CN104376280A - Image code generating method for Google project glass - Google Patents

Abstract

The invention discloses an image code generating method for the Google project glass and relates to the field of information processing. The image code generating method for the Google project glass comprises the steps that an original image obtained from an image reading device is converted into a plurality of unfolded images in the geometrical or physical form, the characteristic information of each image is extracted based on the self-organizing probability scale, then numeralization processing is conducted, and finally an image code is generated. The image code generating method for the Google project glass has the advantages that the storage space, occupied by the characteristic information of the images, of a server is small, the processing efficiency is high, and the image code generating method is suitable for wide-range Internet of Things website systems.

Description

A kind of image code generation method towards Google's glasses

Technical field

The invention belongs to a kind of image code generation method towards Google's glasses in image processing field.

Background technology

In the last few years, by mobile phone, smart mobile phone etc. reads Quick Response Code, and the method then downloading video corresponding to Quick Response Code is widely used.(patent documentation 1)

But, after printed article prints Quick Response Code, not only the attractive in appearance of printed article is damaged, and printing Quick Response Code also needs necessary space.Therefore go out to send consideration from the attractive in appearance of printed article and printing space, adopt eletric watermark technology, the system using smart mobile phone to carry out reading also is widely adopted.The gimmick of representative is relatively had to have " high speed signal detects and dispersion calculates on portable computing equipment] method mentioned of patent file ".(patent documentation 2)

Simultaneously, also use pattern recognition technique is had, carry out image reading, based on the image interconnection network read, the expansion reality technology of the so-called AR (Augmented Reality) that information that the image of reading and image itself are associated shows simultaneously on picture ( www.sbbit.jp).

In order to do one's utmost the capacity reducing the information be associated with characteristics of image, people are also groping the technology by the direct synthetic image code of image, and the patent of having applied for has " image code equipment, method and program ".(patent documentation 3)

Published technical literature

Patent documentation:

[patent documentation 1] (Te Open No. 2009-104164 bulletin)

[patent documentation 2] (Te Open No. 2011-234376 bulletin)

[patent documentation 3] (Te Open No. 2011-94551 bulletin)

[non-patent literature 1] " AR application note " ( www.ar.selkosha-p.com<http: //wwww.ar .selkosha-p.com>)

Access time: on August 6th, 2013

The method described in above-mentioned patent documentation 1, uses portable phone to read the technology of Quick Response Code and be widely used, but Quick Response Code needs private space, Commercial goods labels and packaging upper use time, can damage attractive in appearance.

In the smart mobile phone reading system recorded in above-mentioned patent documentation 2, because Quick Response Code can be hidden in Commercial goods labels, aesthetically impact is not had so to pack etc. the label of commodity, but need in advance Information hiding to be entered, need complicated printing means, printing cost and spended time all can increase, because the operation needing prior information to imbed needs long time so this means are extensively universal.

Meanwhile, the expansion reality technology of the AR (Augmented Reality) mentioned in non-patent literature 1, resolves according to the profile of image and carries out specific to image, then to acquisition related information.Therefore, by using the image of same profile likely to get relevant information, and be difficult to prevent from forging.Further, this gimmick have employed the recognition methods of images match, and single image just needs the images match information of registration several million in internal memory, therefore needs jumbo memory media (internal memory).

The image code mentioned in above-mentioned patent documentation 3, refer to setting level threshold value in the picture in advance, Iamge Segmentation specific in this is become some regions, gray scale in the region each be divided into is straight to be contrasted with the realization criteria threshold set in the picture, by being greater than criteria threshold, still be less than criteria threshold and judge it is information " 1 ", or information " 0 ".The generation method of this image code, with to mention technology in patent documentation 2 the same, all can only be carried out the generation of image code, can not carry out the generation of image code for specific image for specific image.

Summary of the invention

First object of the present invention is that providing a method, extract the characteristic information of image from original image out, is method for distinguishing according to image, retrieves the intrinsic code of original image from the image feature data logged in.

Second object of the present invention is, provides one directly by the method for not specific printing images synthetic image code.

3rd object of the present invention provides a generation method that can be used as false proof specific image code.

In order to solve above-mentioned problem, following technical scheme is proposed:

Towards a generation method for the image code of Google's glasses, be by image conversion part, characteristic information extraction section, image code generating portion composition, there is following feature:

The original image that image conversion part will get from image-reading device, is transformed into a plurality of geometry, or the unfolded image of physics form;

The image of a plurality of expansion that feature extraction section will be produced by image conversion part, based on the probability scale of self-organization, extracts the characteristic information of each image out;

Image code generating portion, by being extracted out the characteristic information of each image by feature extraction section, carries out the process quantized, synthetic image code.

And, above-mentioned image conversion part, described image conversion is the distribution of the gray scale of each pixel according to original image information, original image is transformed into the unfolded image with geometric shape facility, there is the unfolded image of physical energy feature, edge unfolded image, in the unfolded image of similar degree, plural number plants unfolded image.

And, above-mentioned characteristic information extraction section, described self-organization probability scale refers to, comprise normal distribution, multivariate normal distributes, exponential distribution, traffic distribution, Weibull distribution, triangle distribution, in beta distribution, at least one has the parameter of the probability attribute of probability distribution, the central value of described self-organization, refer to the mean value of probability distribution, or expected value.

And above-mentioned feature extraction section, is divided into multiple image-region respectively by above-mentioned plural unfolded image, based on self-organization probability scale, from the field of each Iamge Segmentation, extract the characteristic information of unfolded image out;

Above-mentioned image code generating portion, by the characteristic information of plural unfolded image, by single-bit, or the form of many bits carries out the process that quantizes, and direct synthetic image code.

And above-mentioned feature extraction section, by a plurality of unfolded image, based on the probability scale of self-organization, extracts the characteristic information of each unfolded image out;

Above-mentioned image code generating portion, by the characteristic information of all unfolded images extracted out by feature extraction section, uses a plurality of membership functions defined based on artificial experience, quantizes between 0 to n numerical value, and direct synthetic image code.

And above-mentioned image code generating portion, by the characteristic information of the unfolded image of each original image quantized, as the proper vector of multiple image, and for Login Register to server constituting the characteristic vector space of image;

Calculate and belong to the Euclidean distance (Euclidean distance) of each proper vector logged in the proper vector of the image of characteristic vector space and the characteristic vector space of image, using the proper vector apart from minimum characteristic vector space as the image code of present image.

