CN105844251A - Cartoon video identification method and device - Google Patents

Abstract

The invention discloses a cartoon video identification method and device, belonging to the mode identification technical field. The method comprises: extracting key frames from a video to be identified, obtaining image characteristics from the key frames, calculating the cartoon image membership of the key frames according to the image characteristics of each key frame, and determining whether the video is a carton video according to the cartoon image membership of all the key frames of the video to be identified. The method and device possess the characteristics of simple algorithm and reasonable image characteristic selection, reach an appropriate balance between identification speed and identification accuracy, and are very suitable for identification scenes of a lot of videos.

Description

A kind of cartoon video recognition methods and device

Technical field

The present invention relates to mode identification technology, particularly relate to a kind of cartoon video recognition methods and dress Put.

Background technology

At present, along with network technology and the high speed development of multimedia technology, various multimedia messages are the hugest Rich, greatly enrich daily life.Meanwhile, video as a kind of common multimedia form, Daily life with the mankind is closely bound up, is also to access more a kind of resources mode on network.

According to the 34th time, China Internet center China Internet network state of development statistical report, cut-off In by the end of June, 2014 China held up to 4.39 hundred million internet video user, the user of Internet video Quantity has accounted for the 1/3 of Chinese population sum.Along with increasing of number of users, user is for Online Video Demand is the most increasing.In order to sufficiently meet the demand of user, Ge great portal website constantly expands online Video library, causes sharply increasing of internet video quantity, the number of videos of single portal website to reach To tens billion of more than.It is reported an investigational data of survey institute ComScore, in October, 2011 is excellent extremely Net video playback amount reach 4,600,000,000 times, day the amount of uploading be 70,000.

But, owing to kind and the quantity of video grow with each passing day, so how the video of these magnanimity is entered Row taxonomic revision, enables one to be quickly found out oneself content interested according to classification and becomes for one Individual important problem.To this end, the automatic analysis system of video just arises at the historic moment, the inspection of automatically classifying of video Method of determining and calculating also becomes a study hotspot of area of pattern recognition.

It will be seen that common video type mainly has film, electricity from the visual classification label of happy view Depending on play, physical culture, animation etc. classification, wherein animation i.e. cartoon video.At these video genre In, cartoon video is a kind of special classification, and it is different from " authenticity " of other videos, but passes through The video that craft or computer aided painting are made.At present, animation industry has become a weight the most The cultural industry wanted, therefore the proportion shared by animation video also will be the hugest.Therefore, cartoon video is known The most just become an important research direction in visual classification field.

One important feature of cartoon video is, cartoon video has more significantly edge feature, with Time, the color of cartoon video is the abundantest.Based on these features, more existing cartoons in prior art Video frequency identifying method, it is right such as to be come by the statistics color of video, texture, shape, motion etc. feature Video genre is identified, and wherein " identifies " and uses grader that precondition is good to a certain the most exactly Organize concrete characteristics of image to classify.But, owing to the extraction of characteristics of image can not be divided comprehensively Class device there is also inevitable deviation, and therefore recognition result exists inaccuracy definitely.

Summary of the invention

In view of this, it is an object of the invention to propose a kind of cartoon video recognition methods and device, it is possible to Improve the accuracy rate of cartoon video identification further.

Based on above-mentioned purpose, the technical scheme that the present invention provides is:

A kind of cartoon video recognition methods, the method comprises:

Key frame is extracted from video to be identified；

Characteristics of image is obtained from key frame；

Use the first sorting algorithm according to the cartoon image of this key frame of box counting algorithm of each key frame Degree of membership；

The span of cartoon image degree of membership is divided at least three interval, and adds up all key frames The interval distribution situation of cartoon image degree of membership；

The second sorting algorithm is used to judge whether video to be identified is card intervisibility according to described interval distribution situation Frequently.

Specifically, characteristics of image can comprise color histogram, edge histogram, high luminance pixels ratio, Edge pixel ratio and color moment information, wherein color moment information is calculated by color histogram；Edge Histogram is drawn by gradient direction and the gradient magnitude of statistical pixel point；High luminance pixels ratio refers to HSV V (Value, lightness) parameter in (Hue-Saturation-Value, tone-saturation degree-lightness) space Ratio more than the pixel of threshold X；Edge pixel ratio refers to that gradient magnitude is more than the pixel of threshold value Y The ratio of point.

