CN101853286A - Intelligent selection method of video thumbnails - Google Patents

Abstract

The invention discloses an intelligent selection method of video thumbnails, comprising the following steps: 1) acquiring N numbered thumbnails of a video, wherein N is more than or equal to 2 and less than or equal to 100, and N is a constant; 2) calculating the characteristics of each thumbnail, such as frontal face position and size, profile face position and size, whole body position and size, upper body position and size, lower body position and size, gray level histogram distribution value, gray level histogram variance distribution value, colorful RGB histogram variance distribution value, difference value that a colorful RGB image is changed into a gray level image, the variance value of a gray level image pixel point, the variance value of the colorful RGB image pixel point, the dispersion distance value between the colorful RGB blocks, and the like; 3) calculating the characteristic values of each thumbnail respectively through ambiguity functions; 4) weighting and calculating the characteristic values of each thumbnail to obtain a final fraction; 5) sequencing the N numbered thumbnails according to the fraction of each thumbnail; and 6) selecting the thumbnail with the highest fraction as the thumbnail representing the video.

Description

Intelligent selection method of video thumbnails

Technical field

The present invention relates to the picture choosing method, particularly a kind of intelligent selection method of video thumbnails that is undertaken by Computer Image Processing.

Background technology

In the daily use of video website, it is exactly to extract the representative picture of a two field picture as this video file from video file that an important techniques is arranged, and we claim that this picture is the thumbnail of video file.Our way is to extract eight hypertonic sketch maps from video file at present, and the 4th thumbnail as acquiescence is set.This employing fixed position method is given tacit consent to and the thumbnail that obtains is not the most representative thumbnail in the eight hypertonic sketch maps usually.We have adopted certain methods to address this problem, and can determine the thumbnail of which pictures as acquiescence by self-defining method such as (1) user, and (2) are given the website staff and selected the right of which pictures as the acquiescence thumbnail.Though said method can address this problem, but reality still can't meet the demands, this is by data analysis because of (1), we know the user usually not can or seldom self-defined thumbnail (2) because to upload to the video file quantity of video website every day very numerous, the workload that defines thumbnail by the website staff is very heavy.Based on above-mentioned two reasons, we invent and a kind ofly can select the method for thumbnail of representative thumbnail as acquiescence automatically by the method for Computer Image Processing.

People such as P.Viola and M.Jones " adopting the stacked method for improving of simple feature to carry out quick object detection " (Rapid Object Detection Using a Boosted Cascade of SimpleFeatures.) method that (Proceedings ofComputer Vision and Pattern Recognition, 2001) propose on the computer vision and pattern-recognition international conference of calendar year 2001 detects.

In view of this, those skilled in the art provide a kind of and can select the intelligent selection method of video thumbnails of representative thumbnail as the thumbnail of acquiescence automatically by the Computer Image Processing method at the problems referred to above.

Summary of the invention

The invention provides a kind of intelligent selection method of video thumbnails, overcome the difficulty of prior art, select the purpose of representative thumbnail automatically as the thumbnail of acquiescence to reach by the Computer Image Processing method.

The present invention adopts following technical scheme:

The invention provides a kind of intelligent selection method of video thumbnails, be used for selecting a thumbnail of representing this video, may further comprise the steps from one section video:

(1) obtain the N hypertonic sketch map of one section video, wherein, 2≤N≤100, and N is a constant;

(2) every hypertonic sketch map is calculated the position and the size of its front face, the position and the size of people from side face, the position of whole body human body and size, the position and the size of the position of upper body and the size and the lower part of the body, grey level histogram distribution value, grey level histogram variance distribution value, colored RGB histogram variance distribution value, colored RGB image becomes the difference value of gray level image, the variance yields of gray level image pixel, the variance yields of colored RGB image slices vegetarian refreshments, discrete distance value between the colored RGB piece or the like feature;

(3) every hypertonic sketch map is tried to achieve the score value of above-mentioned each feature respectively by ambiguity function;

(4) the feature score value to every hypertonic sketch map obtains a final mark by the weighting evaluation;

(5) size according to mark sorts to N hypertonic sketch map;

(6) select for use the highest thumbnail of mark as the thumbnail of representing this video.