And, above-mentioned code building portion, in order to generate the image code logged in, with multiple images of the same reading object got different opportunitys, obtain the characteristic information of each fixing multiple unfolded images, as the proper vector of multiple image information, based on self-organization probability scale, calculate central value and the probability scale of the proper vector of present image, form the new characteristic vector space of server log;

For the image feature vector belonging to new characteristic vector space, the central value of the proper vector of each image logged in characteristic vector space and dispersion value is used to carry out the calculating of the distance of the distance probability scale of probability scale, using the nearest image code of central value as present image belonging to the proper vector of characteristic vector space of probability scale.

Accompanying drawing explanation

Fig. 1 uses the read method schematic diagram of Google's glasses

Fig. 2 self-organization treatment scheme schematic diagram

Fig. 3 carries out the schematic diagram of the transform method of unfolded image by geometry or physics

Fig. 4 extracts the schematic diagram of image feature information out by the conversion of energygram picture

Fig. 5 extracts the schematic diagram of image feature information out by the conversion of shape image

Fig. 6 extracts the schematic diagram of image feature information out by the conversion of edge image

Fig. 7 extracts the schematic diagram of image feature information out by the conversion of similar image

Fig. 8 carries out image code schematic diagram by the segmentation of energygram picture

Gray scale adjustment principle schematic between Fig. 9 polychrome figure

Figure 10 is the image construction principle schematic of anti-counterfeiting image code

Figure 11 utilizes micro lens arrays to realize the schematic diagram of the image generating method of anti-counterfeiting image code

Google's glasses system that Figure 12 is used for Internet of Things website forms schematic diagram

Figure 13 image code searching system treatment scheme schematic diagram

The System's composition schematic diagram of Figure 14 image code searching system

(401) energy lofty perch in energygram picture

(402) profile of energygram picture

(501) the rounded nose position of character image

(502) result of calculation of self-organization

(801) centre sphere of energygram picture

(802) field, upper left of energygram picture

(803) the upper right field of energygram picture

(804) field, bottom right of energygram picture

(805) field, lower-left of energygram picture

Embodiment

Below in conjunction with accompanying drawing, the embodiment of the present invention is further described, but embodiment of the present invention is illustrative, instead of determinate.

The present invention solves its technical matters and takes following technical scheme to realize:

Image is used to be described for example of the present invention.

Fig. 1 is the explanation schematic diagram of image code transform method of the present invention and conversion equipment.Fig. 1 is described for Google's glasses.

In the following description; often can be described with Google's glasses; but the present invention is not only confined to Google's glasses; use smart mobile phone, portable phone and other mobile terminal, line loop; the online terminal of dedicated circuit wiring; monitoring camera, the certification camera of automatic ticket checker, the system of credit card terminal etc. can use.

Fig. 1 (a) is the schematic diagram using the Google glasses image to Commercial goods labels or packaging to read, Fig. 1 (b) is for mobile terminals such as Google's glasses, introduces the present invention about having the high-level schematic functional block diagram of the device of the code building function of image.In Fig. 1 (b), necessary functional unit is all adopted to the calcspar of solid line.

Such as, the image reading unit 11 of the mobile terminals 10 such as Google's glasses is towards Commercial goods labels (1-2), and CCD makes a video recording the image reading unit 11 of first-class formation, gets image information, then sends to terminal control portion 14, be then presented at image expressed portion 12.Meanwhile, image information also can be saved in data accumulating portion 15.When receiving signal specific from data input part 13, terminal control portion 14 and image code transformation component 16 co-treatment, get the peculiar unique code of image.Image code, based on the steering order in terminal control portion 14, is saved in data accumulating portion 15, also by Department of Communication Force 17, based on terminal control portion the current command, is sent in the external unit of network connection.Coming from the information of network facet, is by Department of Communication Force 17, and then terminal control portion 14 is presented on image displaying part 12, and the information that image reading section obtains also can be simultaneously displayed on image displaying part 12.

Although Fig. 1 is the example of carrying out with mobile terminal, be connected with the computer of network, POS terminal equipment, and other retrieval specialized equipment etc. is all fine.

Fig. 2 is self-organization treatment scheme.As shown in Figure 2, the self-organized algorithm based on probability scale is made up of following 4 steps.

STEP1: pre-treatment step: M ⁽⁰⁾as initialization probability yardstick, (x ₀, y ₀) ⁽⁰⁾as the central value of self-organization, V is as the convergency value of self-organization, and MN is as self-organization maximum tissue time numerical value, and n=0 is as the current number of times of self-organization.

About M ⁽⁰⁾as initialization probability yardstick and (x ₀, y ₀) ⁽⁰⁾as the determining method of the central value of self-organization, tight setting need not be carried out.By artificial prediction, for final scope, a part is had at least to be included in initialization probability yardstick M ⁽⁰⁾scope in.Initialization probability yardstick M ⁽⁰⁾larger, the time of calculating is longer, otherwise too little, likely can not get correct result.

About the establishing method of V as convergency value, convergency value V is larger, just likely can not get correct result.Be worth less, the time calculating cost is longer.Correct establishing method is about 10% of the probability of final self-organization.

About the establishing method of maximum self-organization number of times MN, be generally 5-10 time just enough.

STEP2: self-organization step: carry out n self-organization process, (x _n, y _n) ⁽ⁿ⁾as self-organization central value, probability scale M ⁽ⁿ⁾as radius, calculate all pixel I (x within radius _i, y _j) (i=1,2 ..., k, j=1,2 ..., 1) gray-scale intensity dispersion value S ⁽ⁿ⁺¹⁾.M ⁽ⁿ⁺¹⁾＝S ⁽ⁿ⁺¹⁾，n＝n+1。

[formula 1]

${X_0}^{\left(n\right)}={X_0}^{(n-1)}+\frac{\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}(x_i-{X_0}^{(n-1)})I(x_i,y_j)}{\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}I(x_i,y_j)}$

${Y_0}^{\left(n\right)}={Y_0}^{(n-1)}+\frac{\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}(y_i-{Y_0}^{(n-1)})I(x_i,y_j)}{\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}I(x_i,y_j)}$

[formula 2]

$S^{\left(n\right)2}=\frac{\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}[{(x_i-{X_0}^{\left(n\right)})}^2+{(y_j-{Y_0}^{\left(n\right)})}^2]I(x_i,y_j)}{\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}I(x_i,y_j)}$

X ₀ ^(n-1), y ₀ ^(n-1)as image I (x last time _i, y _j) (i=1,2 ..., k, j=1,2 ..., 1) self-organization initial point.

