CN102968622B - A kind of TV station symbol recognition method and TV station symbol recognition device - Google Patents

Abstract

The invention discloses a kind of TV station symbol recognition method, described method includes: extract based on 7 front 2 shape facilities that bending moment does not represents during bending moment does not represents the shape facility of described station symbol；Extract 4 the discrete shape facilities describing described station symbol based on Gaussian descriptors；Described station symbol is identified based on station symbol knowledge base according to described front 2 shape facilities that bending moment does not represents and described 4 discrete shape facilities based on Gaussian descriptors.Correspondingly, the invention also discloses a kind of TV station symbol recognition device.Use the present invention, it is possible to simply and efficiently identify station symbol.

Description

A kind of TV station symbol recognition method and TV station symbol recognition device

Technical field

The present invention relates to computer image processing technology field, more particularly, to a kind of TV station symbol recognition method.

Background technology

Station symbol, the i.e. mark of a television station, be to be different from other channels, embodies and does platform theory and the outward manifestation of image Symbol.Station symbol has identification, the basic function distinguished and beautify, and it is also to realize video analysis, the important semanteme understanding, retrieving One of source.Along with developing rapidly of TV tech, each province, city, the program set the most up to a hundred of Television Station of County Level.Generally, once have Illegal signals invades, and television station's station symbol will disappear, and invades so real-time station symbol detection and identification technology are monitoring illegal signals The key entered, has highly important practical value.

Generally after station symbol being detected, perform the feature extraction to station symbol and identification.Existing big in station symbol feature extraction Part is the most all to use to carry shape, color, textural characteristics, and the comparison of counting of the feature of extraction is many；And in extracting shape facility, greatly It is all 7 features using and extracting Hu not bending moment.The method of this station symbol Shape Feature Extraction, computationally intensive, time-consuming；And by Many in the feature comparison of counting extracted, relate to too increasing complexity to later grader.

Summary of the invention

For the problem in correlation technique, the present invention proposes a kind of TV station symbol recognition method and apparatus, and it can quickly be known Do not go out the station symbol extracted.

According to an aspect of the invention, it is provided a kind of TV station symbol recognition method, described method includes:

Extract based on 7 front 2 shape facilities that bending moment does not represents during bending moment does not represents the shape facility of described station symbol；

Extract 4 the discrete shape facilities describing described station symbol based on Gaussian descriptors；

According to described front 2 shape facilities that bending moment does not represents and described 4 discrete shapes based on Gaussian descriptors Feature identifies described station symbol based on station symbol knowledge base.

In an alternative embodiment, described method also includes the color characteristic extracting described station symbol, then according to described front 2 not Shape facility and described 4 discrete shape facilities based on Gaussian descriptors that bending moment represents identify based on station symbol knowledge base The step of described station symbol farther includes: according to described front 2 shape facilities that bending moment does not represents, institute based on Gaussian descriptors The color characteristic stating 4 discrete shape facilities and described station symbol identifies described station symbol based on station symbol knowledge base.

In an alternative embodiment, the color characteristic extracting described station symbol includes: changed by the RGB color of described station symbol Become hsv color space；Special for the color including saturation and brightness describing described station symbol in described hsv color spatial extraction Levy.

In an alternative embodiment, extracting before the step of the shape facility describing described station symbol, also including: receiving Including the signal of station symbol image information, from described signal, extract station symbol.

In an alternative embodiment, according to described front 2 shape facilities that bending moment does not represents and based on Gaussian descriptors Before described 4 discrete shape facilities identify the step of described station symbol based on station symbol knowledge base, also include: after extracting Station symbol gray processing.

According to another aspect of the present invention, additionally providing a kind of TV station symbol recognition device, described device includes:

Shape Feature Extraction module, for extracting based on 7 first 2 during bending moment does not represents the shape facility of described station symbol The shape facility that bending moment does not represents；And extract 4 the discrete shape facilities describing described station symbol based on Gaussian descriptors；

TV station symbol recognition module, according to described front 2 shape facilities that bending moment does not represents and based on Gaussian descriptors described 4 Individual discrete shape facility identifies described station symbol based on station symbol knowledge base.