Preferably, in the described step (1), adopt the extracted at equal intervals method, the video duration according to obtaining in advance is divided into the N equal portions with it, and each equal portions extracts a frame as thumbnail then, and N=8.

Preferably, may further comprise the steps in the described step (2):

(21) the input thumbnail detects face characteristic, comprises position and the position and the size big or small, people from side face of front face, and calculates fractional value;

(22) input thumbnail, the human body feature comprises position and size, the position of upper body and the position and the size of the size and the lower part of the body of whole body human body, and calculates fractional value;

(23) input thumbnail calculates grey level histogram distribution value;

(24) input thumbnail calculates grey level histogram variance distribution value;

(25) the input thumbnail calculates colored RGB histogram variance distribution value;

(26) the input thumbnail calculates the difference value that colored RGB image becomes gray level image;

(27) input thumbnail, the variance yields of calculating gray level image pixel;

(28) import thumbnail, calculate the variance yields of colored RGB image slices vegetarian refreshments;

(29) the input thumbnail calculates the discrete distance value between the colored RGB piece.

Preferably, in the described step (21):

The score value of front face then

W wherein _iWide for front face, n is people's face number, establishing W is the wide of thumbnail;

The score value of people from side face

W wherein _iWide for side people's face, n is people's face number, establishing W is the wide of thumbnail.

Preferably, establish (x in the described step (22) _i, y _i, w _i, h _i), i=1 ..., n, wherein x _i, y _iRepresentative's face position, w _iBehaviour face wide, n is people's face number, establishing W is the wide of thumbnail,

The score value of whole body then

W wherein _iWide for the whole body human body, n is a number, establishing W is the wide of thumbnail;

The score value of upper body

W wherein _iWide for the whole body human upper limb, n is a number, establishing W is the wide of thumbnail;

The score value of the lower part of the body

W wherein _iWide for the whole body human body lower limbs, n is a number, establishing W is the wide of thumbnail.

Preferably, input thumbnail in the described step (23), establishing its grey level histogram is gray_hist=(gh_1, gh_2, .., gh_n), n is histogrammic number, t1, t2 is the integer threshold value, T1 and T2 be histogram and threshold value, base_score is the benchmark mark, then the computing formula of grey level histogram distribution value score_hist_distrib is:

Wherein, n=10, t1=2, t2=8, T ₁=0.7, T ₁=0.8, base_score=1.0.

Preferably, input thumbnail in the described step (24), establish its grey level histogram and be gray_hist=(gh_1, gh_2 .., gh_n), n is histogrammic number, Max_num=max (gh_1, gh_2 .., gh_n), base_score is the benchmark mark, the score_gray_hist_std computing formula of then calculating grey level histogram variance distribution value is:

Wherein, n=64, base_score=1.0.

Preferably, histogram variance distribution value on R plane, G plane, B plane respectively in the described step (25), averaging then obtains colored RGB histogram variance distribution value score_RGB_hist_std;

$\mathrm{score}\_\mathrm{RGB}\_\mathrm{hist}\_\mathrm{distrib}=\frac{\mathrm{score}\_R\_\mathrm{hist}\_\mathrm{distrib}+\mathrm{score}\_G\_\mathrm{hist}\_\mathrm{distrib}+\mathrm{score}\_B\_\mathrm{hist}\_\mathrm{distrib}}{3}$

Wherein, n=64, base_score=1.0.

Preferably, input thumbnail in the described step (26), the pixel value of establishing gray level image is x _Ij, i=1 ..., W, j=1 ..., H, the pixel value (r of colored RGB image _Ij, g _Ij, b _Ij), i=1 ..., W, j=1 ..., H, wherein W, H is respectively the wide and high of thumbnail, and base_score1 and base_score2 are the benchmark mark, and t1 is a threshold value, establishes then to calculate the computing formula that colored RGB image becomes the difference value score_rgb2gray_diff of gray level image and be:

Wherein,

Base_score1=1.0, base_score2=2.0, t ₁=-0.9.