STEP3: self-organization discriminating step.Self-organization process reaches maximum times (N>=MN) or self-organization process convergence (M ⁽ⁿ⁾-M ⁽ⁿ⁺¹⁾≤ V), as being YES, just do not carry out the self-organization process of next time, self-organization terminates to jump to STEP4.If NO, just jump to STEP2 and proceed self-organization process.

STEP4: self-organization process terminates.

In the present invention, the x of formula 1 ₀ ⁽ⁿ⁾, Y ₀ ⁽ⁿ⁾as common geometric coordinate center, or mechanics center of gravity physically, or mathematical digital average value, or probability is average, the probability center of expectation value.

Probability scale M ⁽ⁿ⁾it is a parameter with the probability statistics of multiple attributes.For example normal distribution, exponential distribution, Erlangian distribution, Weibull distribution, triangle distribution, beta distribution etc.

Such as probability scale M ⁽ⁿ⁾just can as the dispersion value of normal distribution.

Example: the example of a normal probability distribution.With x ₁, x ₂..., x _kas sampled data, with x ₁, x ₂..., x _kmean value as self-organization central value.

[formula 3]

$X=\frac{1}{k}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{x}_i$

The x of probability distribution ₁, x ₂..., x _kdispersion value

[formula 4]

$S^2=\frac{1}{k-1}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{(x_i-X)}^2$

The present invention, for the original image of the object of image code, it not the feature directly extracting original image, but will the Feature Mapping of image to geometry or physics aspect, be there is by plural number the image of geometry or physical ones, carry out image feature extraction, and then synthetic image code.

For the original image of the object of image code, the feature of image is passed through the membership function of geometry or physical model, carry out encode, the characteristic information of such image just can be described with the numeral of 0 ~ 9.

Or in above-mentioned transformed geometry or physics image, be divided into multiple field, based on the probability scale of self-organization, for each field, take out respectively and can to use but bit or many bits represent and the value of information of current area be transformed into image code.

Fig. 3 is the schematic diagram being carried out the transform method of unfolded image by geometry or physics.

As shown in the figure, image (a) is the original image of the object of image code, in order to carry out the encode of image, not the original image of the object being directly used as image code, but by original image with geometry or physical mathematical model, object images is transformed into physical energygram picture, edge image, similar degree image, thermodynamic diagram picture, cycle image or geometric shape image etc.

For sake of convenience, this instructions definition " unfolded image " for above-mentioned by original image with geometry or physical mathematical model, the general name of the various images be transformed into.

Fig. 4 is the schematic diagram being extracted out image feature information by the conversion of energygram picture.

As shown in the figure, the original image (a) of the object of the image code of Fig. 4, image (b) after the energy conversion of Fig. 4, (401) in Fig. 4 are as energy peak, when using camera to read image, the brightness value of environment is irradiated according to camera, the energy value of (401) in Fig. 4 may have very large fluctuation, use traditional Euclidean distance as threshold value, be the fluctuation that the difference that can't resolve because of shooting environmental causes the grey value profile of original image, make the unstable result of code building.

In order to improve the problems referred to above, use the self-organized algorithm of the probability scale shown in Fig. 4, can the distribution density of gray scale of image as threshold values, and do not rely on the size of gradation of image, so just can offset the code building instability problem caused due to the fluctuation of image intensity value, the threshold values that therefore the present invention introduces the probability scale obtained by the self-organized algorithm of probability scale is extracted out as the information for image.

Concrete example, P (x _i, y _j) (i=1,2 ..., k, j=1,2 ..., 1) as the function of energygram picture, calculate the average value P of energy according to following formula 5.

[formula 5]

$P=\frac{1}{k1}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{1}{\mathrm{\&Sigma;}}}P(x_i,y_j)$

Power dissipation S is calculated according to following formula ²

[formula 6]

${S_P}^2=\frac{1}{(k1-1)}\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{1}{\mathrm{\&Sigma;}}}{\&lsqb;P(x_i,y_j)-P\&rsqb;}^2$

For unfolded image Fig. 4 (b) image of energygram picture, use above-mentioned self-organized algorithm, the energy threshold in probability scale can be calculated.Extract the pixel that probability scale upper threshold value is larger out, thus obtain profile (402).

The physical meaning of the threshold value of above-mentioned energy probability scale is, the benchmark of the image intensity value of the energy aggregation of more than 70%.The threshold value of energy probability scale is the parameter relevant with the density of the density of image, and the height of the gray scale of image it doesn't matter, therefore do not rely on the reading environment of image.

By artificial intervention, the membership function of a performance energygram picture can be defined.First, energygram as in the image of Fig. 4 (b), P ' (x in profile (402) _i, y _j) (i=1,2 ..., h, j=1,2, ..., g), time the highest according to energy value, the solution of membership function will close to 9, time energy value is minimum, the membership function of energy density close to the subjective requirement of 0, by artificial experience, can will be defined by following formula by the solution of membership function.

[formula 7]

$M_P=\frac{9}{{\left({256}^h\right)}^g}\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{g}{\mathrm{\&Sigma;}}}{P}^{,}(x_i,y_j)$

The meaning of top formula, (256 ^h) ^gequal image P ' (x _i, y _j) in maximal value.P ' (x _i, y _j)=(256 ^h) ^gtime, M _p=9.P ' (x _i, y _j) become less, M _pmore close to 0.

Use the solution of the membership function of energy density, can be used as the characteristic information quantized of the object images of image code, also can become an eigenwert of energygram picture.

With reference to foregoing, when the area that the image outline (402) of Fig. 4 comprises is consistent with all areas of the image (b) of image 4, the solution of membership function will close to 9, when differing maximum, the solution of membership function will close to 0, and the membership function of the energy area features of profile (402) also can define with reference to above-mentioned formula.

Utilize the solution of the membership function of energy area, also can become a characteristic information quantized of the object images of image code, also can become an eigenwert of energygram picture.

Fig. 5 extracts the schematic diagram of image feature information out by the conversion of shape image.

Lower example, as shown in Figure 5, Fig. 5 (a) is the original image of the object of image code, and Fig. 5 (b) is contour images, Fig. 5 (501) is the image of the head portion as people, and Fig. 5 (502) is the result of the self-organized algorithm as probability scale.