Specifically, the calculation that edge histogram represents in the matrix form is:

B_mn=L_m×G_n,

In formula, B_mnRepresent the edge histogram matrix of m row n row, L_mRepresent the gradient magnitude Nogata of m row 1 row Figure matrix, G_nRepresent the gradient orientation histogram matrix of 1 row n row；

The calculation of above-mentioned gradient orientation histogram matrix is:

$G_n=\frac{1}{N}\underset{j=1}{\overset{N}{\Σ}}{Q}_{nj},$

In formula: N represents by the sum of statistical pixel point；Q_njIt is the matrix of 1 row n row, represents jth The pixel contribution to each gradient direction interval；

Above-mentioned Q_njCalculation be: the gradient direction setting jth pixel falls into quantized interval S, then it is right Quantized interval S and quantized interval T all has contribution, and to other quantized intervals all without contribution, described amount Changing interval T is quantization district minimum with the gradient direction angle of jth pixel in addition to quantized interval S Between；The jth pixel contribution v to quantized interval S_SWith the contribution v to quantized interval T_TCount as the following formula Calculate:

$v_S=\frac{{\mathrm{\θ}}_T}{{\mathrm{\γ}}_{ST}},$

$v_T=\frac{{\mathrm{\θ}}_S}{{\mathrm{\γ}}_{ST}},$

In formula, γ_STRepresent what the direction that characterized, midpoint of quantized interval S and the midpoint of quantized interval T were characterized The angle in direction, θ_SRepresent what the gradient direction of jth pixel and the midpoint of quantized interval S were characterized The angle in direction, θ_TRepresent what the gradient direction of jth pixel and the midpoint of quantized interval T were characterized The angle in direction.

Specifically, color histogram carries out equal interval quantizing to color in HSV space and draws, color Square information is made up of the front third moment of color histogram, i.e. first moment (mean value Mean), second moment (variance Variance) and third moment (degree of bias Skewness).

Specifically, gradient direction interval division in the range of 0～180 degree with it at 180～360 degree of models Enclose interior interval division specular.

Specifically, key frame is both from effective section of video to be identified, and effective section is to be identified regarding Frequency removes the remainder behind the beginning part and ending, and the duration of effective section is at least whole waiting to be known The 50% of other video duration, and the duration of the beginning part and ending all at least whole video to be identified The 8% of duration.Advantage of this is that: no matter general cartoon video is also that non-cartoon video all can have head And run-out, and head and run-out be probably captions, this can affect the algorithm recognition result to video, therefore Preferably the beginning and end of video is removed.

Specifically, characteristics of image is all from the effective coverage of key frame, and the area of effective coverage is at least For the 25% of whole key frame area, and effective coverage is positioned at the desirable region of key frame；Desirable region For having the similar figures of the key frame of common geometric center with key frame, and the area in desirable region is crucial The 64% of frame area.Advantage of this is that: may regard no matter cartoon video is also non-cartoon video The marginal portion of frequency has black surround and/or captions, in order to avoid black surround and the interference of captions, best during identification The geometry mid portion of selecting video key frame.

A kind of cartoon video identification device, comprises:

Key-frame extraction module: for extracting key frame from video to be identified；

Image characteristics extraction module: for obtaining characteristics of image from key frame；

First sort module: for using the first sorting algorithm according to the box counting algorithm of each key frame The cartoon image degree of membership of this key frame；

Degree of membership distribution statistics module: for the span of cartoon image degree of membership is divided at least three Interval, and add up the interval distribution situation of the cartoon image degree of membership of all key frames；

Second sort module: for using the second sorting algorithm to judge to wait to know according to described interval distribution situation Whether other video is cartoon video.

Specifically, characteristics of image can comprise color histogram, edge histogram, high luminance pixels ratio, Edge pixel ratio, and the color moment information obtained by color histogram；Edge histogram is by statistics Gradient direction and the gradient magnitude of pixel draw；High luminance pixels ratio refers to V parameter in HSV space Ratio more than the pixel of threshold X；Edge pixel ratio refers to that gradient magnitude is more than the pixel of threshold value Y The ratio of point.

