CN106874942A - A kind of object module fast construction method semantic based on regular expression - Google Patents

Abstract

The invention discloses a kind of object module fast construction method semantic based on regular expression, belong to machine vision and area of pattern recognition.Image where identified object is pre-processed first, to improve the quality of feature extraction, the local feature of image object is extracted thereafter by Harris feature detection algorithms, regular expressions finally by amplification definition are semantic, and the object matching model of identified thing is depicted with reference to the picture material storehouse for defining.The method " regularization theory " theoretic to traditional computer has carried out amplification definition, make it possible to be adapted to images steganalysis field, its recall precision characteristic high is remained, has extraordinary effect in the case where the physical characteristics such as identified thing profile, color are constant.

Description

A kind of object module fast construction method semantic based on regular expression

Technical field

The present invention relates to machine vision technique and mode identification technology, and in particular to a kind of image based on regular expression Target identification method

Background technology

It is exactly to be processed and analyzed to characterizing the various forms of information that obtain of thing in image to the identification of image object. The process for being described to object, recognized, classifying and being explained.Target identification can be divided into has classifying for supervision to divide with unsupervised Two kinds of class, both main difference is that, whether the classification belonging to each experiment sample previously known.Supervised classification is also known as training point Class method, its principle is to be gone to recognize the process of other unknown classification pixels with the sample pixel for being identified classification.Led to before classification The visually means such as judgement are crossed, has priori to the objective attribute target attribute on image.In general the classification for having supervision is generally required The sample of a large amount of known class is provided, according to the sample that known training center is provided, by selecting characteristic parameter, characteristic parameter is obtained Used as decision rule, the image classification for setting up discriminant function to be carried out to each image to be sorted is a kind of method of pattern-recognition. It is required that training region has typicalness and representativeness.By summary for many years, people design multiple supervised training features and carry The method for taking, summary gets up to be divided into four classes：Bottom global characteristics, bottom local feature, middle level features and attributive character.It is early The feature extractions such as Tamura textures, color histogram, Harris operators, intensive sampling SIFT, Texton that the phase proposes are mainly Bottom global characteristics and bottom local feature.What global characteristics were extracted is the information such as global color, texture, the shape of image, right The development of the image indexing system that the image understanding task of promotion early stage is based particularly on content has played very big impetus.With Going deep into for research, it has been found that overall Vision feature can not fully meet image classification task and lifting nicety of grading is wanted Ask, and the local feature with SIFT feature as representative has stronger descriptive power such that it is able to Man Zheng.In recent years due to part Feature is increasingly perfect to vision descriptive power, and the research of local visual feature starts to turn to how to ensure the premise of descriptive power The lower extraction efficiency for improving local feature.At the same time, set up and vision content mechanism information is described on the basis of local feature Middle level features also begin to obtain the attention of academia, and the attributive character for combining figure supervision message in itself is also studied as another Focus.

The content of the invention

The purpose of the present invention be to image in identified thing be identified, set up matching template during object matching.For This proposes a kind of object module fast construction method semantic based on regular expression.

The technical solution adopted by the present invention is：

A kind of object module fast construction method semantic based on regular expression, comprises the following steps：

Step 1：Target identification thing position image is obtained, image preprocessing is then carried out：IMAQ point is chosen, is obtained Take the very big constant physical features of identified thing；Step 2：Target's feature-extraction, including color characteristic, locus feature, with And the extraction and selection of Corner Feature；Step 3：The picture element database of image is set up, and amplification definition is carried out to regular expression：It is first First, the physical characteristic according to general object sets up the picture element database that can describe object features, mainly includes lines storehouse, shape Storehouse, color libraries, spatial positional information storehouse；These pixel elements are organized followed by canonical semanteme, is assigned them and is described thing The ability of body characteristicses；Step 4：Object matching model is depicted using the regular grammar of amplification definition.

Further, the step 1 is that under special scenes, selection obtains target image from positive side, so can guarantee that and obtains The same class target object got has maximum physical features similitude；Image filtering and edge strengthening：Image is entered first The treatment of row gray processing, is then filtered using medium filtering to ambient noise, and side is carried out to image finally by canny operators Edge is strengthened.