In an alternative embodiment, described device also includes color feature extracted module, for extracting the color of described station symbol Feature；Described TV station symbol recognition module, is additionally operable to according to described front 2 shape facilities based on not bending moment, based on Gaussian descriptors Described 4 discrete shape facilities and the color characteristic of described station symbol identify described station symbol based on station symbol knowledge base.

In an alternative embodiment, the color characteristic of the described station symbol of described color feature extracted module extraction includes: by described The RGB color of station symbol is converted into hsv color space；And be used for describing described station symbol in described hsv color spatial extraction The color characteristic including saturation and brightness.

In an alternative embodiment, described device also includes station symbol extraction module, includes station symbol image information for receiving Signal, extracts station symbol from described signal.

In an alternative embodiment, described device includes: storage has the station symbol knowledge base of multiple station symbol SVM trained.

The present invention can be substantially reduced station symbol by extracting front 2 shape facilities in 7 shape facilities based on not bending moment The amount of calculation of feature extraction and time, can ensure that extraction at 4 shape facilities extracting station symbol based on Gaussian descriptors Front 2 moment characteristics have more robustness.Using the TV station symbol recognition method of the present invention, TV station symbol recognition rate also is able to improve further.

Accompanying drawing explanation

Fig. 1 is the flow chart of a kind of TV station symbol recognition method according to embodiments of the present invention.

Fig. 2 is the schematic diagram for explaining Gaussian descriptors principle.

Fig. 3 is the block diagram of a kind of TV station symbol recognition device according to embodiments of the present invention.

Detailed description of the invention

Below in conjunction with the accompanying drawings the embodiment of the present invention is described in detail.

Fig. 1 is the flow chart of a kind of TV station symbol recognition method according to embodiments of the present invention.As described in Figure 1, the method includes S101, extracts based on 7 front 2 shape facilities that bending moment does not represents during bending moment does not represents the shape facility of described station symbol.

Wherein, step S101 includes sub-step S101-1: the station symbol gray processing after extracting.The back of the body of the station symbol after gray processing Scape is black.Step S101 also includes sub-step S101-2: the station symbol after gray processing extracts normalized not bending moment, the most only Extract front 2 eigenvalues in 7 eigenvalues of not bending moment.

Moment characteristics mainly characterizes the geometric properties of image-region, is also called geometric moment.Due to its have rotation, translation, The invariant features of the characteristics such as yardstick, so being called again not bending moment.In image procossing, geometric invariant moment can be as a weight The feature wanted, to represent object, feature can carry out the operations such as classification accordingly to image.Geometric moment is by Hu (Visualpattern recognition by momentinvariants, IRE TRANSA CTIONS ON INFORMA TION THEORY) 1962 years propose:

Be provided with consecutive image two-dimensional function f (x, y), its (p+q) rank square is defined as follows:

$m_{\mathrm{pq}}=\underset{x}{\mathrm{\Σ}}\underset{y}{\mathrm{\Σ}}{x}^p{y}^{q}f(x,y)$ P, q=0,1,2... (1)

Above-mentioned moment characteristics amount carries out place normalization obtain image f (x, central moment y) is

${\mathrm{\μ}}_{\mathrm{pq}}=\underset{x}{\mathrm{\Σ}}\underset{y}{\mathrm{\Σ}}{(x-\stackrel{\‾}{x})}^p{(y-\stackrel{\‾}{y})}^{q}f(x,y)$ P, q=0,1,2... (2)

WhereinFor the barycenter of image,

Central moment carries out size normalization again, and definition normalization central moment is:

R=(p+q)/2+1p+q=2,3,4... (3)

Utilize second order and third central moment to construct 7 and there is translation, scaling and the moment invariants of invariable rotary shape, specifically Expression formula is as follows:

$\left\{\begin{array}{c}\\ {\mathrm{\φ}}_1={\mathrm{\η}}_{20}+{\mathrm{\η}}_{02}\\ {\mathrm{\φ}}_2={({\mathrm{\η}}_{20}-{\mathrm{\η}}_{02})}^2+4{\mathrm{\η}}_{11}^2\\ {\mathrm{\φ}}_3={({\mathrm{\η}}_{30}-3{\mathrm{\η}}_{12})}^2+{(3{\mathrm{\η}}_{21}-{\mathrm{\η}}_{03})}^2\\ {\mathrm{\φ}}_4={({\mathrm{\η}}_{30}+{\mathrm{\η}}_{12})}^2+{({\mathrm{\η}}_{21}+{\mathrm{\η}}_{03})}^2\\ {\mathrm{\φ}}_5=({\mathrm{\η}}_{30}-3{\mathrm{\η}}_{12})({\mathrm{\η}}_{30}+{\mathrm{\η}}_{12})[{({\mathrm{\η}}_{30}+{\mathrm{\η}}_{12})}^2-3{({\mathrm{\η}}_{21}+{\mathrm{\η}}_{03})}^2]+\\ (3{\mathrm{\η}}_{21}-{\mathrm{\η}}_{03})({\mathrm{\η}}_{21}+{\mathrm{\η}}_{03})[3{({\mathrm{\η}}_{30}+{\mathrm{\η}}_{12})}^2-{({\mathrm{\η}}_{21}+{\mathrm{\η}}_{03})}^2]\\ {\mathrm{\φ}}_6=({\mathrm{\η}}_{20}-{\mathrm{\η}}_{02})[{({\mathrm{\η}}_{30}+{\mathrm{\η}}_{12})}^2-{({\mathrm{\η}}_{12}-{\mathrm{\η}}_{03})}^2]+4{\mathrm{\η}}_{11}({\mathrm{\η}}_{30}+{\mathrm{\η}}_{12})({\mathrm{\η}}_{21}+{\mathrm{\η}}_{03})\\ {\mathrm{\φ}}_7=(3{\mathrm{\η}}_{21}-{\mathrm{\η}}_{03})({\mathrm{\η}}_{30}+{\mathrm{\η}}_{21})[{({\mathrm{\η}}_{30}+{\mathrm{\η}}_{12})}^2-3{({\mathrm{\η}}_{21}+{\mathrm{\η}}_{03})}^2]+\\ (3{\mathrm{\η}}_{21}-{\mathrm{\η}}_{30})({\mathrm{\η}}_{21}+{\mathrm{\η}}_{03})[3{({\mathrm{\η}}_{30}+{\mathrm{\η}}_{12})}^2-{({\mathrm{\η}}_{21}+{\mathrm{\η}}_{03})}^2]\end{array}\right.---\left(4\right)$

Moment characteristics has obvious physical significance, null matrix m of target₀₀Reflecting the area of target, first moment reflects The centroid position of target, just can avoid the shadow to characteristics of image because of article size and change in displacement hence with the two square amount Ring.

Due to 7 of same image not bending moment φ_k(k=1,2 ..., 7) the order of magnitude differ greatly, for the ease of follow-up The identification of image, the available method taking natural logrithm carries out scope of data compression, simultaneously takes account of not bending moment and likely occur The situation of negative value, therefore the not bending moment of actual employing is:

Φ_k=| ln | φ_k| | k=1,2 ..., 7 (5)

In the embodiment of the present invention, extract φ based on formula (5)₁And φ₂For representing the shape facility of station symbol.

S102, extracts 4 the discrete shape facilities describing described station symbol based on Gaussian descriptors.

Wherein, step S102 includes sub-step S102-1: extract the edge of station symbol.Edge extracting is image processing field Common technology.Such as carry out edge extracting with wavelet transformation.Step S102 also includes sub-step S102-2: calculate total arc of station symbol Long l_total；And sub-step S102-3 calculates 4 centrifugal pumps based on the Gaussian descriptors characteristic vector with pie graph picture.