For the contour images that Fig. 5 is such, need the circular contour of the head of the image paying attention to people, the whether immediate circle of contour images of head carries out considering.

First, for the contour images that Fig. 5 (b) is such, calculate the length of all connecting lines between all pixels, and center.The feature of round wire is exactly that the pixel of the surrounding of round wire is interconnected the similar length of line, the center of connecting line all can at the near center location of round wire, also have the angle of each connecting line different, the self-organized algorithm of the probability scale utilizing above-mentioned Fig. 2 to introduce, in conjunction with the feature of round wire, the immediate circle of the contouring head image of Fig. 5 just can be calculated.

Then, do a membership function, realize each pixel of contouring head image and round wire 502 major part close time, functional value just becomes " 9 ", and when away from round wire, functional value is close to " 0 ".

Equally, according to the define method of above-mentioned membership function, the membership function of a definition shape image, obtain the result of the membership function of shape image, utilize the solution of the membership function of shape image, also can become a characteristic information quantized of the object images of image code, also can become an eigenwert of shape image.

Fig. 6 is the schematic diagram being extracted out image feature information by the conversion of edge image

The edge image formed according to the grey scale change of image is the important method of 1 of the characteristic information extracting image out.

Under such as shown in Fig. 6, Fig. 6 (a) is the original image of the object of image code, and Fig. 6 (b) is the edge image filtered with linear 1 space differentiation obtained after vertical differential by carrying out level to image.

Same with Fig. 4 (b), the gray-scale value of each pixel of the outline map of Fig. 6 (b), depend on the impact of brightness value that camera irradiates environment, use traditional more stiff threshold value to extract the algorithm of pixel out, for the problem that there is instability image code.

With above-mentioned same disposal route, use and be only correlated with the distribution density of the pixel relevant with the gray-scale value of the edge image of Fig. 6 (b), and the self-organized algorithm of the substantially incoherent probability scale of the gray-scale value of the edge image of same Fig. 6 (b), calculate the threshold value of the edge image paying attention to density probability.

For the edge image of Fig. 6 (b), extract the pixel larger than the edge image threshold value of probability scale out, form edge image E (x _i, y _j) (i=1,2 ... v, j=1,2 ..., w), with the process of above-mentioned energygram picture unlike, the edge image E (x of binaryzation _i, y _j) major part disperses all very much, consider this feature, first, it is an edge pixel set that the pixel larger than the edge image threshold value of probability scale is extracted in order out, the membership function of reference energy density, can define the membership function that belongs to the density of all pixels in edge pixel set.The membership function of reference energy area in addition, can also write out the membership function that belongs to the area of all pixels of edge pixel set.

Fig. 7 is the schematic diagram being extracted out image feature information by the conversion of similar image.

As shown in Figure 7, from the original image (a) of image code object, can obtain four points of domain pictures respectively, (a) is K domain picture, c () is C domain picture, (d) for M domain picture and (e) be Y domain picture.K domain picture and image (c) the i.e. C domain picture of point domain (b) of original image (a) can be calculated, image (d) i.e. M version and the similarity relation of image (e) namely between Y domain picture, thus form 3 similar images, (c) image can also be calculated and C domain picture is the similarity relation between M domain picture with image (d) simultaneously, and (d) image and M domain picture and the similarity relation of image (e) namely between Y domain picture can be calculated, (c) image and the similarity relation between C domain picture and image (e) Y domain picture can also be calculated, thus 3 similar images of having got back.

Example: about the multivalue image F (x of K version _i, y _j) and the multivalue image C (x of C version _i, y _j) (i=1,2 ..., n, j=1,2 ..., similarity relation m), for the purpose of calculating easily, import the concept of Hamming distance, can be calculated by formula 8.

[formula 8]

$S_{\mathrm{KC}}=9[1-\frac{1}{{\left(255\right)}^{\mathrm{nm}}}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}|K(x_i,y_j)-C(x_i,y_j)\left|\right]$

For computed image F (x more accurately _i, y _j) and C (x _i, y _j) (i=1,2 ..., h, j=1,2 ..., similar degree g), the mean value of Hamming distance can be calculated according to formula 9.

[formula 9]

$A=\frac{1}{\mathrm{gh}}\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{g}{\mathrm{\&Sigma;}}}|K(x_i,y_j)-C(x_i,y_j)|$

Identical with above-mentioned, the dispersion value of Hamming distance can be calculated by formula 10.

[formula 10]

${S_{\mathrm{KC}}}^2=\frac{1}{\mathrm{hg}-1}\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{g}{\mathrm{\&Sigma;}}}{\left[\right|K(x_i,y_j)-C(x_i,y_j)|-A]}^2$

The mean value of the Hamming distance of formula 9 is as self-organization central value, and the dispersion value of the Hamming distance of formula 10 can be brought in the algorithm of the self-organization of probability scale, utilizes the last probability scale obtained, can find out the correct F (x of the probability of more than 70% _i, y _j) and C (x _i, y _j) between similar degree, using the characteristic information of the solution of similar degree as similar image, copy the define method of above-mentioned membership function again, the characteristic information of similar image is carried out the process quantized, utilize the solution of the membership function of similar image, also can become a characteristic information quantized of the object images of image code, also can become an eigenwert of similar image.

As shown in Figure 7: an image is divided into the image of CMYK tetra-colors, for each color image, utilize the similar degree between above-mentioned method calculating color image, 6 similar images can be obtained, 6 characteristic informations can be obtained, and the eigenwert of 6 similar images.

Based on above summary of the invention the present invention the 1st embodiment be: for the unfolded image possessing geometry or physics key element, based on the probability scale of self-organization, by artificial intervention, adopt the membership function based on artificial experience, the numerical value of the characteristic information of unfolded image by 0-9 is quantized, by the direct synthetic image code of image feature information quantized.

Be considered as the label of commodity, usually frame line is had, in addition, the read direction of camera is fixing, based on such application, utilizes the feature of the probability scale of above-mentioned self-organization, the present invention proposes by carrying out a plurality of segmentations to unfolded image, respectively the image of every field is carried out to the extraction of image feature information, the process quantized, thus the method for direct synthetic image code.

Fig. 8 is the method for carrying out image code according to energy Iamge Segmentation.

Fig. 8 (a) is energygram picture, and image (b) is image image (a) being divided into (801) (802) (803) (804) (805) (806) (807) (808) 8 fields.

One of method of the present invention the 2nd embodiment is: the threshold value using the probability scale gone out according to self-organization computation, calculates the energygram picture of binaryzation.