Further, the step 1 also includes：

1) Mesophyticum of each point value is replaced in a neighborhood the value of any in image sequence with the point, allows the pixel of surrounding The close actual value of value：With the two-dimentional sleiding form of square square structure, pixel in plate is ranked up according to the size of pixel value, Generation monotone increasing (or decline) is 2-D data sequence；Two dimension median filter is output as g (x, y)=med { f (x-k, y- L), (k, l ∈ W) }, wherein, f (x, y), g (x, y) is respectively image after original image and treatment, and W is two dimension pattern plate, is chosen to be It is 3*3 regions；

2) brightness step higher is relatively likely to be edge, but the definite value of neither one to limit great brightness Gradient is much edges, and the effective gradient scope of target is obtained by having used hysteresis threshold, dynamic regulation in Canny operators； Assuming that the important edges in image are all continuous curves, the part obscured in given curve can be thus tracked, and keep away Exempt from for the noise pixel without constituent curve to treat as edge；Since a larger threshold value, this will identify and compares what is firmly believed True edge, whole edge is tracked since these real edges in the picture；It is smaller using one when tracking Threshold value, thus can be with the blurred portions of aircraft pursuit course until returning to starting point.

Further, the step 2 is specifically included：

Color feature extracted：Single order, second order and third moment are extracted to each color passage to count, if h_ijRepresent i-th The probability that gray scale occurs for the pixel of j in individual Color Channel component, n is total number-of-pixels, then the three of color moment low-order moment number Learning expression formula is：

This 3 low-order moments are referred to as average, variance and degree of skewness；

Space characteristics are extracted：In order to improve the description one after another to positional information, when characteristic vector positional information is calculated, adopt Calculated with D4 models：

D₄(P, Q)=| x_p-x_q|+|y_p-y_q|

D4 distances i.e. city block distance, it only selects both direction anyhow to calculate relative distance；

Corner Feature based on Harris operators is extracted：The detection of Harris angles is found on image by mathematical computations A kind of algorithm of angle character, and it has the speciality of rotational invariance；Setting up the feature regular expression of images match Before, Harris angles are first passed through and detect characteristics of image " Character table ", mathematical principle is as follows：

Wherein W (x, y) represents moving window, and I (x, y) represents grey scale pixel value intensity, and scope is 0~255；According to Taylor Series calculates single order to the partial derivative of N ranks, finally gives a Harris Matrix Formula：

Matrix computations matrix exgenvalue λ according to Harris₁,λ₂, then calculate Harris angular response values：

R=detM-K (traceM)²

DetM=λ₁λ₂

TraceM=λ₁+λ₂

Wherein K be coefficient value, usual span be 0.04~0.06 between.

Further, in the step 3,

The lines storehouse is included by linear division：Horizontal linear, oblique line, right-angle line, camber line, S-shaped camber line, and other are certainly Define linear；The shape library includes square, and rectangle is circular, and semicircle, rhombus is heart-shaped, and other are customized non- Conventional pattern；The color libraries are indicated by way of character and digit variable Wk, wherein first letter w represents colour system, after Face numerical variable k represents brightness, and span is between 0-255；

The positional information storehouse describes the relative position information in image by complete customized symbol, " | (x1)-> (x2) | " level terminates to the right to pixel X2 always since pixel x1 positions for representative；" | (x1)-^ (k) (x2) | | " is represented from picture First x1 starts at the k of its vertical lower pixel x2, and wherein K is variable coefficient, can take numerical value between 0-1.

Further, the basis definition of the horizontal and vertical position information for being given in the step 3, the delimiter of positional information Number and the visual actual conditions self-defining of definition rule and addition；This place it is described as element database be an open element Storehouse, the technical indicator for describing physical characteristic can be as a base values in pixel element storehouse, such as angle information, temperature Information, timbre information, vibration information can be extended as an element database index item of description target.

Further, in the step 4, connective connection is carried out to targeted peripheral characteristic point first, then in connection segment The internal cutting that pixel element is carried out according to maximum similarity principle.Local feature after cutting can find in as element database Most like " pixel element ".

The beneficial effects of the invention are as follows：

The conventional method that object matching template is obtained is completed by substantial amounts of supervised training.In general there is supervision Classification generally requires to provide the sample of a large amount of known class, according to the sample that known training center is provided, by selecting characteristic parameter, Characteristic parameter is obtained as decision rule, the image classification for setting up discriminant function to be carried out to each image to be sorted, this process Need to consume substantial amounts of computing resource, flexibility is not high, train the sample for coming also not necessarily preferable.By amplification in the present invention " the canonical algorithm " of definition, with reference to the pixel element material database for pre-defining, can quick definition go out the matching mould of identified thing Type, while determination methods can in time adjust model by visual observation, with very strong flexibility, exports, production in high speed The matching for constructing identified thing that can be quickly under the little application scenarios of the identified thing change in physical properties such as line conveyer belt Model, can be greatly improved the efficiency of target identification.