The principle of Gaussian descriptors is described below and describes the enforcement of step S102 in detail.

If curve C is by curve C₁, C₂..., C_n(n≤1) forms.From the relevant knowledge of calculus, the barycenter of curve C Coordinate can be expressed as with the 1st type curvilinear integral

$\stackrel{\‾}{x}=\frac{{\∫}_c\mathrm{xds}}{{\∫}_c\mathrm{ds}},$ $\stackrel{\‾}{y}=\frac{{\∫}_c\mathrm{yds}}{{\∫}_c\mathrm{ds}}---\left(6\right)$

On curve C degree of scatter a little can use variances sigma²Describe:

${\mathrm{\σ}}^2=\frac{{\∫}_c[{(x-\stackrel{\‾}{x})}^2+{(y-\stackrel{\‾}{y})}^2]\mathrm{ds}}{{\∫}_c\mathrm{ds}}---\left(7\right)$

With r represent on curve C the meansigma methods of distance a little and between barycenter:

$r=\frac{{\∫}_c\sqrt{{(x-\stackrel{\‾}{x})}^2+{(y-\stackrel{\‾}{y})}^2}\mathrm{ds}}{{\∫}_c\mathrm{ds}}---\left(8\right)$

Claim withFor the center of circle, the circle with r as radius is the reference circle of curve C.

To arbitrary λ ∈ (0 ,+∞), remember curve C^(λ)=(x, y) | (x, y) ∈ C, and ${(x-\stackrel{\‾}{x})}^2+{(y-\stackrel{\‾}{y})}^2\≤{\left(\mathrm{\λr}\right)}^2\}.$

To arbitrary θ ∈ [0,2 π] and λ ∈ (0 ,+∞), defined function

$f(\mathrm{\θ},\mathrm{\λ})=\frac{{\∫}_{c^{\left(\mathrm{\λ}\right)}}{e}^{-\frac{{(x-\stackrel{\‾}{x}-\mathrm{\λ}r\cos \mathrm{\θ})}^2+{(y-\stackrel{\‾}{y}-\mathrm{\λ}r\sin \mathrm{\θ})}^2}{2{\mathrm{\σ}}^2}\mathrm{ds}}}{{\∫}_c\mathrm{ds}}---\left(9\right)$

The meaning of function f (θ, λ) is as shown in Figure 2.

In Fig. 1, part edge curve is positioned at round inner arc AmB and arc CnD, i.e. curve C^(λ)=AmB+CnD.To C^(λ)On appoint Anticipate 1 Q, if its coordinate be (x, y), then the spacing of point P and Q square is:

${\left|\mathrm{PQ}\right|}^2={(x-\stackrel{\‾}{x}-\mathrm{\λ}r\cos \mathrm{\θ})}^2+{(y-\stackrel{\‾}{y}-\mathrm{\λ}r\sin \mathrm{\θ})}^2$

" potential function " in f (θ, λ) analog Neo-Confucianism, replaces simple inverse proportion letter with 2 dimension Gaussian functions the most here Number, and the point (particle in analog Neo-Confucianism, point charge etc.) on circle outward flange curve is masked.F (θ, λ) is C^(λ)Upper institute There is the summation of " the acting on " of point-to-point P.∫ in denominator_cThe effect of ds is by yardstick standardization.Generally function f (θ, λ) is claimed For Gauss potential function.For given θ and λ, f (θ, λ) about translation and Scale invariant.

Defined function

$l\left(\mathrm{\λ}\right)=\frac{{\∫}_0^{2\mathrm{\π}}f(\mathrm{\θ},\mathrm{\λ})\mathrm{d\θ}}{2\mathrm{\π}}---\left(10\right)$

When λ fixes, Gauss potential function f (θ, λ) is the continuous function of θ.I (λ) can be interpreted as f (θ, λ) intuitively Meansigma methods on interval [0,2 π].It may also be said that I (λ) is the meansigma methods of each point on circumference " gesture ".So, I (λ) portrays Edge image is positioned atFor the center of circle, the shape facility of part in the circle with λ r as radius.I (λ) is referred to as the height of curve C This describes son.