Be an image pair by the image in 8 fields of segmentation according to (801) and (802) again, (803) be an image pair with (804), (805) and (806) be an image to and (807) be an image pair with (808).

If (801) pixel value of the image in (802) (803) (804) (805) (806) (807) (808) 8 fields is the individual numerical value of " 1 " is N _a1, N _a2, N _a3, N _a4, N _a1, N _a5, N _a6, N _a7and N _a8.

If (801) N of the image in field and the image in (802) field _a1> N _a2then can extract information " 1 ", on the contrary can extract information " 0 ",

If (803) N of the image in field and the image in (804) field _a3> N _a4then can extract information " 1 ", on the contrary can extract information " 0 ",

If (805) N of the image in field and the image in (806) field _a5> N _a6then can extract information " 1 ", on the contrary can extract information " 0 ",

If (807) N of the image in field and the image in (808) field _a7> N _a8then can extract information " 1 ", on the contrary can extract information " 0 ".

As implied above, a geometric or physical unfolded image, can extract the information of 4 bits out after being divided into 8 fields.

Two of the method for the present invention the 2nd embodiment is: for obtaining more stable image code, first for the image in (801) (802) (803) (804) (805) (806) (807) and (808) 8 fields, respectively based on the self-organized algorithm of probability scale, obtain center of energy value and the probability scale value of energygram picture, the pixel of all energygram pictures within the scope of probability scale is expressed as P ₁(x _i, y _j), P ₂(x _i, y _j), P ₃(x _i, y _j), P _{4 (}x _i, y _j), P ₅(x _i, y _j), P ₆(x _i, y _j), P ₇(x _i, y _j) and P ₈(x _i, y _j), and the energy value that can obtain every field image is, V _p1, V _p2, V _p3, V _p4, V _p5, V _p6, V _p7, and V _p8.

If (801) V of the image in field and the image in (802) field _p1> V _p2then can extract information " 1 ", on the contrary can extract information " 0 ",

If (803) V of the image in field and the image in (804) field _p3> V _p4then can extract information " 1 ", on the contrary can extract information " 0 ",

If (805) V of the image in field and the image in (806) field _p5> V _p6then can extract information " 1 ", on the contrary can extract information " 0 ",

If (807) V of the image in field and the image in (808) field _p7> V _p8then can extract information " 1 ", on the contrary can extract information " 0 ".

Equally, a geometric or physical unfolded image, can extract the information of 4 bits out after being divided into 8 fields.

By the original image of the object of image code, carry out point version and can obtain black and white, R, G, B or C, M, Y, the shades of colour of K, then press energygram picture, edge image, similar degree image, thermodynamic diagram picture, the minimum image conversion that can carry out more than 18 kinds of cycle image etc., therefore can form the image code of more than 72bit.

3rd implementation method of the present invention, using multiple characteristic informations of the image after above-mentioned formation quantizes as proper vector, can Login Register on the server, the characteristic vector space of composing images.The Euclidean distance (Euclidean distance) of each proper vector of the characteristic vector space of the image recognition that the proper vector of the image of the object of computed image encode and server log in, wherein using the image code of proper vector value as present image belonging to characteristic vector space that Euclidean distance is minimum.

4th implementation method of the present invention, in order to more correct carries out image code, by the proper vector of multiple images calculated image same under varying environment study repeatedly, based on the probability scale of self-organization, calculate central value and the probability scale of the proper vector of present image, the central value of proper vector and probability scale all log in into server, form new characteristic vector space.

Time the image logged in new characteristic vector space carries out encode, for central value and the probability scale of the proper vector of each image of characteristic vector space, can using the code value of central value as present image belonging to the proper vector of characteristic vector space minimum for the distance of probability scale.

Define the distance of probability scale here.First set original image as q, the number of unfolded image is p, and i-th original image be set as image feature vector and V by the image feature information after quantizing _i1, V _i2, V _i3, V _i4, V _i5, V _i6, V _i7and V _i8, the image feature vector space so logging in q original image in the server can by following Determinant Expressions.

[formula 11]

V ₁₁，V ₁₂，…，V _1p

V ₂₁，V ₂₂，…，V _2p

V _q1，V _q2，…，V _qp

If set the proper vector of the original image of the object of i-th image code as p _i1, p _i2..., p _iq, the Euclidean distance so between the proper vector of i-th original image and the vector logging in each image of image feature vector space is in the server;

[formula 12]

$E_i={\left(\underset{j=1}{\overset{P}{\mathrm{\&Sigma;}}}{(P_{\mathrm{ij}}-V_{\mathrm{ij}})}^2\right)}^{\frac{1}{2}}$

Establish again log in server there is the proper vector of image, central value and probability scale, new image graph is M as the probability scale in characteristic vector space _ij, (i=1,2 ..., q; J=1,2 ..., p) C _ij(i=1,2 ..., q; J=1,2 ..., p), if the proper vector of the original image of the object of i-th image code is p _i1, p _i2..., p _iqif, so (p _ij-C _ij) absolute value be greater than M _ij, then from (p _ij-C _ij) absolute value in deduct M _ijvalue, otherwise, then allow (p _ij-C _ij)=0.According to such process, reference formula 12, just can obtain the distance of probability scale.

Above-mentioned physical significance is, image code retrieval is carried out for the image logged in the characteristic vector space of image recognition, the proper vector of the image calculated, when retrieving in characteristic vector space, first calculate the Euclidean distance between the vector of each image that logged in current signature vector and characteristic vector space, if time the Euclidean distance value between the eigenwert of characteristic vector space is worth little than probability scale, Euclidean distance is 0, other situations, deduct by Euclidean distance the distance value that whole result that probability scale obtains just can be used as probability scale.

Relevant to characteristics of image is various geometric, or physical image conversion method has a lot.Further, for the image of conversion, determine a lot of various membership function by artificially getting involved.Image feature information quantize not only between 0-9, in addition can consider any scope of such as 0-n, above-mentioned all related contents all belong within scope of the present invention.

Fig. 9 is the adjustment principle schematic of gray scale degree between multiple color image.

As shown in the figure, Fig. 9 (a) is as original image, and Fig. 9 (b) is as by C, M, Y, K tetra-original image of forming of look, and equally, Fig. 9 (d) is as by C, M, Y tri-original image of forming of look.

All grey scale pixel values of the K color image of C, M, Y image three-colo(u)r (d) are all 0.

