (2), the sample graph layer that use image capture module (1) collects is as input information, by being calculated Multiple transmission map data mining platform relevant to sample image；

Characteristic extracting module (3), uses transmission map data mining platform to put forward the characteristic vector of image capture module (1) collection figure layer Take；

Tagsort module (4), for carrying out classification storage to the characteristic vector of figure layer；

Identification module (5), uses the classification results of figure layer characteristic vector to be identified and the classification results of sample graph layer characteristic vector Contrast, it is determined that whether figure layer to be identified is two-dimension code area.

2. the localization method of the two-dimension code area alignment system of the improvement used described in claim 1, it is characterised in that bag Include following steps:

A, use image capture module (1) gather the sample map data mining platform trained, and form sample set；

B, training module (2) calculate the probability that transmission figure layer jth node is 0 or 1,

$P(y_j=1|x)={(1+e^{-{\mathrm{\Σ}}_i{k}_{ij}{x}_i-a_j})}^{-1},$

Wherein, x is sample graph layer, and y is transmission figure layer, and k is the degree of association between sample graph layer and transmission figure layer；

C, training module (2) calculate, according to the information of transmission figure layer, the probability that sample graph layer jth node is 0 or 1,

$P\left(x\right|y^0)=\underset{i}{\mathrm{\Π}}P\left(x_i\right|y),$

$P(x_i=1|y^0)={(1+e^{-{\mathrm{\Σ}}_i{k}_{ij}{x}_i-b_i})}^{-1};$

D, training module (2) calculate the δ making L (δ) must be worth maximum,

$\mathrm{\δ}=\{k,a,b\},L\left(\mathrm{\δ}\right)=\frac{1}{N}{\mathrm{\Σ}}_{n=1}^N{\mathrm{logP}}_{\mathrm{\θ}}\left(x^{\left(n\right)}\right)-\frac{\mathrm{\λ}}{N}\left|\right|k|{|}_F^2,$

The information of transmission figure layer is calculated by δ；

E, using known transmission figure layer as transmission figure layer, repeat step B～D, show that several transmit map data mining platform；

F, characteristic extracting module (3) use transmission map data mining platform to carry out the characteristic vector of image capture module (1) collection figure layer Extract；

G, tagsort module (4) carry out classification storage to the characteristic vector of figure layer；

H, figure layer to be identified is brought in step F and G, obtain the classification results of figure layer characteristic vector to be identified, use sample graph The layer classification results of characteristic vector contrasts with the classification results of figure layer characteristic vector to be identified, it is determined that whether figure layer to be identified For two-dimension code area；If figure layer to be identified is two-dimension code area, realize the location of two-dimension code area, if figure layer to be identified is not Two-dimension code area then repeats step H, till two-dimension code area being detected.

Localization method the most according to claim 2, it is characterised in that: in step E, transmission figure layer and sample graph layer are carried out Adjust.

Localization method the most according to claim 2, it is characterised in that: in step F, extract characteristic vector and include following step Suddenly,

F1, each transmission figure layer is used to obtain the transformation matrix between transmission figure layer and target figure layer respectively；

F2, obtain the characteristic vector of each transformation matrix；

F3, the most linear incoherent characteristic vector of extraction are as the characteristic vector of target figure layer.

Localization method the most according to claim 2, it is characterised in that: in step G, include following for characteristic vector classification Step,

G1, obtain the eigenmatrix of target figure layer gray scale, obtain the characteristic vector of eigenmatrix；

The each characteristic vector of target figure layer obtained in G1, use step F contrasts with the characteristic vector of eigenmatrix, asks Take its similarity；

The characteristic vector of target figure layer is divided into 10 groups～20 groups by G3, height according to similarity.