For the edge of station symbol image, by formula (6) to formula (10) discretization.To this end, first provide total arc length of calculated curve C ∫_cDs (is designated as l_total) discretization formula.Utilize the definition of the 1st type curvilinear integral, available:

$l_{\mathrm{total}}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\mathrm{\Δ}{s}_i---\left(11\right)$

Formula (6) is the most corresponding following various to the discretization result of formula (9): $\stackrel{\‾}{x}=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_i\mathrm{\Δ}{s}_i}{l_{\mathrm{total}}},$ $\stackrel{\‾}{y}=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_i\mathrm{\Δ}{s}_i}{l_{\mathrm{total}}},$ (x_i, y_i)∈c (12) ${\mathrm{\σ}}^2=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}[{(x_i-\stackrel{\‾}{x})}^2+{(y_i-\stackrel{\‾}{y})}^2]\mathrm{\Δ}{s}_i}{l_{\mathrm{total}}},$ (x_i, y_i)∈c (13) $r=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\sqrt{{(x_i-\stackrel{\‾}{x})}^2+{(y_i-\stackrel{\‾}{y})}^2}\mathrm{\Δ}{s}_i}{l_{\mathrm{total}}},$ (x_i, y_i)∈c (14) $f(\mathrm{\θ},\mathrm{\λ})=\frac{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{e}^{-\frac{{(x_i-\stackrel{\‾}{x}-\mathrm{\λ}r\cos \mathrm{\θ})}^2+{(y_i-\stackrel{\‾}{y}-\mathrm{\λ}r\sin \mathrm{\θ})}^2}{2{\mathrm{\σ}}^2}\mathrm{\Δ}{s}_i}}{l_{\mathrm{total}}},$ (x_i, y_i)∈C^(λ) (15)

In above formula, Δ s_iIt is " distance " on contour curve between two neighbor pixels: if two adjacent Pixel is horizontally oriented or vertical direction arrangement, then Δ s_i=1；If two neighbor pixels are 45 ° or 135 ° of directions Arrangement, then

Interval [0,2 π] is bisected into K part, then formula (5) can get according to the correlation theory of definite integral:

$I\left(\mathrm{\λ}\right)\≈\frac{\frac{2\mathrm{\π}}{K}\underset{i=1}{\overset{K}{\mathrm{\Σ}}}f(\frac{2\mathrm{\π}}{K}i,\mathrm{\λ})}{2\mathrm{\π}}=\frac{\underset{i=1}{\overset{K}{\mathrm{\Σ}}}(\frac{2\mathrm{\π}}{K}i,\mathrm{\λ})}{K}---\left(16\right)$

Result shows, can select K value according to practical situation between 3≤K≤8, and K is integer.In the embodiment of the present invention In, K value takes 8.

In order to obtain enough information to portray the shape facility of object, thus improve identification/recall ratio, take 4 differences λ value, such as: λ₁=0.25r, λ₃=0.75r, λ₅=1.25r, λ₇=1.75r.So, correspondingly can get Gaussian descriptors 4 centrifugal pumps.Then 4 shape eigenvectors based on Gaussian descriptors are constituted for station symbol by these 4 centrifugal pumps Identify/retrieval.

Step 103, according to according to described front 2 shape facilities that bending moment does not represents and based on Gaussian descriptors described 4 Individual discrete shape facility identifies station symbol based on station symbol knowledge base.

In one embodiment, SVM technology (support vector machine) is used to identify station symbol.SVM is to manage at statistical learning On the basis of Lun, a kind of new learning machine structural theory that developed.In order to solve this multi-class problem of TV station symbol recognition, adopt equally Use one-to-many strategy, i.e. train a SVM for each television channel, enable each SVM to distinguish each channel station symbol.