Specifically, the calculation that edge histogram represents in the matrix form is:

B_mn=L_m×G_n,

In formula, B_mnRepresent the edge histogram matrix of m row n row, L_mRepresent the gradient magnitude Nogata of m row 1 row Figure matrix, G_nRepresent the gradient orientation histogram matrix of 1 row n row；

The calculation of above-mentioned gradient orientation histogram matrix is:

$G_n=\frac{1}{N}\underset{j=1}{\overset{N}{\Σ}}{Q}_{nj},$

In formula: N represents by the sum of statistical pixel point；Q_njIt is the matrix of 1 row n row, represents jth The pixel contribution to each gradient direction interval；

Above-mentioned Q_njCalculation be: the gradient direction setting jth pixel falls into quantized interval S, then it is right Quantized interval S and quantized interval T all has contribution, and to other quantized intervals all without contribution, described amount Changing interval T is quantization district minimum with the gradient direction angle of jth pixel in addition to quantized interval S Between；The jth pixel contribution v to quantized interval S_SWith the contribution v to quantized interval T_TCount as the following formula Calculate:

$v_S=\frac{{\mathrm{\θ}}_T}{{\mathrm{\γ}}_{ST}},$

$v_T=\frac{{\mathrm{\θ}}_S}{{\mathrm{\γ}}_{ST}},$

In formula, γ_STRepresent the direction that characterized, midpoint of quantized interval S and the midpoint institute table of quantized interval T The angle in the direction levied, θ_SRepresent the gradient direction of jth pixel and the midpoint institute table of quantized interval S The angle in the direction levied, θ_TRepresent the gradient direction of jth pixel and the midpoint institute table of quantized interval T The angle in the direction levied.

Specifically, color histogram carries out equal interval quantizing to color in HSV space and draws, color Square information is made up of the front third moment of color histogram, i.e. first moment (mean value Mean), second moment (variance Variance) and third moment (degree of bias Skewness).

Specifically, gradient direction interval division in the range of 0～180 degree with it at 180～360 degree of models Enclose interior interval division specular.

Specifically, key frame is both from effective section of video to be identified, and effective section is to be identified regarding Frequency removes the remainder behind the beginning part and ending, and the duration of effective section is at least whole waiting to be known The 50% of other video duration, and the duration of the beginning part and ending all at least whole video to be identified The 8% of duration.Advantage of this is that: no matter general cartoon video is also that non-cartoon video all can have head And run-out, and head and run-out be probably captions, this can affect the algorithm recognition result to video, therefore Preferably the beginning and end of video is removed.

Specifically, characteristics of image is all from the effective coverage of key frame, and the area of effective coverage is at least For the 25% of whole key frame area, and effective coverage is positioned at the desirable region of key frame；Desirable region For having the similar figures of the key frame of common geometric center with key frame, and the area in desirable region is crucial The 64% of frame area.Advantage of this is that: may regard no matter cartoon video is also non-cartoon video The marginal portion of frequency has black surround and/or captions, in order to avoid black surround and the interference of captions, best during identification The geometry mid portion of selecting video key frame.

From the above it can be seen that the beneficial effects of the present invention is:

Cartoon image degree of membership has been divided into multiple interval, and the cartoon figure to all key frames by the present invention As the distribution situation of degree of membership is added up, then statistics is inputted the second grader and has carried out again Subseries, thus improve the judgment accuracy of cartoon video, still maintain relatively low algorithm multiple simultaneously Miscellaneous degree, is a kind of important improvement to prior art.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to reality Execute the required accompanying drawing used in example or description of the prior art to be briefly described, it should be apparent that below, Accompanying drawing in description is only some embodiments of the present invention, for those of ordinary skill in the art, On the premise of not paying creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings.

Fig. 1 is a kind of flow chart of the inventive method embodiment；

Fig. 2 is a kind of structural representation of apparatus of the present invention embodiment；

Fig. 3 is a kind of angular interval dividing mode schematic diagram in the embodiment of the present invention.

Detailed description of the invention

For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with being embodied as Example, and referring to the drawings, the present invention is described in more detail.

It should be noted that the statement of all uses " first " and " second " is equal in the embodiment of the present invention It is for the parameter of the entity or non-equal distinguishing two same names non-equal, it is seen that " first " " the Two " only for the convenience of statement, should not be construed as the restriction to the embodiment of the present invention, subsequent embodiment pair This illustrates the most one by one.

A kind of cartoon video recognition methods, the method comprises:

Extracting key frame from video to be identified, key frame can use ffmpeg etc. software to extract；

Characteristics of image is obtained from key frame；

Use the first sorting algorithm according to the cartoon image of this key frame of box counting algorithm of each key frame Degree of membership；

The span of cartoon image degree of membership is divided at least three interval, and adds up all key frames The interval distribution situation of cartoon image degree of membership；

The second sorting algorithm is used to judge whether video to be identified is card intervisibility according to described interval distribution situation Frequently.