Here, image (c) is the C domain picture of image (b), image (d) is the M domain picture of image (b), image (e) is the Y domain picture of image (b), and image (f) is the K domain picture of image (b).Image (h) is the C domain picture of image (g) in addition, image (i) is the M domain picture of image (g), image (j) is the Y domain picture of image (g), and image (k) is the K domain picture of image (g).

Foregoing is not only to allow image not have K look, only represent with C, M, Y3 look, and image does not change a lot.Can also when the image quality of original image is off guard perceive, by adjusting the gray-scale value of each color image, between the image realizing shades of colour, gray scale adjusts mutually.

Figure 10 is the image construction principle schematic of anti-counterfeiting image code.

As shown in Figure 10, image (c) is the C domain picture of image (b), image (d) is the M domain picture of image (b), image (e) is the Y domain picture of image (b), and image (f) is the K domain picture of image (b).Image (h) is the C domain picture of image (g) in addition, image (i) is the M domain picture of image (g), image (j) is the Y domain picture of image (g), and image (k) is the K domain picture of image (g).

The original image (a) of Figure 10 is become C, 3 color images (b) of M, Y, the gray-scale value of whole pixels of the K look of 3 color images (b) of C, M, Y is 0.

Use scanner (d) to read 3 color images (b) of above-mentioned C, M, Y, the image of scanning is (g).

It is noted herein that, C, M, the difference of the image (e) that 3 color images (b) of Y and scanning obtain is, when originally not scanning, all pixels of the K version of image (b) are " 0 ", but in the image (e) obtained after scanning, K domain picture just there occurs change.

Usually, time illegal person carries out forging, what scanner must be used to carry out identifying copies, and therefore by judging whether the K domain picture of identification image changes, just can judge to be whether the mark of forgery.

Time Google's glasses read information, whether can change according to above-mentioned K look, judge whether counterfeit.In addition, in the scope that parallax allows, by the adjustment of original image gray scale between shades of colour image, namely adjust C, M, Y, whether the gray scale of arbitrary monochrome image in K, can be realized original image and image code can be made when being copied by scanner to change, be changed by them, judge whether to forge, do not change again the quality of original image simultaneously.

Figure 11 is the schematic diagram utilizing micro lens arrays to realize the image generating method of anti-counterfeiting image code

As shown in the figure, (a) as Google's glasses, (b), as the camera of Google's glasses, (c), as the arrangement of micro lens, (d), as printing images, (e) is as printed medium.

The effect of above-mentioned micro lens is, the direction of accurate control representation printing images.Such as, by forming 3D rendering, be just difficult to forge, Google's glasses and smart mobile phone etc. carry the terminal of camera, also can read.

Figure 12 is the Google's glasses system formation schematic diagram for Internet of Things website

As shown in figure 12, Google's glasses are by network (g), and server (h) UNICOM mutually.Fig. 7 (a) is the camera using Google's glasses, reads Commercial goods labels, carries out the encode conversion of image from the image of Commercial goods labels, then the code value of image is obtained, based on this value, be connected to network and go, just can see relevant merchandise news on the display screen of Google's glasses.In this case, commodity sign will become a new network broadcast media, can receive different Commdity advertisement information or consumer's video frequency program loved by all at any time by Google's glasses.

(b) and (c) of Figure 12 is, use the camera of Google's glasses, read the label of commodity, the encode conversion of image is carried out from the image of Commercial goods labels, then the code value of image is obtained, based on this value, find from network the people reading this label equally, carry out social interactions.This form that will be social networks of new generation, people and thing gather by commodity sign, the fan of commodity can be brought together, carry out social interaction.

Figure 12 (d) is, use the camera of Google's glasses, read the label of commodity, the encode conversion of image is carried out from the image of Commercial goods labels, then the code value of image is obtained, based on this value, can by these commodity of Online Shopping, this is by network sale system new for formation one, as long as the commodity that people often use just can buy this commodity automatically by the mark of the required commodity of Google's glasses direct-view.

Figure 12 (e) is, uses the camera of Google's glasses, reads the label of commodity, carry out the encode conversion of image from the image of Commercial goods labels, then obtain the code value of image, based on this value, by the place of production information of these commodity of Network Capture, sales region etc. information.Consulting merchandise news from now on has not needed the website logging in producer to carry out information search, and the website of producer is just based upon in the mark of commodity.

Figure 12 (f) is, uses the camera of Google's glasses, reads the label of commodity, carries out the encode conversion of image, then obtain the code value of image, just can directly surf the Net based on this value, the information required for retrieval from the image of Commercial goods labels.At this moment commodity sign is as Web portal, enters network by commodity sign, will become general popular Main Means, will change the form of current network retrieval website.

As noted above, Commercial goods labels has just possessed multiple function, the entrance of network, homepage address, netcast media, social networks gather client etc.

The Google glasses system schematic flow sheet of Figure 13 centered by article

As shown in the figure, the Google's glasses system flow process centered by article is made up of following step.

Step 1, the tag read steps of commodity.Mainly use the camera of Google's glasses, the label of commodity is read.

Step 2, the code steps of recognition image.The black white image of label image, R, G, B or C, in each color image of M, Y, K, more than one original image, is transformed into unfolded image by multiple spatial mappings, then obtains the numerical value of the 0-n quantized by membership function, and synthetic image code.

Step 3, function automatic switchover step.The differences such as Google's glasses and smart mobile phone, do not have the button operated, but sometimes need the automatic switchover of function.So, the present invention, by identifying the action of human eye and the action etc. of eyelid, as function information, the switching of automatic practical function.Or Google's eyes load a gyro accelerometer sensor, identify the acceleration of headwork, as function information, the automatic switchover of practical function.Whole idea is exactly by reading once, as function information, and the automatic switchover of practical function.Also can consider that the knee-action of user's tooth causes the action of jawbone, produce controlling functions handover information.

Step 4, function treatment step.According to the function determined, concrete carrying out processes, and is social networks, or the advertisement of current commodity, or the network selling of current commodity, or the related information of current commodity etc.

The automatic switchover step of above-mentioned steps 3, can be placed in initial step 1 or other step and realize.

The problem that image changes of code into important is, the stability of code value.Therefore, the present invention is for the image that will carry out encode, use above-mentioned probability scale self-organized algorithm, calculate the threshold value of energygram picture, for the threshold value of edge image, the threshold value of similar image, the threshold value of thermodynamic diagram picture, threshold value of cycle image etc., the result calculated has to be offset by the very large effect of the overall variation of inconsistent the caused image intensity value of the shooting environmental of image.