From station caption sample library, take out all samples of the first channel, be designated as class I, owning other all channels Sample is designated as class II, using all samples of having marked classification as input sample, trains a SVM, supported accordingly to Amount and corresponding (broad sense) optimal classification surface, be designated as A by this SVM sequence number, show A SVM be used to distinguish first channel and Remaining channel.

From station caption sample library, take out all samples of second channel, be designated as class I, by the institute of other all channels Having sample point to be designated as class II, the SVM sequence number trained, as the operation of previous step, is finally designated as B by following step, should SVM is used to distinguish second channel and other all channels.

Repeat above step, until all channels in traversal Sample Storehouse.Finally obtain m SVM.

Setting up station symbol knowledge base, it can include the station symbol SVM trained, and television station's title, background information and program take To etc. knowledge.

If only having m channel in Sample Storehouse to be known, and each channel has a corresponding SVM trained.When defeated Enter a new sample to be known, simultaneously by this in time knowing sample and classified by m SVM: if this sample is assigned to by i-th SVM In class I, remaining SVM is classified in class II, then judge that this sample belongs to the i-th channel in Sample Storehouse；If have simultaneously This sample is assigned in class I by multiple SVM, and remaining SVM is assigned in class II, then be judged as wrong point；If all SVM should Sample is assigned in class II, is the most also judged as wrong point.So the most a certain sample is identified, just can be by correspondence in station symbol knowledge base Related content carries out semantic tagger to it.

In the TV station symbol recognition method of alternative embodiment of the present invention, also include: extract the color characteristic of station symbol.Then, root According to described front 2 shape facilities based on not bending moment, described 4 discrete shape facilities based on Gaussian descriptors and described Target color characteristic identifies described station symbol based on station symbol knowledge base.

Wherein, the color characteristic extracting described station symbol includes: the RGB color of described station symbol is converted into hsv color Space.Hsv color space is the mode of a kind of color characteristic describing station symbol.HSV is expressed as three attribute colour signal: color Adjust H, saturation S and brightness V.

Then, special for the color including saturation and brightness describing described station symbol in described hsv color spatial extraction Levy.

Fig. 3 is the block diagram of a kind of TV station symbol recognition device that the embodiment of the present invention provides.Device 10 includes: Shape Feature Extraction Module 110, for extracting front 2 shape facilities in 7 shape facilities for describing described station symbol based on not bending moment；With And extract 4 discrete shape facilities for describing described station symbol based on Gaussian descriptors；And TV station symbol recognition module 130, according to described front 2 shape facilities based on not bending moment and described 4 discrete shape facilities based on Gaussian descriptors Described station symbol is identified based on station symbol knowledge base 140.

In an alternative embodiment, described device 10 also includes color feature extracted module 120, for extracting described station symbol Color characteristic.TV station symbol recognition module 130, is additionally operable to according to described front 2 shape facilities based on not bending moment, describes based on Gauss Described 4 discrete shape facilities of son and the color characteristic of described station symbol identify described station symbol based on station symbol knowledge base 140.

The color characteristic that described color feature extracted module 120 extracts described station symbol includes: by the RGB color of described station symbol Space is converted into hsv color space；And described hsv color spatial extraction for describe described station symbol include saturation and The color characteristic of brightness.

In an alternative embodiment, described device also includes station symbol extraction module, includes station symbol image information for receiving Signal, extracts station symbol from described signal.

In an alternative embodiment, station symbol knowledge base 140 storage has multiple station symbol SVM trained.

Traditional Shape Feature Extraction based on Hu not bending moment is all based on 7 not bending moments, tests through substantial amounts of experiment Card, 7 constant Hu squares all participate in the extraction of feature, and the error sometimes produced is relatively big, and time-consuming, is carrying out identification below Time the most also can produce misclassification.In order to ensure that front 2 the Hu moment characteristics extracted have more robustness, the present invention introduces again height This describes son, utilizes Gaussian descriptors to extract 4 shape facilities of station symbol, and it is little that Gaussian descriptors has amount of calculation, to appropriateness Edge variation and insensitive for noise feature.In order to embody the strikingly color feature of station symbol, it is special that the present invention additionally extracts 2 colors Levy.