Preferably, input thumbnail in the described step (27), the pixel value of establishing gray level image is x _Ij, i=1 ..., W, j=1 ..., H, W wherein, H is respectively the wide and high of thumbnail, and the score_gray_std computing formula of then calculating the variance yields of gray level image pixel is:

$\mathrm{score}\_\mathrm{gray}\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}}.$

Preferably, in the described step (28) respectively on the R plane, the variance yields of G plane, B plane epigraph pixel, averaging then obtains the variance yields score_RGB_std of colored RGB image slices vegetarian refreshments;

$\mathrm{score}\_R\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}};$

$\mathrm{score}\_G\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}};$

$\mathrm{score}\_B\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}};$

$\mathrm{score}\_\mathrm{RGB}\_\mathrm{std}=\frac{\mathrm{score}\_R\_\mathrm{std}+\mathrm{score}\_G\_\mathrm{std}+\mathrm{score}\_B\_\mathrm{std}}{3}.$

Preferably, import thumbnail in the described step (29), establish the pixel value (r of colored RGB image _i, g _i, b _i), i=1 ..., W * H, W wherein, H is respectively the wide and high of thumbnail, and base_score is the benchmark mark, and the computing formula of then calculating the discrete distance value score_rgb_dist between the colored RGB piece is:

Wherein, base_score=0.0.

Preferably, final score value final_score is calculated by following formula in the described step (4):

final_score＝w ₁×score_face_frontal+w ₂×score_face_profile+w ₃×score_full_body+w ₄×score_upper_body+w ₅×score_lower_body+w ₆×score_hist_distrib+w ₇×score_gray_hist_std+w ₈×score_RGB_hist_std+w ₉×score_rgb2gray_diff+w ₁₀×score_gray_std+w ₁₁×score_RGB_std；

Wherein, w ₁=1.0, w ₂=0.5, w ₃=0.5, w ₄=0.5, w ₅=0.5, w ₆=0.5, w ₇=0.5, w ₈=1.2, W ₉=1.2, w ₁₀=0.5, w ₁₁=1.2.

Owing to adopted above-mentioned technology, compared with prior art, the present invention has following advantage: the present invention can select the thumbnail of representative thumbnail as acquiescence automatically by the Computer Image Processing method.

Further specify the present invention below in conjunction with drawings and Examples.

Description of drawings

Fig. 1 is the process flow diagram of intelligent selection method of video thumbnails of the present invention.

Embodiment

Introduce a kind of specific embodiment of the present invention below by Fig. 1.

As shown in Figure 1, a kind of intelligent selection method of video thumbnails may further comprise the steps:

(1) obtain the N hypertonic sketch map of a video, the acquisition methods of N hypertonic sketch map is to adopt the extracted at equal intervals method, promptly according to the video duration that obtains in advance, it is divided into the N equal portions, and each equal portions extracts a frame as thumbnail then.Here get N=8, but be not limited to N=8, N can be between 2 to 100.

(2) every hypertonic sketch map is calculated the position and the size of its front face, the position and the size of people from side face, the position of whole body human body and size, the position and the size of the position of upper body and the size and the lower part of the body, grey level histogram distribution value, histogram variance distribution value, colored RGB histogram variance distribution value, colored RGB image becomes the difference value of gray level image, the variance yields of gray level image pixel, the variance yields of colored RGB image slices vegetarian refreshments, discrete distance value between the colored RGB piece or the like feature.

(3) try to achieve the score value of above-mentioned feature respectively by ambiguity function.

(4) obtain a final mark by the weighting evaluation.

(5) size according to mark sorts to N hypertonic sketch map.

(6) thumbnail that score value is the highest is as the thumbnail of acquiescence.

Above-mentioned steps (2), carry out as follows:

(21) the input thumbnail detects face characteristic, includes position and the position and the size big or small, people from side face of front face; Calculate fractional value;

(22) input thumbnail, the human body feature includes the position of whole body human body and size, the position of upper body and the position and the size of the size and the lower part of the body; Calculate fractional value;

(23) input thumbnail calculates grey level histogram distribution value;

(24) input thumbnail calculates grey level histogram variance distribution value;

(25) the input thumbnail calculates colored RGB histogram variance distribution value;

(26) the input thumbnail calculates the difference value that colored RGB image becomes gray level image;

(27) input thumbnail, the variance yields of calculating gray level image pixel;

(28) import thumbnail, calculate the variance yields of colored RGB image slices vegetarian refreshments;

(29) the input thumbnail calculates the discrete distance value between the colored RGB piece.