Brief description of the drawings

The present invention is described in further details with reference to the accompanying drawings and detailed description：

Fig. 1 is target identification flow chart.

Fig. 2 is the design sketch after certain image preprocessing.

Fig. 3 is that certain characteristics of image angle point chooses icon example.

Fig. 4 is according to the pixel element extracted as the definition in element database.

Fig. 5 is that object matching model is depicted using the regular grammar of amplification definition.

Specific embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Site preparation is described.

Fig. 1 is target identification flow chart, and wherein image preprocessing purpose is filtering environment noise spot, strengthens extract wheel Wide feature；Feature extraction includes the color feature extracted based on color moment, the locus feature extraction based on constant geometry, shape Shape feature extraction；Shape Feature Extraction uses Harris Corner Feature extraction algorithms, in the characteristic angle point set chosen Selection can most represent the angle point of identified thing geometric properties, finally by regular expressions semantic rules, using the basis of pixel element Material database is depicted and the immediate object matching model of target signature.

Step 1：Target identification thing position image is obtained, image preprocessing is then carried out

(1) IMAQ point is chosen, the very big constant physical features of identified thing are obtained.

Specific method is under special scenes, such as to export at a high speed, and selection obtains target image from positive side, can so protect The same class target object that card gets has maximum physical features similitude, so as to improve Matching Model, the match is successful Rate.

(2) image filtering and edge strengthening

Gray processing treatment is carried out to image first, ambient noise is filtered using medium filtering then, finally by Canny operators carry out edge strengthening to image, so as to improve the quality of Edge Gradient Feature.

Morphology processing is further used, is specifically included：

1) Mesophyticum of each point value is replaced in a neighborhood the value of any in image sequence with the point, allows the pixel of surrounding The close actual value of value, so as to eliminate isolated noise spot.Method is the two-dimentional sleiding form with square square structure, by picture in plate Element is ranked up according to the size of pixel value, and generation monotone increasing (or decline) is 2-D data sequence.Two dimension median filter G (x, y)=med { f (x-k, y-l), (k, l ∈ W) } is output as, wherein, f (x, y), g (x, y) is respectively original image and treatment Image afterwards.W is two dimension pattern plate, and it is 3*3 regions to be chosen to be.

2) brightness step higher is relatively likely to be edge, but the definite value of neither one to limit great brightness Gradient is much edges, by having used hysteresis threshold in Canny operators --- high threshold and Low threshold, can be obtained with dynamic regulation Take the effective gradient scope of target.Assuming that the important edges in image are all continuous curves, given song can be thus tracked The part obscured in line, and avoid for the noise pixel without constituent curve treating as edge.So from a larger threshold value Start, this will identify the true edge for comparing and firmly believing, can in the picture be tracked since these real edges whole Edge.When tracking, using a less threshold value, thus can be with the blurred portions of aircraft pursuit course until having returned to Point.Fig. 2 is by obtaining contour edge after rim detection and filtering, so as to strengthen to these contour edges.

Step 2：The extraction and choosing of target's feature-extraction, including color characteristic, locus feature, and Corner Feature Take.(1) color feature extracted

Because colouring information is concentrated mainly in low-order moment, therefore need to only single order, second order and three be extracted to each color passage Rank square is counted.If h_ijGray scale is the probability of the pixel appearance of j in representing i-th Color Channel component, and n is total pixel number Mesh, then the three of color moment low-order moment mathematic(al) representation be：

This 3 low-order moments are referred to as average, variance and degree of skewness.

(2) space characteristics are extracted

In order to improve the description one after another to positional information, when characteristic vector positional information is calculated, carried out using D4 models Calculate：

D₄(P, Q)=| x_p-x_q|+|y_p-y_q| (4)

D4 distances i.e. city block distance, it only selects both direction anyhow to calculate relative distance, on positioning describing more Plus it is easy.