[embodiment]

Figure 14 is the functional area figure of the image indexing system 20 that an example of the present invention is correlated with.Have in the terminal device 19 of image code mapping function, have and have with the mobile terminal 10 of Fig. 1 the number that the part of same function adopts and Fig. 1 is same.

Terminal device 19 has the function same with function illustrated in fig. 1.Image reading section 11 comes in image reading, then shows in image displaying part 12.When carrying out image code process, and the instruction of terminal control part, the image portrait read is sent to image conversion part 16.

Image code conversion fraction 16, has image conversion part 25, feature extraction section 26, and the part such as image code generating portion 27.By image conversion part 25, utilize image-reading device, obtain the intensity profile of each pixel of image information, based on geometry or physical key element, be transformed to multiple unfolded image information.Feature extraction section 26, from the unfolded image information of top, based on the probability scale of self-organization, extracts various characteristic information out, then quantizes.Finally, the numerical value of the characteristic information of each unfolded image that image code generating portion 27 utilizes above-mentioned feature extraction section 26 to obtain, synthetic image code.

Like this, the image conversion part 25 of image code conversion fraction 16, the image information needing to carry out image code, the black and white of such as image information, RGB, or the assorted image information of CMYK, project to energy space, differential space, shape space, similar degree space, thermodynamics space, cycle space etc., be transformed to energygram picture, differential map picture, shape image, similar degree image, thermodynamic diagram picture, cycle image etc.

For the shades of colour figure of above-mentioned image, project to the result in multiple space, the unfolded image information of more than 18 kinds can be obtained.All it doesn't matter for the quantity of unfolded image information, and when many, the capacity of image code is just larger, calculates and needs to spend the long time.Here, need to consider practical image code capacity and computing velocity.

The present invention is as the 1st implementation method, the feature extraction section 26 of image code conversion fraction 16, from the unfolded image information of more than 18 kinds, use the probability scale of self-organization and fixing multiple membership functions, extract out for the relative energygram picture of the image of each color, differential map picture, shape image, similar degree image, thermodynamic diagram picture, the characteristic information of the unfolded image that cycle image etc. is more than 18 kinds, quantizes with the numeral of 0-n.

Image code generating portion 27, more than the 18 kinds image feature informations obtained by the extraction section 26 of above-mentioned conversion fraction 16, utilize the characteristic information quantized of all unfolded images, directly generate 1 image code.

As the 2nd implementation method of the present invention, the feature extraction section 26 of above-mentioned conversion fraction 16, the energygram picture that the image for each color is relative, differential map picture, shape image, similar degree image, thermodynamic diagram picture, the unfolded image of cycle image etc. more than 18 kinds, is divided into multiple field.Based on the probability scale of self-organization, for the every field of segmentation, extract the characteristic information of wherein unfolded image out, as 1bit or many bit information, carry out digitizing.

Above-mentioned image code generating portion, uses the numerical value of the characteristic information of the 18 kinds of unfolded images extracted out, the code value of direct synthetic image.

According to the image code generated in implementation method in the 1st, the another kind of method obtained more accurately is exactly the 3rd implementation method of the present invention.3rd implementation method of the present invention, by the feature extraction section 26 of image code conversion fraction 16, obtain the characteristic information of each image quantized, the proper vector of these characteristic informations as image is logged on the server, the characteristic vector space of composing images.

Euclidean distance between each proper vector in the characteristic vector space of the proper vector of the image of the object of computed image encode and the image of login, according to nearest proper vector, generates the code value of present image.

Compare with above-mentioned 3rd implementation method, what image code was evolved more is the 4th implementation method of the present invention.Use the characteristic information scalarization method of the unfolded image of the feature extraction section 26 of the image code conversion fraction 16 of above-mentioned 1st implementation method, in the environment that irradiating angle of brightness of illumination and camera etc. is different, for the same image of Login Register, through 10 study images of the example that study (example is the reading of camera 10 times) repeatedly obtains, for the characteristic information of the unfolded image of more than 18 of these images, based on the probability scale of self-organization, calculate central value and the dispersion value of 1 proper vector of present image.Then central value and dispersion value are logged in into server, build a new vector space.

Time the image logged in new characteristic vector space carries out encode, for central value and the dispersion value of the proper vector of each image in characteristic vector space, the central value of the proper vector in characteristic vector space minimum for the distance of probability scale, as the image code of present image.

The control information sent from communications portion 17 and image code, through repeater 21, via network, send to central computer 23.The control information that central computer 23 sends based on terminal device 19, the code dependent information of retrieving images, obtains photographed image-related information, delivers letters to terminal device 19.Do not need to send image code to central computer 23 from terminal device 19, by terminal device 19, based on image code data, direct access data services device 24, gets the data message that image is relevant.

As shown in figure 14, terminal device 19, not only by radio communication, also connects repeater 21 by loop and carries out communication.

Further, in the example of Figure 14, direct configuration image code conversion part 16 is on terminal device 19.In fact, image code conversion fraction 16 also can be configured on repeater 21 or central computer 23 etc., also can be configured in the outside of terminal device 19.But now, in the communication before image code conversion, must send image information, therefore, it is additional that code conversion can increase communication.

The center of proper vector is propped up and dispersion value, and the proper vector as image carries out Login Register, in the image code conversion fraction 16 of terminal device 19, detects the central value of the proper vector of present image, synthetic image code.Then delivered letters by network, based on the information that image code retrieval is relevant.

Further, based on the image code calculated through image code part, or the central value of proper vector, by network, the server that visit data stores, can download Commdity advertisement animation, merchandise news, social network information, network selling information etc.

Data input unit is can perception human eye action by input, and tooth action, the sensor information of the action of lip, realizes the operation of terminal.

By the signal of sound import identification, based on the result of voice recognition, realize the operation of terminal.

By input gyro signal, the action acceleration of human body is utilized to operate terminal.

Voice output importation, sound guidance, the functions such as music.Additionally by lift-launch GPS, the guiding of cartographic information can be realized, on-site display of equipment etc.Further, by loading various chip, various function is realized.