Embodiments of the present invention are as follows:

Continuation is referring to Fig. 1, and a kind of intelligent selection method of video thumbnails of the present invention is used for selecting a thumbnail of representing this video from one section video, may further comprise the steps:

Step (1) is obtained the N hypertonic sketch map of one section video, wherein, and 2≤N≤100, and N is constant, adopts the extracted at equal intervals method, and the video duration according to obtaining in advance is divided into the N equal portions with it, and each equal portions extracts a frame as thumbnail then, and N=8.

Step (2) is calculated the position and the size of its front face to every hypertonic sketch map, the position and the size of people from side face, the position of whole body human body and size, the position and the size of the position of upper body and the size and the lower part of the body, grey level histogram distribution value, grey level histogram variance distribution value, colored RGB histogram variance distribution value, colored RGB image becomes the difference value of gray level image, the variance yields of gray level image pixel, the variance yields of colored RGB image slices vegetarian refreshments, discrete distance value between the colored RGB piece or the like feature;

Described step may further comprise the steps in (2):

Step (21) input thumbnail detects face characteristic, comprises position and the position and the size big or small, people from side face of front face, and calculates fractional value;

Step (22) input thumbnail, the human body feature comprises position and size, the position of upper body and the position and the size of the size and the lower part of the body of whole body human body, and calculates fractional value;

Step (23) input thumbnail calculates grey level histogram distribution value;

Step (24) input thumbnail calculates grey level histogram variance distribution value;

Step (25) input thumbnail calculates colored RGB histogram variance distribution value;

Step (26) input thumbnail calculates the difference value that colored RGB image becomes gray level image;

Step (27) input thumbnail, the variance yields of calculating gray level image pixel;

Step (28) is imported thumbnail, calculates the variance yields of colored RGB image slices vegetarian refreshments;

Step (29) input thumbnail calculates the discrete distance value between the colored RGB piece.

In the described step (21): the score value of front face then W wherein _iWide for front face, n is people's face number, establishing W is the wide of thumbnail; The score value of people from side face

W wherein _iWide for side people's face, n is people's face number, establishing W is the wide of thumbnail.

Described step is established (x in (22) _i, y _i, w _i, h _i), i=1 ..., n, wherein x _i, y _iRepresentative's face position, w _iBehaviour face wide, n is people's face number, establishing W is the wide of thumbnail,

The score value of whole body then

W wherein _iWide for the whole body human body, n is a number, establishing W is the wide of thumbnail;

The score value of upper body

W wherein _iWide for the whole body human upper limb, n is a number, establishing W is the wide of thumbnail;

The score value of the lower part of the body

W wherein _iWide for the whole body human body lower limbs, n is a number, establishing W is the wide of thumbnail.

Input thumbnail in the described step (23), if its grey level histogram is gray_hist=(gh_1, gh_2 .., gh_n), n is histogrammic number, t1, t2 are the integer threshold value, T1 and T2 be histogram and threshold value, base_score is the benchmark mark, and then the computing formula of grey level histogram distribution value score_hist_distrib is:

Wherein, n=10, t1=2, t2=8, T ₁=0.7, T ₁=0.8, base_score=1.0.

Input thumbnail in the described step (24), establish its grey level histogram and be gray_hist=(gh_1, gh_2 .., gh_n), n is histogrammic number,

Max_num=max (gh_1, gh_2 .., gh_n), base_score is the benchmark mark, the score_gray_hist_std computing formula of then calculating grey level histogram variance distribution value is:

Wherein, n=64, base_score=1.0.

Histogram variance distribution value on R plane, G plane, B plane respectively in the described step (25), averaging then obtains colored RGB histogram variance distribution value score_RGB_hist_std;

$\mathrm{score}\_\mathrm{RGB}\_\mathrm{hist}\_\mathrm{distrib}=\frac{\mathrm{score}\_R\_\mathrm{hist}\_\mathrm{distrib}+\mathrm{score}\_G\_\mathrm{hist}\_\mathrm{distrib}+\mathrm{score}\_B\_\mathrm{hist}\_\mathrm{distrib}}{3}$

Wherein, n=64, base_score=1.0.