(3) Corner Feature based on Harris operators is extracted

The detection of Harris angles is a kind of algorithm for finding angle character on image by mathematical computations, and it has The speciality of rotational invariance.Before the feature regular expression for setting up images match, first pass through Harris angles and detect figure As feature " Character table ", mathematical principle is as follows：

Wherein W (x, y) represents moving window, and I (x, y) represents grey scale pixel value intensity, and scope is 0~255.According to Taylor Series calculates single order to the partial derivative of N ranks, finally gives a Harris Matrix Formula：

Matrix computations matrix exgenvalue λ according to Harris₁,λ₂, then calculate Harris angular response values：

Wherein K be coefficient value, usual span be 0.04~0.06 between.Fig. 3 is the characteristic point for extracting.

Step 3：That sets up image carries out amplification definition as element database and to regular expression

First, the physical characteristic according to general object sets up the picture element database that can describe object features, mainly including line Bar storehouse, shape library, color libraries, spatial positional information storehouse：Then these pixel elements can be organized using canonical semanteme, is assigned The ability for giving them to describe object features.

Wherein lines storehouse can be divided into as linear：The routine lines such as horizontal linear, oblique line, right-angle line, camber line, S-shaped camber line And other are self-defined linear etc., shape library is included, square, and rectangle is circular, semicircle, rhombus, the conventional shape such as heart Shape, and other customized unconventional figures.Color libraries are indicated by way of character and digit, such as:Wk, wherein first Letter w represents colour system, behind numeral k represent brightness, span is between 0-255.Positional information storehouse is by completely customized Symbol describes the relative position information in image, such as " | (x1)->(x2) | " represent since pixel x1 positions always level to The right side is terminated to pixel X2, and " | (x1)-^ (k) (x2) | | " is then represented since the pixel x1 at the k of its vertical lower pixel x2, Wherein K is variable coefficient, can take numerical value between 0-1, is such as taken between 0.25 expression x1 to x2 at a quarter of distance, herein It has been merely given as the basis definition of horizontal and vertical position information, the definition symbol and the visual actual feelings of definition rule of positional information Condition self-defining and addition.This place it is described as element database be an open element database, for describing physical characteristic Technical indicator can be as a base values in pixel element storehouse, such as angle information, temperature information, timbre information, vibration letter Breath etc. can be extended as an element database index item of description target.

Table 1 is as element database

Then these pixel elements can be organized using canonical semanteme, assigns the ability that they describe object features. Regular expression, also known as regular expression, a concept of computer science.Canonical table is usually used to retrieval, replaces those Meet the text of certain pattern (rule).Regular expression is a kind of logical formula to string operation, is exactly to be determined with prior The group of some good specific characters of justice and these specific characters, conjunction, constitute one " regular character string ", this " regular character String " is used for expressing a kind of filter logic to character string.The core concept of regular expression is to carry out abstract to being described thing And classification.A given regular expression and another character string, we can reach following purpose：

1. whether given character string meets the filter logic (referred to as " matching ") of regular expression；

2. the specific part that we want can be obtained from character string by regular expression；

The traditional regular grammar of table 2

It is for describing the part of characteristics of image " pixel element ", the plain language according to regular expression of these pixels in table 1 Method is organized, and can quickly define the local feature of image.Such as：|●->● | | the expression formula can represent image The circular part of two black in middle matching horizontal direction.Can certainly be used as defining text regular expression " the pixel element " of more succinct character representative image canonical.

Step 4：Object matching model is depicted using the regular grammar of amplification definition：

Selection is best able to describe the characteristic point of characteristics of image on the basis of Fig. 3, is extracted with reference to as the definition in element database Pixel element, such as Fig. 4.Connective connection is carried out to targeted peripheral characteristic point first, then according to maximum phase inside connection segment The cutting of pixel element is carried out like degree principle.Local feature after cutting can find most like " pixel in as element database Element ".Shape pixel can be extracted in Fig. 4 after dicing：Trapezoidal, rectangle, circle.The color pixel for extracting is：w12 (white No. 12), B10 (black 10), the position image element information for extracting be " |-^ | | " (from upper position to lower position), " |-^ (0.25) | | " (at from upper position to a quarter of lower position), " |-^ | (0.75) | " are (from upper position under At 3/4ths of side position).