[utilizing possibility in industry]

Use the means of above-mentioned image code, compare with traditional Commercial goods labels, directly can log in these commodity on network, the label of all commodity can as Web portal, consumer uses, and smart mobile phone or Google's glasses etc., read the label of commodity, just can judge the true and false of commodity, understand the place of production of commodity, the purchasing method of commodity, the advertisement of using method etc. merchandise news and commodity, commodity related personnel network social intercourse information, network selling information of commodity etc.

In addition, by carrying out encode to the image of the human body informations such as record face, just can replace credit card, adopting human body codes implement to do shopping without card.

Above-mentioned, by carrying out encode to the image of human body information such as record face etc., also can realize automatic ticket checking system without card system.

Advantageous effect of the present invention is:

According to image code generation method of the present invention, from the image that image-reading device reads, the printed article of such as Commercial goods labels or packaging, landscape or other image, can be simple and quick be transformed to unique code.

Such as, use image code equipment of the present invention, various image conversion is become distinctive code, the information that this image is associated and peculiar code carry out network associate registration, by this method, the quick-searching of the photographed image-related information that image-reading device obtains can just be realized.

In addition, according to the present invention, from the printed article image information of Commercial goods labels and packaging, the distinctive unique code of image can be obtained, as Quick Response Code, the original picture quality of image can not be destroyed because of extracode, also can realize the quick-searching of information.Meanwhile, do not need in advance to imbedding stealthy code inside printed article yet.

According to the present invention, the image read can be transformed to tens of bits to the distinctive unique code of hundreds of bit.Therefore, compare with images match AR technology, the quantity of information that communication adds is very little, can realize retrieval at a high speed.All images can be transformed to the distinctive unique code of image about hundreds of bit, search a large amount of data such as the match information not needing to log in view data or image in order to what carry out image, only need the capacity of very little logon data just can realize proper retrieval.

Such as, read label or the packaging of displaying merchandise by image-reading device, be transformed into image code.According to the code value of image code as retrieval coding, the place of production of commodity, the various information such as commodity selling price log in, and just can provide the additional information of product.

Image code generation method of the present invention is well suited for building and uses Google's glasses, and the mobile image indexing system of smart mobile phone or portable phone etc. equipment and additional information provide system.Simultaneously, at use Google glasses, when smart mobile phones etc. carry out image retrieval, all images have all carried out encode, when such visit information, will the traffic be reduced, improve network communication efficiency, and image codeization can carry out dispersion treatment, the retrieval of the image of a large amount of wide area can be carried out simultaneously.

Table 1

Table 1 is exactly the contrast of image code technology of the present invention and ITC (Image To Code) technology and AR technology.About the capacity that the characteristic information of a registration view data is necessary, AR technology needs several Mbytes, and the present invention only needs tens of Bytes.

In addition, another one feature of the present invention have exactly prevent forge function.AR technology is for black and white picture, and extract profile diagram out, common duplicating machine just can be forged, and not possesses antiforge function.

Further, if picture shape too simple simon says, AR technology is easy to identify by mistake.The present invention understands the original color information of reference picture, there is the ability of decomposing image information, figure code technology of comparing with original AR technology can corresponding great amount of images, only need little image feature information amount, just effectively can carry out the network processes of wide area, be suitable for the Internet of things system globalized.

Claims (7)

1., towards a generation method for the image code of Google's glasses, be by image conversion part, characteristic information extraction section, image code generating portion composition, there is following feature:

The original image that image conversion part will get from image-reading device, is transformed into a plurality of geometry, or the unfolded image of physics form;

The image of a plurality of expansion that feature extraction section will be produced by image conversion part, based on the probability scale of self-organization, extracts the characteristic information of each image out;

Image code generating portion, by being extracted out the characteristic information of each image by feature extraction section, carries out the process quantized, synthetic image code.

2. the generation method of the image code towards Google's glasses according to claim 1, is characterized in that:

Above-mentioned image conversion part, described image conversion is the distribution of the gray scale of each pixel according to original image information, original image is transformed into the unfolded image with geometric shape facility, there is the unfolded image of physical energy feature, edge unfolded image, in the unfolded image of similar degree, plural number plants unfolded image.

3. the generation method of the image code towards Google's glasses according to claim 1, is characterized in that:

Above-mentioned characteristic information extraction section, described self-organization probability scale refers to, comprises normal distribution, and multivariate normal distributes, exponential distribution, traffic distribution, Weibull distribution, triangle distribution, in beta distribution, at least one has the parameter of the probability attribute of probability distribution, the central value of described self-organization, refers to the mean value of probability distribution, or expected value.

4. the generation method of the image code towards Google's glasses according to claim 1, is characterized in that:

Above-mentioned feature extraction section, is divided into multiple image-region respectively by above-mentioned plural unfolded image, based on self-organization probability scale, from the field of each Iamge Segmentation, extracts the characteristic information of unfolded image out;

Above-mentioned image code generating portion, by the characteristic information of plural unfolded image, by single-bit, or the form of many bits carries out the process that quantizes, and direct synthetic image code.

5. the generation method of the image code towards Google's glasses according to claim 1, is characterized in that:

Above-mentioned feature extraction section, by a plurality of unfolded image, based on the probability scale of self-organization, extracts the characteristic information of each unfolded image out;

Above-mentioned image code generating portion, by the characteristic information of all unfolded images extracted out by feature extraction section, uses a plurality of membership functions defined based on artificial experience, quantizes between 0 to n numerical value, and direct synthetic image code.

6. the generation method of the image code towards Google's glasses according to claim 1, is characterized in that:

Above-mentioned image code generating portion, by the characteristic information of the unfolded image of each original image quantized, as the proper vector of multiple image, and for Login Register to server constituting the characteristic vector space of image;

Calculate and belong to the Euclidean distance (Euclidean distance) of each proper vector logged in the proper vector of the image of characteristic vector space and the characteristic vector space of image, using the proper vector apart from minimum characteristic vector space as the image code of present image.

7. the generation method of the image code towards Google's glasses according to claim 1, is characterized in that:

Above-mentioned Code Generation section, in order to generate the image code logged in, with multiple images of the same reading object got different opportunitys, obtain the characteristic information of each fixing multiple unfolded images, as the proper vector of multiple image information, based on self-organization probability scale, calculate central value and the probability scale of the proper vector of present image, form the new characteristic vector space of server log;

For the image feature vector belonging to new characteristic vector space, the central value of the proper vector of each image logged in characteristic vector space and dispersion value is used to carry out the calculating of the distance of the distance probability scale of probability scale, using the nearest image code of central value as present image belonging to the proper vector of characteristic vector space of probability scale.