Specifically, characteristics of image comprises color histogram, edge histogram, high luminance pixels ratio, edge Pixel ratio, and the color moment information obtained by color histogram；Edge histogram passes through statistical pixel Gradient direction and the gradient magnitude of point draw, specifically can use Sobel operator；High luminance pixels ratio Refer to V in HSV (Hue-Saturation-Value, tone-saturation degree-lightness) space (Value, bright Degree) parameter is more than the ratio of pixel of threshold X；Edge pixel ratio refers to that gradient magnitude is more than threshold value The ratio of the pixel of Y.

Specifically, the calculation that edge histogram represents in the matrix form is:

B_mn=L_m×G_n,

In formula, B_mnRepresent the edge histogram matrix of m row n row, L_mRepresent the gradient magnitude Nogata of m row 1 row Figure matrix, G_nRepresent the gradient orientation histogram matrix of 1 row n row；

The calculation of above-mentioned gradient orientation histogram matrix is:

$G_n=\frac{1}{N}\underset{j=1}{\overset{N}{\Σ}}{Q}_{nj},$

In formula: N represents by the sum of statistical pixel point；Q_njIt is the matrix of 1 row n row, represents jth The pixel contribution to each gradient direction interval；

Above-mentioned Q_njCalculation be: the gradient direction setting jth pixel falls into quantized interval S, then it is right Quantized interval S and quantized interval T all has contribution, and to other quantized intervals all without contribution, described amount Changing interval T is quantization district minimum with the gradient direction angle of jth pixel in addition to quantized interval S Between；The jth pixel contribution v to quantized interval S_SWith the contribution v to quantized interval T_TCount as the following formula Calculate:

$v_S=\frac{{\mathrm{\θ}}_T}{{\mathrm{\γ}}_{ST}},$

$v_T=\frac{{\mathrm{\θ}}_S}{{\mathrm{\γ}}_{ST}},$

In formula, γ_STRepresent what the direction that characterized, midpoint of quantized interval S and the midpoint of quantized interval T were characterized The minimum positive angle in direction, θ_SRepresent the gradient direction of jth pixel and the midpoint institute of quantized interval S The minimum positive angle in the direction characterized, θ_TRepresent that the gradient direction of jth pixel is with quantized interval T's The minimum positive angle in the direction that midpoint is characterized.

Such as, being illustrated in figure 3 a kind of angular interval dividing mode, the most every two adjacent solid lines are one Individual angular interval, two dotted lines in figure represent that the angle of the first angular interval and the second angular interval is put down respectively Separated time, the chain-dotted line in figure represents the gradient direction of a certain pixel, and it falls in the second angular interval, Simultaneously also falling in the angular range of two dotted lines, therefore this pixel is to the first angular interval and second jiao All there is contribution in degree interval, and the contribution of the first angular interval is by itAnd the tribute to the second angular interval Offer for

Certainly, the statistics about gradient direction also has simpler mode, the gradient of the most a certain pixel Direction falls in the second angular interval, then this pixel only has contribution to the second angular interval, and to other Angular interval is all without contribution.Both statistical methods all can be applicable to all detailed description of the invention of the present invention In.

Specifically, color histogram carries out equal interval quantizing to color in HSV space and draws, color Square information is made up of the front third moment of color histogram, i.e. first moment (mean value Mean), second moment (variance Variance) and third moment (degree of bias Skewness), its calculation is respectively as follows:

First moment:

Second moment:

Third moment:

Wherein, N represents the sum of pixel, and i represents image channel, and (i.e. H in HSV space leads to Road, channel S or V passage), p_ijRepresent key frame jth pixel gray value under passage i.

Specifically, gradient direction interval division in the range of 0～180 degree with it at 180～360 degree of models Enclose interior interval division specular, such as start even circumferential is divided into 8 intervals from 0 degree.

Specifically, key frame is both from effective section of video to be identified, and effective section is to be identified regarding Frequency removes the remainder behind the beginning part and ending, and the duration of effective section is at least whole waiting to be known The 50% of other video duration, and the duration of the beginning part and ending all at least whole video to be identified The 8% of duration.Advantage of this is that: no matter general cartoon video is also that non-cartoon video all can have head And run-out, and head and run-out be probably captions, this can affect the algorithm recognition result to video, therefore Preferably the beginning and end of video is removed.