The pixel element such as following table for being extracted：

The basic pixel element of table 3

Based on the pixel element in table 3, the object matching regular expressions based on pixel element can be constructed:

Expression formula carries out the description of object module according to priority from left to right.One description of content representation in " { } " Overall, i.e. the description of wherein positional information is relative to for the first pixel element in bracket.Description in above formula can be with It is the trapezoidal beginning of w12 to be interpreted as with a color number, and this trapezoidal vertical lower connects the color lump element that a color number is B10, Vertical direction is a Description of " { } " form below, and the content in Description is the rectangle that a color W12 starts, at it There is the circle that a color number is B10 at the 1/4 of lower position, have a circle of color B10 again at following 3/4.It is retouched The graphic feature stated is as shown in Figure 5.

Preferred embodiment：

An optimal specific embodiment of the invention：Image collecting device is set in the complete face frame of target object, is got The greatest physical characteristic of image, is strengthened object edge and noise reduction pretreatment, so by medium filtering and canny operators The acquisition quality of characteristic point can be improved, the edge angle point in image is detected followed by harris operators, selection most can The angle point of the identified thing physical features of sign, colouring information is proposed by color moment, and D4 operators calculate image relative position information. Finally using establish as element database by amplify the regular expression semantic description of definition go out closest to identification target With model.

Regular expression based on character unit is high because of its matching efficiency, and suitability is good, is obtained in text character retrieval It is widely applied, although there is larger difference in matching content in images match and character match, can on matching process To find it, concomitant, i.e., the method that Matching Model can be built by regular organization foundation element, is carried out to target Identification.It is concomitant based on this point, the semanteme of regular expression has been carried out into amplification definition by assigning graphic attribute feature herein, The new ideas and semantic rules of " pixel element " are introduced, experiment shows, by writing images match regular expression, is fully able to Quickly, Image Feature Matching model is easily set up, the purpose of object matching is realized.

In the description of this specification, reference term " one embodiment ", " some embodiments ", " illustrative examples ", The description of " example ", " specific example " or " some examples " etc. means to combine specific features, the knot that the embodiment or example are described Structure, material or feature are contained at least one embodiment of the invention or example.In this manual, to above-mentioned term Schematic representation is not necessarily referring to identical embodiment or example.And, the specific features of description, structure, material or spy Point can in an appropriate manner be combined in one or more any embodiments or example.

Although an embodiment of the present invention has been shown and described, it will be understood by those skilled in the art that：Not Can these embodiments be carried out with various changes, modification, replacement and modification in the case of departing from principle of the invention and objective, this The scope of invention is limited by claim and its equivalent.

Claims (7)

1. a kind of object module fast construction method semantic based on regular expression, it is characterised in that comprise the following steps：

Step 1：Target identification thing position image is obtained, image preprocessing is then carried out：IMAQ point is chosen, quilt is obtained Recognize the very big constant physical features of thing；Step 2：Target's feature-extraction, including color characteristic, locus feature, and angle The extraction and selection of point feature；Step 3：The picture element database of image is set up, and amplification definition is carried out to regular expression：First, Physical characteristic according to general object sets up the picture element database that can describe object features, mainly includes lines storehouse, shape library, face Color storehouse, spatial positional information storehouse；These pixel elements are organized followed by canonical semanteme, is assigned them and is described object features Ability；Step 4：Object matching model is depicted using the regular grammar of amplification definition.

2. a kind of object module fast construction method semantic based on regular expression according to claim 1, its feature It is that the step 1 is that under special scenes, selection obtains target image from positive side, so can guarantee that get same Class target object has maximum physical features similitude；Image filtering and edge strengthening：Image is carried out at gray processing first Reason, is then filtered using medium filtering to ambient noise, and edge strengthening is carried out to image finally by canny operators.

3. a kind of object module fast construction method semantic based on regular expression according to claim 1, its feature It is that the step 1 also includes：

1) Mesophyticum of each point value is replaced in a neighborhood the value of any in image sequence with the point, allows the pixel value of surrounding to connect Near actual value：With the two-dimentional sleiding form of square square structure, pixel in plate is ranked up according to the size of pixel value, generated Monotone increasing (or decline) is 2-D data sequence；Two dimension median filter be output as g (x, y)=med f (x-k, y-l), (k, L ∈ W) }, wherein, f (x, y), g (x, y) is respectively image after original image and treatment, and W is two dimension pattern plate, and it is 3*3 areas to be chosen to be Domain；

2) brightness step higher is relatively likely to be edge, but the definite value of neither one to limit great brightness step It is much edges, the effective gradient scope of target is obtained by having used hysteresis threshold, dynamic regulation in Canny operators；Assuming that Important edges in image are all continuous curves, can thus track the part obscured in given curve, and avoid by The noise pixel for not having constituent curve treats as edge；Since a larger threshold value, this will identify and compares firmly believe true Edge, whole edge is tracked since these real edges in the picture；When tracking, a less threshold is used Value, thus can be with the blurred portions of aircraft pursuit course until returning to starting point.