Input thumbnail in the described step (26), the pixel value of establishing gray level image is x _Ij, i=1 ..., W, j=1 ..., H, the pixel value (r of colored RGB image _Ij, g _Ij, b _Ij), i=1 ..., W, j=1 ..., H, wherein W, H is respectively the wide and high of thumbnail, and base_score1 and base_score2 are the benchmark mark, and t1 is a threshold value, establishes then to calculate the computing formula that colored RGB image becomes the difference value score_rgb2gray_diff of gray level image and be:

Wherein, Base_score1=1.0, base_score2=2.0, t ₁=-0.9.

Input thumbnail in the described step (27), the pixel value of establishing gray level image is x _Ij, i=1 ..., W, j=1 ..., H, W wherein, H is respectively the wide and high of thumbnail, and the score_gray_std computing formula of then calculating the variance yields of gray level image pixel is:

$\mathrm{score}\_\mathrm{gray}\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}}.$

In the described step (28) respectively on the R plane, the variance yields of G plane, B plane epigraph pixel, averaging then obtains the variance yields score_RGB_std of colored RGB image slices vegetarian refreshments;

$\mathrm{score}\_R\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}};$

$\mathrm{score}\_G\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}};$

$\mathrm{score}\_B\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}};$

$\mathrm{score}\_\mathrm{RGB}\_\mathrm{std}=\frac{\mathrm{score}\_R\_\mathrm{std}+\mathrm{score}\_G\_\mathrm{std}+\mathrm{score}\_B\_\mathrm{std}}{3}.$

Import thumbnail in the described step (29), establish the pixel value (r of colored RGB image _i, g _i, b _i), i=1 ..., W * H, W wherein, H is respectively the wide and high of thumbnail, and base_score is the benchmark mark, and the computing formula of then calculating the discrete distance value score_rgb_dist between the colored RGB piece is:

$\mathrm{score}\_\mathrm{rgb}\_\mathrm{dist}=-\mathrm{base}\_\mathrm{score}+\underset{i=1}{\overset{W\×H}{\mathrm{\Σ}}}\underset{j=1,j\≠i}{\overset{W\×H}{\mathrm{\Σ}}}\left(\right|r_i-r_j|+|g_i-g_j|+|b_i-b_j\left|\right)/(3\×W\×H\×255),$ Wherein, base_score=0.0.

Step (3) is tried to achieve the score value of above-mentioned each feature respectively by ambiguity function to every hypertonic sketch map;

Step (4) obtains a final mark to the feature score value of every hypertonic sketch map by the weighting evaluation, and final score value final_score is calculated by following formula:

final_score＝w ₁×score_face_frontal+w ₂×score_face_profile+w ₃×score_full_body+w ₄×score_upper_body+w ₅×score_lower_body+w ₆×score_hist_distrib+w ₇×score_gray_hist_std+w ₈×score_RGB_hist_std+w ₉×score_rgb2gray_diff+w ₁₀×score_gray_std+w ₁₁×score_RGB_std；

Wherein, w ₁=1.0, w ₂=0.5, w ₃=0.5, w ₄=0.5, w ₅=0.5, w ₆=0.5, w ₇=0.5, w ₈=1.2, W ₉=1.2, w ₁₀=0.5, w ₁₁=1.2.

Step (5) sorts to N hypertonic sketch map according to the size of mark;

Step (6) selects for use the highest thumbnail of mark as the thumbnail of representing this video.

In summary, owing to adopted above-mentioned technology, the present invention has following advantage: the present invention can select the thumbnail of representative thumbnail as acquiescence automatically by the Computer Image Processing method.

Above-described embodiment only is used to illustrate technological thought of the present invention and characteristics, its purpose is to make those skilled in the art can understand content of the present invention and implements according to this, can not only limit claim of the present invention with present embodiment, be all equal variation or modifications of doing according to disclosed spirit, still drop in the claim of the present invention.

Claims (13)

1. an intelligent selection method of video thumbnails is used for selecting a thumbnail of representing this video from one section video, it is characterized in that: may further comprise the steps:

(1) obtain the N hypertonic sketch map of one section video, wherein, 2≤N≤100, and N is a constant;

(2) every hypertonic sketch map is calculated the position and the size of its front face, the position and the size of people from side face, the position of whole body human body and size, the position and the size of the position of upper body and the size and the lower part of the body, grey level histogram distribution value, grey level histogram variance distribution value, colored RGB histogram variance distribution value, colored RGB image becomes the difference value of gray level image, the variance yields of gray level image pixel, the variance yields of colored RGB image slices vegetarian refreshments, discrete distance value between the colored RGB piece or the like feature;

(3) every hypertonic sketch map is tried to achieve the score value of above-mentioned each feature respectively by ambiguity function;

(4) the feature score value to every hypertonic sketch map obtains a final mark by the weighting evaluation;

(5) size according to mark sorts to N hypertonic sketch map;

(6) select for use the highest thumbnail of mark as the thumbnail of representing this video.