Specifically, characteristics of image is all from the effective coverage of key frame, and the area of effective coverage is at least For the 25% of whole key frame area, and effective coverage is positioned at the desirable region of key frame；Desirable region For having the similar figures of the key frame of common geometric center with key frame, and the area in desirable region is crucial The 64% of frame area.Advantage of this is that: may regard no matter cartoon video is also non-cartoon video The marginal portion of frequency has black surround and/or captions, in order to avoid black surround and the interference of captions, best during identification The geometry mid portion of selecting video key frame.

A kind of cartoon video identification device, it comprises:

Key-frame extraction module: for extracting key frame from video to be identified；

Image characteristics extraction module: for obtaining characteristics of image from key frame；

First sort module: for using the first sorting algorithm according to the box counting algorithm of each key frame The cartoon image degree of membership of this key frame；

Degree of membership distribution statistics module: for the span of cartoon image degree of membership is divided at least three Interval, and add up the interval distribution situation of the cartoon image degree of membership of all key frames；

Second sort module: use the second sorting algorithm to judge that video to be identified is according to interval distribution situation No for cartoon video.

Specifically, characteristics of image comprises color histogram, edge histogram, high luminance pixels ratio, edge Pixel ratio, and the color moment information obtained by color histogram；Edge histogram passes through statistical pixel Gradient direction and the gradient magnitude of point draw；High luminance pixels ratio refers to that in HSV space, V parameter is more than The ratio of the pixel of threshold X；Edge pixel ratio refers to that gradient magnitude is more than the pixel of threshold value Y Ratio.

Specifically, the calculation that edge histogram represents in the matrix form is:

B_mn=L_m×G_n,

In formula, B_mnRepresent the edge histogram matrix of m row n row, L_mRepresent the gradient magnitude Nogata of m row 1 row Figure matrix, G_nRepresent the gradient orientation histogram matrix of 1 row n row；

The calculation of above-mentioned gradient orientation histogram matrix is:

$G_n=\frac{1}{N}\underset{j=1}{\overset{N}{\Σ}}{Q}_{nj},$

In formula: N represents by the sum of statistical pixel point；Q_njIt is the matrix of 1 row n row, represents jth The pixel contribution to each gradient direction interval；

Above-mentioned Q_njCalculation be: the gradient direction setting jth pixel falls into quantized interval S, then it is right Quantized interval S and quantized interval T all has contribution, and to other quantized intervals all without contribution, described amount Changing interval T is quantization district minimum with the gradient direction angle of jth pixel in addition to quantized interval S Between；The jth pixel contribution v to quantized interval S_SWith the contribution v to quantized interval T_TCount as the following formula Calculate:

$v_S=\frac{{\mathrm{\θ}}_T}{{\mathrm{\γ}}_{ST}},$

$v_T=\frac{{\mathrm{\θ}}_S}{{\mathrm{\γ}}_{ST}},$

In formula, γ_STRepresent what the direction that characterized, midpoint of quantized interval S and the midpoint of quantized interval T were characterized The minimum positive angle in direction, θ_SRepresent the gradient direction of jth pixel and the midpoint institute of quantized interval S The minimum positive angle in the direction characterized, θ_TRepresent that the gradient direction of jth pixel is with quantized interval T's The minimum positive angle in the direction that midpoint is characterized.

Certainly, the statistics about gradient direction also has simpler mode, the gradient of the most a certain pixel Direction falls in the second angular interval, then this pixel only has contribution to the second angular interval, and to other Angular interval is all without contribution.Both statistical methods all can be applicable to the concrete of all embodiments of the present invention In practice.

Specifically, color histogram carries out equal interval quantizing to color in HSV space and draws, color Square information is made up of the front third moment of color histogram, i.e. first moment (mean value Mean), second moment (variance Variance) and third moment (degree of bias Skewness).

Specifically, gradient direction interval division in the range of 0～180 degree with it at 180～360 degree of models Enclose interior interval division specular.

Specifically, key frame is both from effective section of video to be identified, and effective section is to be identified regarding Frequency removes the remainder behind the beginning part and ending, and the duration of effective section is at least whole waiting to be known The 50% of other video duration, and the duration of the beginning part and ending all at least whole video to be identified The 8% of duration.Advantage of this is that: no matter general cartoon video is also that non-cartoon video all can have head And run-out, and head and run-out be probably captions, this can affect the algorithm recognition result to video, therefore Preferably the beginning and end of video is removed.