4. a kind of object module fast construction method semantic based on regular expression according to claim 1, its feature It is that the step 2 is specifically included：

Color feature extracted：Single order, second order and third moment are extracted to each color passage to count, if h_iJ represents i-th face The probability that gray scale occurs for the pixel of j in the component of chrominance channel, n is total number-of-pixels, then the three of color moment low-order moment mathematical table It is up to formula：

${\mathrm{\μ}}_i=\frac{1}{n}{\mathrm{\Σ}}_{j=1}^n{h}_{ij}$

${\mathrm{\σ}}_i={\left(\frac{1}{n}{\mathrm{\Σ}}_{j=1}^n{\left(h_{ij}-{\mathrm{\μ}}_i\right)}^2\right)}^{\frac{1}{2}}$

$S_i={\left(\frac{1}{n}{\mathrm{\Σ}}_{j=1}^n{\left(h_{ij}-{\mathrm{\μ}}_i\right)}^3\right)}^{\frac{1}{3}}$

This 3 low-order moments are referred to as average, variance and degree of skewness；

Space characteristics are extracted：In order to improve the description one after another to positional information, when characteristic vector positional information is calculated, using D4 Model is calculated：

D₄(P, Q)=| x_p-x_q|+|y_p-y_q|

D4 distances i.e. city block distance, it only selects both direction anyhow to calculate relative distance；

Corner Feature based on Harris operators is extracted：The detection of Harris angles is that angle is found on image by mathematical computations A kind of algorithm of feature, and it has the speciality of rotational invariance；Before the feature regular expression for setting up images match, First pass through Harris angles and detect characteristics of image " Character table ", mathematical principle is as follows：

$E(u,v)=\underset{x,y}{\Σ}w(x,y){\[I(x+u,y+v)-I(x,y)\]}^2$

Wherein W (x, y) represents moving window, and I (x, y) represents grey scale pixel value intensity, and scope is 0~255；According to Taylor series Single order is calculated to the partial derivative of N ranks, a Harris Matrix Formula is finally given：

$M=\underset{x,y}{\Σ}w(x,y)\left[\begin{array}{cc}{I}_x^2& I_x{I}_y\\ I_x{I}_y& I_y^2\end{array}\right]$

Matrix computations matrix exgenvalue λ according to Harris₁,λ₂, then calculate Harris angular response values：

R=detM-K (traceM)²

DetM=λ₁λ₂

TraceM=λ₁+λ₂

Wherein K be coefficient value, usual span be 0.04~0.06 between.

5. a kind of object module fast construction method semantic based on regular expression according to claim 1, its feature It is, in the step 3,

The lines storehouse is included by linear division：Horizontal linear, oblique line, right-angle line, camber line, S-shaped camber line, and other are self-defined It is linear；The shape library includes square, and rectangle is circular, and semicircle, rhombus is heart-shaped, and other are customized unconventional Figure；The color libraries are indicated by way of character and digit variable Wk, wherein first letter w represents colour system, behind number Word variable k represents brightness, and span is between 0-255；

The positional information storehouse describes the relative position information in image by complete customized symbol,

“|(x1)->(x2) | " level terminates to the right to pixel X2 always since pixel x1 positions for representative；

" | (x1)-^ (k) (x2) | | " represents that wherein K is variable system to the k of its vertical lower pixel x2 places since the pixel x1 Number, can take numerical value between 0-1.

6. a kind of object module fast construction method semantic based on regular expression according to claim 5, its feature It is the basis definition of the horizontal and vertical position information be given in the step 3, the definition symbol of positional information and definition are advised Then visual actual conditions self-defining and addition；It is described as element database is an open element database, for describing thing The technical indicator for managing characteristic can be as a base values in pixel element storehouse, such as angle information, temperature information, tone color letter Breath, vibration information can be extended as an element database index item of description target.

7. a kind of object module fast construction method semantic based on regular expression according to claim 1, its feature It is in the step 4, connective connection to be carried out to targeted peripheral characteristic point first, then according to most inside connection segment Big similarity principle carries out the cutting of pixel element, that is, the local feature after cutting can find most like in as element database " pixel element ".