2. intelligent selection method of video thumbnails as claimed in claim 1 is characterized in that: in the described step (1), adopt the extracted at equal intervals method, according to the video duration that obtains in advance, it is divided into the N equal portions, and each equal portions extracts a frame as thumbnail then, and N=8.

3. intelligent selection method of video thumbnails as claimed in claim 2 is characterized in that: described step may further comprise the steps in (2):

(21) the input thumbnail detects face characteristic, comprises position and the position and the size big or small, people from side face of front face, and calculates fractional value;

(22) input thumbnail, the human body feature comprises position and size, the position of upper body and the position and the size of the size and the lower part of the body of whole body human body, and calculates fractional value;

(23) input thumbnail calculates grey level histogram distribution value;

(24) input thumbnail calculates grey level histogram variance distribution value;

(25) the input thumbnail calculates colored RGB histogram variance distribution value;

(26) the input thumbnail calculates the difference value that colored RGB image becomes gray level image;

(27) input thumbnail, the variance yields of calculating gray level image pixel;

(28) import thumbnail, calculate the variance yields of colored RGB image slices vegetarian refreshments;

(29) the input thumbnail calculates the discrete distance value between the colored RGB piece.

4. intelligent selection method of video thumbnails as claimed in claim 3 is characterized in that: in the described step (21):

The score value of front face then

W wherein _iWide for front face, n is people's face number, establishing W is the wide of thumbnail;

The score value of people from side face

W wherein _iWide for side people's face, n is people's face number, establishing W is the wide of thumbnail.

5. intelligent selection method of video thumbnails as claimed in claim 3 is characterized in that: described step is established (x in (22) _i, y _i, w _i, h _i), i=1 ..., n, wherein x _i, y _iRepresentative's face position, w _iBehaviour face wide, n is people's face number, establishing W is the wide of thumbnail,

The score value of whole body then

W wherein _iWide for the whole body human body, n is a number, establishing W is the wide of thumbnail;

The score value of upper body

W wherein _iWide for the whole body human upper limb, n is a number, establishing W is the wide of thumbnail;

The score value of the lower part of the body

W wherein _iWide for the whole body human body lower limbs, n is a number, establishing W is the wide of thumbnail.

6. intelligent selection method of video thumbnails as claimed in claim 3, it is characterized in that: input thumbnail in the described step (23), establishing its grey level histogram is gray_hist=(gh_1, gh_2, .., gh_n), n is histogrammic number, t1, t2 is the integer threshold value, T1 and T2 be histogram and threshold value, base_score is the benchmark mark, then the computing formula of grey level histogram distribution value score_hist_distrib is:

Wherein, n=10, t1=2, t2=8, T ₁=0.7, T ₁=0.8, base_score=1.0.

7. intelligent selection method of video thumbnails as claimed in claim 3 is characterized in that: input thumbnail in the described step (24), establish its grey level histogram and be gray_hist=(gh_1, gh_2 .., gh_n), n is histogrammic number, Max_num=max (gh_1, gh_2 .., gh_n), base_score is the benchmark mark, the score_gray_hist_std computing formula of then calculating grey level histogram variance distribution value is:

When max num ≠ 0 wherein, n=64, base_score=1.0.

8. intelligent selection method of video thumbnails as claimed in claim 3, it is characterized in that: histogram variance distribution value on R plane, G plane, B plane respectively in the described step (25), averaging then obtains colored RGB histogram variance distribution value score_RGB_hist_std;

When max_num ≠ 0;

When max_num ≠ 0;

When max_num ≠ 0;

$\mathrm{score}\_\mathrm{RGB}\_\mathrm{hist}\_\mathrm{distrib}=\frac{\mathrm{score}\_R\_\mathrm{hist}\_\mathrm{distrib}+\mathrm{score}\_G\_\mathrm{hist}\_\mathrm{distrib}+\mathrm{score}\_B\_\mathrm{hist}\_\mathrm{distrib}}{3}$ Wherein, n=64, base_score=1.0.