Specifically, characteristics of image is all from the effective coverage of key frame, and the area of effective coverage is at least For the 25% of whole key frame area, and effective coverage is positioned at the desirable region of key frame；Desirable region For having the similar figures of the key frame of common geometric center with key frame, and the area in desirable region is crucial The 64% of frame area.Advantage of this is that: may regard no matter cartoon video is also non-cartoon video The marginal portion of frequency has black surround and/or captions, in order to avoid black surround and the interference of captions, best during identification The geometry mid portion of selecting video key frame.

As an embodiment of the inventive method, as it is shown in figure 1, a kind of cartoon video recognition methods, The method comprises the steps of:

Step 101, intercepts the centre 80% duration part of video to be identified；

Step 102, extracts key frame from intercepting part；

Step 103, intercepts the region of 70% in the middle part of the length and width of key frame；

Step 104, is transformed into HSV space by the image intercepting region；

Step 105, in HSV space, statistics intercepts the color histogram in region, and H parameter is drawn equably Being divided into 8 intervals, S and V parameter is evenly divided into 6 intervals respectively；Statistics intercepts region simultaneously Edge histogram, wherein edge histogram is the gradient direction by statistical pixel point and gradient magnitude obtains Going out, gradient direction and gradient magnitude are all evenly divided into 8 intervals；

Step 106, adds up the high luminance pixels ratio intercepting region according to color histogram and edge histogram With edge pixel ratio；Wherein high luminance pixels ratio is the ratio of the V parameter pixel more than 0.5, limit Edge pixel ratio is the ratio that gradient magnitude is more than the pixel of threshold value 0.087；

Step 107, uses SVM (Support Vector Machine, support vector that precondition is crossed Machine) grader to intercept region classify, calculate intercept region cartoon image degree of membership；

Step 108, is uniformly divided into the span of degree of membership 5 intervals, adds up all key frames Degree of membership is distributed；

Step 109, inputs another SVM classifier through precondition by degree of membership distribution, thus counts Calculate the probability that this video is cartoon video, if probability is more than 50%, determine that it is cartoon video；

Step 110, exports result of determination.

As an embodiment of apparatus of the present invention, as in figure 2 it is shown, a kind of device, this cartoon video is known Other device 2 comprises key-frame extraction module 201, image characteristics extraction module the 202, first sort module 203, degree of membership distribution statistics module 204 and the second sort module 205；Degree of membership distribution statistics module 204 for being divided into 5 intervals by the span of cartoon image degree of membership, and add up cartoon image and be subordinate to The interval distribution situation of degree, uses the second sorting algorithm to classify further according to interval distribution situation, thus Judge whether video to be identified is cartoon video.During use, by video input key-frame extraction mould to be identified Block 201, the key frame extracted is passed to image characteristics extraction module by key-frame extraction module 201 202, image characteristics extraction module 202 extracts characteristics of image from key frame, and its extracting method can use Any of which mentioned in the inventive method, then image characteristics extraction module 202 is by characteristics of image Passing to the first sort module 203, the first sort module 203 calculates the cartoon image of each key frame Degree of membership, and pass the result to degree of membership distribution statistics module 204, degree of membership distribution statistics module 204 Statistics passes to the second sort module 205, and the classified calculating through the second sort module 205 is sentenced Whether disconnected video to be sorted is cartoon video, finally exports video type.

Being readily apparent that, implementing of apparatus of the present invention both can be a kind of special equipment, it is also possible to be The equipment installing specific software on the smart machines such as computer, mobile phone, flat board and formed.

It should be noted that scope is made that in described above the parameters of restriction, within the range Choose any endpoint value or median is all desirable, and the different valued combinations of each parameter are also feasible 's.After recognizing specific embodiment of the invention limited range, those skilled in the art need not Paying any creative work and each parameter can be carried out concrete value, it is obtained Application effect is all without departing from the scope described in the present invention, and therefore, in order to save length, inventor is not Again various possible values and possible combination thereof are enumerated.

The device of above-described embodiment is used for the corresponding method in previous embodiment that realizes, and has corresponding The beneficial effect of embodiment of the method, does not repeats them here.

Those of ordinary skill in the field are it is understood that the discussion of any of the above embodiment is merely illustrative , it is not intended that hint the scope of the present disclosure (including claim) is limited to these examples；At this Under bright thinking, between the technical characteristic in above example or different embodiment, group can also be carried out Closing, step can realize with random order, and there is the many of the different aspect of the present invention as above Other change, for they not offers in details simple and clear.