By means of the invention it is possible to greatly reduce amount of calculation and the time of station symbol feature extraction, and discrimination also have into One step improves.

The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all essences in the present invention Within god and principle, any modification, equivalent substitution and improvement etc. made, should be included within the scope of the present invention.

Claims (8)

1. a TV station symbol recognition method, described method includes:

Extract based on 7 front 2 shape facilities that bending moment does not represents during bending moment does not represents the shape facility of described station symbol；

Extract 4 the discrete shape facilities describing described station symbol based on Gaussian descriptors；

According to described front 2 shape facilities that bending moment does not represents and described 4 discrete shape facilities based on Gaussian descriptors Described station symbol is identified based on station symbol knowledge base；

Wherein, described 4 discrete shape facilities be 4 centrifugal pumps based on Gaussian descriptors with the feature of pie graph picture to Amount；

Described method also includes:

Extract the color characteristic of described station symbol；

According to described front 2 shape facilities that bending moment does not represents and described 4 discrete shape facilities based on Gaussian descriptors The step identifying described station symbol based on station symbol knowledge base farther includes:

According to described front 2 shape facilities that bending moment does not represents, described 4 discrete shape facilities based on Gaussian descriptors and The color characteristic of described station symbol identifies described station symbol based on station symbol knowledge base；

Wherein, extract 4 the discrete shape facilities describing described station symbol based on Gaussian descriptors to include:

Step S101, extracts the edge of described station symbol；

Step S102, calculates total arc length l of described station symbol_total；And

Step S103, calculates 4 centrifugal pumps based on the Gaussian descriptors characteristic vector with pie graph picture；

Wherein, according to described front 2 shape facilities that bending moment does not represents and described 4 discrete shapes based on Gaussian descriptors Feature identifies described station symbol based on station symbol knowledge base and includes:

From station caption sample library, take out all samples of the first channel, be designated as class I, by all samples of other all channels It is designated as class II, using all samples of having marked classification as input sample, trains a support vector machine, supported accordingly The corresponding optimal classification surface of vector sum, is designated as A by described support vector machine sequence number, shows that A support vector machine is used to distinguish First channel and remaining channel；

From described station caption sample library, take out all samples of second channel, be designated as class I, by the institute of other all channels Have sample point to be designated as class II, following step as the operation of previous step, the described support vector machine that finally will train Being designated as B, described support vector machine is used to distinguish second channel and other all channels；

Repeat above step, until all channels traveled through in described Sample Storehouse, finally obtain m support vector machine；

Setting up station symbol knowledge base, it can include the described station symbol support vector machine trained, television station's title, background information and Program is orientated；

If only having m channel in Sample Storehouse to be known, and each channel has a corresponding described support vector trained Machine, when one new sample to be known of input, simultaneously by described in time knowing sample and classified by m support vector machine: if i-th This sample is assigned in class I by support vector machine, and remaining support vector machine is classified in class II, then judge that described sample belongs to I-th channel in Sample Storehouse；If there is multiple support vector machine to be assigned in class I by this sample simultaneously, and remaining support to Amount machine is assigned in class II, then be judged as wrong point；If described sample is assigned in class II by all support vector machine, the most also judge Dividing for mistake, the most a certain sample is identified, and just by related content corresponding in station symbol knowledge base, it can be carried out semantic mark Note.

TV station symbol recognition method the most according to claim 1, it is characterised in that the color characteristic extracting described station symbol includes:

The RGB color of described station symbol is converted into hsv color space；

In described hsv color spatial extraction for describing the color characteristic including saturation and brightness of described station symbol.

TV station symbol recognition method the most according to claim 1, it is characterised in that extracting the shape for describing described station symbol Before the step of feature, also include: receive the signal including station symbol image information, from described signal, extract station symbol.