9. intelligent selection method of video thumbnails as claimed in claim 3 is characterized in that: input thumbnail in the described step (26), the pixel value of establishing gray level image is x _Ij, i=1 ..., W, j=1 ..., H, the pixel value (r of colored RGB image _Ij, g _Ij, b _Ij), i=1 ..., W, j=1 ..., H, wherein W, H is respectively the wide and high of thumbnail, and base_score1 and base_score2 are the benchmark mark, and t1 is a threshold value, establishes then to calculate the computing formula that colored RGB image becomes the difference value score_rgb2gray_diff of gray level image and be:

Wherein, Base_score1=1.0, base_score2=2.0, t ₁=-0.9.

10. intelligent selection method of video thumbnails as claimed in claim 3 is characterized in that: input thumbnail in the described step (27), the pixel value of establishing gray level image is x _Ij, i=1 ..., W, j=1 ..., H, W wherein, H is respectively the wide and high of thumbnail, and the score_gray_std computing formula of then calculating the variance yields of gray level image pixel is:

$\mathrm{score}\_\mathrm{gray}\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}\stackrel{}{{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}}.$

11. intelligent selection method of video thumbnails as claimed in claim 3, it is characterized in that: in the described step (28) respectively on the R plane, the variance yields of G plane, B plane epigraph pixel, averaging then obtains the variance yields score_RGB_std of colored RGB image slices vegetarian refreshments;

$\mathrm{score}\_R\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}\stackrel{}{{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}};$

$\mathrm{score}\_G\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}\stackrel{}{{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}};$

$\mathrm{score}\_B\_\mathrm{std}=-\mathrm{base}\_\mathrm{score}+\sqrt{\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}\stackrel{}{{(x_{\mathrm{ij}}-\stackrel{\‾}{x})}^2}}/255,$ Wherein $\stackrel{\‾}{x}=\underset{i=1}{\overset{W}{\mathrm{\Σ}}}\underset{j=1}{\overset{H}{\mathrm{\Σ}}}{x}_{\mathrm{ij}};$

$\mathrm{score}\_\mathrm{RGB}\_\mathrm{std}=\frac{\mathrm{score}\_R\_\mathrm{std}+\mathrm{score}\_G\_\mathrm{std}+\mathrm{score}\_B\_\mathrm{std}}{3}.$

12. intelligent selection method of video thumbnails as claimed in claim 3 is characterized in that: import thumbnail in the described step (29), establish the pixel value (r of colored RGB image _i, g _i, b _i), i=1 ..., W * H, W wherein, H is respectively the wide and high of thumbnail, and base_score is the benchmark mark, and the computing formula of then calculating the discrete distance value score_rgb_dist between the colored RGB piece is:

$\mathrm{score}\_\mathrm{rgb}\_\mathrm{dist}=-\mathrm{base}\_\mathrm{score}+\underset{i=1}{\overset{W\×H}{\mathrm{\Σ}}}\underset{j=1,j\≠i}{\overset{W\×H}{\mathrm{\Σ}}}\left(\right|r_i-r_j|+|g_i-g_j|+|b_i-b_j\left|\right)/(3\×W\×H\×255),$ Wherein, base_score=0.0.

13. intelligent selection method of video thumbnails as claimed in claim 2 is characterized in that: score value final_score final in the described step (4) is calculated by following formula:

final_score＝w ₁×score_face_frontal+w ₂×score_face_profile+w ₃×score_full_body+w ₄×score_upper_body+w ₅×score_lower_body+w ₆×score_hist_distrib+w ₇×score_gray_hist_std+w ₈×score_RGB_hist_std+w ₉×score_rgb2gray_diff+w ₁₀×score_gray_std+w ₁₁×score_RGB_std；

Wherein, w ₁=1.0, w ₂=0.5, w ₃=0.5, w ₄=0.5, w ₅=0.5, w ₆=0.5, w ₇=0.5, w ₈=1.2, W ₉=1.2, w ₁₀=0.5, w ₁₁=1.2.

Images