It addition, for simplifying explanation and discussing, and in order to obscure the invention, provided Accompanying drawing in can illustrate or can not illustrate with integrated circuit (IC) chip and other parts known Power supply/grounding connection.Furthermore, it is possible to illustrate device in block diagram form, in order to avoid making the present invention be difficult to Understand, and this have also contemplated that following facts, i.e. the details about the embodiment of these block diagram arrangements is Depend highly on will implement the present invention platform (that is, these details should be completely in this area skill In the range of the understanding of art personnel).Elaborating that detail (such as, circuit) is to describe showing of the present invention In the case of example embodiment, it will be apparent to those skilled in the art that can there is no this In the case of a little details or these details change in the case of implement the present invention.Therefore, These descriptions are considered as illustrative and not restrictive.

Although invention has been described to have been incorporated with the specific embodiment of the present invention, but according to front The description in face, a lot of replacements, amendment and the modification of these embodiments are for those of ordinary skills Will be apparent from.Such as, other memory architecture (such as, dynamic ram (DRAM)) can The embodiment discussed with use.

Embodiments of the invention be intended to fall within the broad range of claims all so Replacement, amendment and modification.Therefore, all within the spirit and principles in the present invention, any province done Summary, amendment, equivalent, improvement etc., should be included within the scope of the present invention.

Claims (14)

1. a cartoon video recognition methods, it is characterised in that comprise:

Key frame is extracted from video to be identified；

Characteristics of image is obtained from key frame；

Use the first sorting algorithm according to the cartoon image of this key frame of box counting algorithm of each key frame Degree of membership；

The span of described cartoon image degree of membership is divided at least three interval, and adds up all keys The interval distribution situation of the cartoon image degree of membership of frame；

The second sorting algorithm is used to judge whether video to be identified is card intervisibility according to described interval distribution situation Frequently.

Cartoon video recognition methods the most according to claim 1, it is characterised in that: described image is special Levy and comprise color histogram, edge histogram, high luminance pixels ratio, edge pixel ratio and color moment letter Breath, wherein color moment information is calculated by described color histogram；Described edge histogram is by statistics Gradient direction and the gradient magnitude of pixel draw；

Described high luminance pixels ratio refers to that in HSV space, V parameter is more than the ratio of the pixel of threshold X； Described edge pixel ratio refers to that gradient magnitude is more than the ratio of the pixel of threshold value Y.

Cartoon video recognition methods the most according to claim 2, it is characterised in that: described edge is straight The calculation just scheming to represent in the matrix form is:

B_mn=L_m×G_n,

In formula, B_mnRepresent the edge histogram matrix of m row n row, L_mRepresent the gradient magnitude of m row 1 row Histogram matrix, G_nRepresent the gradient orientation histogram matrix of 1 row n row；

The calculation of described gradient orientation histogram matrix is:

$G_n=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{Q}_{nj},$

In formula: N represents by the sum of statistical pixel point；Q_njIt is the matrix of 1 row n row, represents the J pixel is to the interval contribution of each gradient direction；

Q_njCalculation be: the gradient direction setting jth pixel falls into quantized interval S, then it is right Quantized interval S and quantized interval T all has contribution, and to other quantized intervals all without contribution, described amount Changing interval T is quantization district minimum with the gradient direction angle of jth pixel in addition to quantized interval S Between；The jth pixel contribution v to quantized interval S_SWith the contribution v to quantized interval T_TCount as the following formula Calculate:

$v_S=\frac{{\mathrm{\θ}}_T}{{\mathrm{\γ}}_{ST}},$

$v_T=\frac{{\mathrm{\θ}}_S}{{\mathrm{\γ}}_{ST}},$

In formula, γ_STRepresent the direction that characterized, midpoint of quantized interval S and the midpoint institute table of quantized interval T The angle in the direction levied, θ_SRepresent the gradient direction of jth pixel and the midpoint institute table of quantized interval S The angle in the direction levied, θ_TRepresent the gradient direction of jth pixel and the midpoint institute table of quantized interval T The angle in the direction levied.

Cartoon video recognition methods the most according to claim 2, it is characterised in that: described color is straight Side's figure carries out equal interval quantizing to color in HSV space and draws, described color moment information is by described face The front third moment composition of Color Histogram.

Cartoon video recognition methods the most according to claim 2, it is characterised in that: described gradient side To the interval division in the range of 0～180 degree and its interval division mirror image in the range of 180～360 degree Symmetrical.