TV station symbol recognition method the most according to claim 3, it is characterised in that extract based on 7 bending moment do not represent described Before the step of front 2 the not shape facilities that bending moment represents in the shape facility of station symbol, also include:

Station symbol gray processing after extracting.

5. a TV station symbol recognition device, described device includes:

Shape Feature Extraction module, for extract based on 7 during bending moment does not represents the shape facility of described station symbol front 2 constant The shape facility that square represents；And extract 4 the discrete shape facilities describing described station symbol based on Gaussian descriptors, its In, described 4 discrete shape facilities are 4 centrifugal pumps based on the Gaussian descriptors characteristic vectors with pie graph picture；

TV station symbol recognition module, according to described front 2 shape facilities that bending moment does not represents and based on Gaussian descriptors described 4 from The shape facility dissipated identifies described station symbol based on station symbol knowledge base；

Wherein, described device also includes color feature extracted module, for extracting the color characteristic of described station symbol；

Described TV station symbol recognition module, is additionally operable to according to described front 2 shape facilities that bending moment does not represents, based on Gaussian descriptors The color characteristic of described 4 discrete shape facilities and described station symbol identifies described station symbol based on station symbol knowledge base；

Wherein, described Shape Feature Extraction module includes:

First extracts submodule, for extracting the edge of described station symbol；

First calculating sub module, for calculating total arc length l of described station symbol_total；And

Second calculating sub module, for calculating 4 centrifugal pumps based on the Gaussian descriptors characteristic vector with pie graph picture；

Wherein, described TV station symbol recognition module includes:

First takes out discriminating module, for taking out all samples of the first channel from station caption sample library, is designated as class I, will All samples of other all channels are designated as class II, using all samples having marked classification as inputting sample, train one Hold vector machine, supported the corresponding optimal classification surface of vector sum accordingly, described support vector machine sequence number is designated as A, shows A Number support vector machine is used to distinguish first channel and remaining channel；

Second takes out discriminating module, for taking out all samples of second channel from described station caption sample library, is designated as All sample points of other all channels are designated as class II by class I, and following step, as the operation of previous step, finally will The described support vector machine trained is designated as B, and described support vector machine is used to distinguish second channel and other all channels 's；

Replicated blocks, are used for repeating described first and take out discriminating module and described second taking-up discriminating module, until traversal is described All channels in Sample Storehouse, have finally obtained m support vector machine；

Setting up module, be used for setting up station symbol knowledge base, it can include the described station symbol support vector machine trained, television station's name Title, background information and program orientation；

TV station symbol recognition submodule, is used for setting in Sample Storehouse to be known only m channel, and each channel has a corresponding instruction The described support vector machine perfected, when one new sample to be known of input, simultaneously by described sample to be known by m support vector machine When classifying: if this sample is assigned in class I by i-th support vector machine, remaining support vector machine is classified into class II In, then judge that described sample belongs to the i-th channel in Sample Storehouse；If there is multiple support vector machine that this sample is assigned to class simultaneously In I, and remaining support vector machine is assigned in class II, then be judged as wrong point；If all support vector machine are by described sample Assigning in class II, be the most also judged as mistake point, the most a certain sample is identified, just can be with corresponding being correlated with in station symbol knowledge base Content carries out semantic tagger to it.

TV station symbol recognition device the most according to claim 5, it is characterised in that described color feature extracted module is extracted described The color characteristic of station symbol includes:

The RGB color of described station symbol is converted into hsv color space；

In described hsv color spatial extraction for describing the color characteristic including saturation and brightness of described station symbol.

TV station symbol recognition device the most according to claim 5, it is characterised in that described device also includes station symbol extraction module, For receiving the signal including station symbol image information, from described signal, extract station symbol.

TV station symbol recognition device the most according to claim 5, it is characterised in that described device also includes that storage has multiple training The station symbol knowledge base of good station symbol SVM.