Cartoon video recognition methods the most according to claim 1, it is characterised in that: described key frame Both from effective section of described video to be identified, described effective section is that video to be identified removes beginning Remainder after part and ending, the duration of described effective section is at least whole video to be identified The 50% of duration, and the duration of described the beginning part and described ending is all at least whole to be identified regard Frequently the 8% of duration.

Cartoon video recognition methods the most according to claim 1, it is characterised in that: described image is special Levying the effective coverage all from key frame, the area of described effective coverage is at least whole crucial pattern Long-pending 25%, and effective coverage is positioned at the desirable region of key frame；Described desirable region is and key frame There are the similar figures of the key frame of common geometric center, and the area in desirable region is key frame area 64%.

8. a cartoon video identification device, it is characterised in that comprise:

Key-frame extraction module: for extracting key frame from video to be identified；

Image characteristics extraction module: for obtaining characteristics of image from key frame；

First sort module: for using the first sorting algorithm according to the box counting algorithm of each key frame The cartoon image degree of membership of this key frame；

Degree of membership distribution statistics module: for the span of described cartoon image degree of membership is divided at least Three intervals, and add up the interval distribution situation of the cartoon image degree of membership of all key frames；

Second sort module: use the second sorting algorithm to judge that video to be identified is according to described interval distribution situation No for cartoon video.

Cartoon video identification device the most according to claim 8, it is characterised in that: described image is special Levy and comprise color histogram, edge histogram, high luminance pixels ratio, edge pixel ratio and color moment letter Breath, wherein color moment information is calculated by color histogram；Described edge histogram passes through statistical pixel Gradient direction and the gradient magnitude of point draw；Described high luminance pixels ratio refers to that in HSV space, V parameter is big Ratio in the pixel of threshold X；Described edge pixel ratio refers to that gradient magnitude is more than the picture of threshold value Y The ratio of vegetarian refreshments.

Cartoon video identification device the most according to claim 9, it is characterised in that: described edge The calculation that histogram represents in the matrix form is:

B_mn=L_m×G_n,

In formula, B_mnRepresent the edge histogram matrix of m row n row, L_mRepresent the gradient magnitude of m row 1 row Histogram matrix, G_nRepresent the gradient orientation histogram matrix of 1 row n row；

The calculation of described gradient orientation histogram matrix is:

$G_n=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{Q}_{nj},$

In formula: N represents by the sum of statistical pixel point；Q_njIt is the matrix of 1 row n row, represents the J pixel is to the interval contribution of each gradient direction；

Q_njCalculation be: the gradient direction setting jth pixel falls into quantized interval S, then it is right Quantized interval S and quantized interval T all has contribution, and to other quantized intervals all without contribution, described amount Changing interval T is quantization district minimum with the gradient direction angle of jth pixel in addition to quantized interval S Between；Quantized interval S is contributed vS and the contribution v to quantized interval T by jth pixel_TCount as the following formula Calculate:

$v_S=\frac{{\mathrm{\θ}}_T}{{\mathrm{\γ}}_{ST}},$

$v_T=\frac{{\mathrm{\θ}}_S}{{\mathrm{\γ}}_{ST}},$

In formula, γ_STRepresent the direction that characterized, midpoint of quantized interval S and the midpoint institute table of quantized interval T The angle in the direction levied, θ_SRepresent the gradient direction of jth pixel and the midpoint institute table of quantized interval S The angle in the direction levied, θ_TRepresent the gradient direction of jth pixel and the midpoint institute table of quantized interval T The angle in the direction levied.

11. cartoon video identification devices according to claim 9, it is characterised in that: described color Histogram carries out equal interval quantizing to color in HSV space and draws, described color moment information is by described The front third moment composition of color histogram.

12. cartoon video identification devices according to claim 9, it is characterised in that: described gradient Direction interval division in the range of 0～180 degree and its interval division mirror in the range of 180～360 degree As symmetrical.

13. cartoon video identification devices according to claim 9, it is characterised in that: described key Frame is both from effective section of described video to be identified, and described effective section is that video to be identified removes out Remainder after head part and ending, the duration of described effective section is at least whole to be identified regard Frequently the 50% of duration, and the duration of described the beginning part and described ending is all at least whole to be identified The 8% of video duration.

14. cartoon video identification devices according to claim 9, it is characterised in that: described image Feature is at least whole key frame all from the effective coverage of key frame, the area of described effective coverage The 25% of area, and effective coverage is positioned at the desirable region of key frame；Described desirable region is and key Frame has the similar figures of the key frame of common geometric center, and the area in desirable region is key frame area 64%.