CN106294690A - Image/video search platform based on content - Google Patents

Image/video search platform based on content Download PDF

Info

Publication number
CN106294690A
CN106294690A CN201610639960.1A CN201610639960A CN106294690A CN 106294690 A CN106294690 A CN 106294690A CN 201610639960 A CN201610639960 A CN 201610639960A CN 106294690 A CN106294690 A CN 106294690A
Authority
CN
China
Prior art keywords
image
video
search
value
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201610639960.1A
Other languages
Chinese (zh)
Inventor
刘靖峰
万伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
CLOUDS (GUANGDONG) CLOUD COMPUTING TECHNOLOGY Co Ltd
Original Assignee
CLOUDS (GUANGDONG) CLOUD COMPUTING TECHNOLOGY Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by CLOUDS (GUANGDONG) CLOUD COMPUTING TECHNOLOGY Co Ltd filed Critical CLOUDS (GUANGDONG) CLOUD COMPUTING TECHNOLOGY Co Ltd
Priority to CN201610639960.1A priority Critical patent/CN106294690A/en
Publication of CN106294690A publication Critical patent/CN106294690A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/71—Indexing; Data structures therefor; Storage structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/783—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Library & Information Science (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

The present invention provides a kind of image/video search platform based on content, including: video storage server, for storing the video of recording；Feature extraction server, is used for: 1) to the video extraction key frame of video recorded；2) by the feature extraction algorithm preset, key frame of video is carried out image characteristics extraction, and the characteristics of image of extraction is sent to video index data base preservation；3) by the feature extraction algorithm preset, the search graph picture of user is carried out image characteristics extraction, and returns the characteristics of image of search graph picture to search server；Search server, for providing the network interface searching video with image；Searching in video index data base after obtaining the characteristics of image of search graph picture, and the characteristics of image of the video preserved in video index data base does and mates, if finding occurrence, video index server returns corresponding list of videos；The present invention searches for compared to video artefacts, and accuracy and search efficiency are greatly improved.

Description

Image/video search platform based on content
Technical field
The present invention relates to a kind of video search platform, a kind of image/video search platform based on content.
Background technology
The amount of the video that current various monitoring devices preserve is increasing.Massive video is difficult to management, and to regard from magnanimity Find a certain section of content the most difficult in Pin.
Therefore, current people search the specific image of needs from video, are the most also dependent on artificial cognition, the most time-consumingly consume , there is again the probability of omission in power.
Summary of the invention
It is an object of the invention to overcome the deficiencies in the prior art, it is provided that a kind of image/video based on content is searched Suo Pingtai, it is possible to feature and video features that user inputs search graph picture contrast, finds with to search for picture material similar Image place video；The technical solution used in the present invention is:
A kind of image/video search platform based on content, including:
Video storage server, for storing the video of recording；
Feature extraction server, is used for:
1) to the video extraction key frame of video recorded；
2) by the feature extraction algorithm preset, key frame of video is carried out image characteristics extraction, and the image that will extract Feature is sent to video index data base and preserves；
3) by the feature extraction algorithm preset, the search graph picture of user is carried out image characteristics extraction, and to search clothes Business device returns the characteristics of image of search graph picture；
Search server, for providing the network interface searching video with image；And the search graph picture of user's input is sent Image characteristics extraction is carried out to feature extraction server；Look in video index data base after obtaining the characteristics of image of search graph picture Look for, and the characteristics of image of the video preserved in video index data base does and mates, if finding occurrence, video index service Device returns corresponding list of videos；
Video index data base, for preserving the characteristics of image of video；And return Search Results to search server, will Return closest to the image place list of videos of characteristics of image with the characteristics of image of search graph picture.
Further, the described video extraction key frame of video to recording, specifically include:
In H264 and H265 video, frame of video is divided into I, P, B frame, and I frame is fully retained；Add up the average of P frame sign Value, if the size of P frame is more than the 120% of meansigma methods, retains current P frame and otherwise abandons current P frame；B frame is abandoned completely.
The invention provides four kinds of default feature extraction algorithms, be color layout feature extraction algorithm respectively, profile is special Levy extraction algorithm, outline color compound characteristics extraction algorithm, gradient orientation histogram feature extraction algorithm；Due to different algorithms Adapt to different scenes, so when a kind of algorithm effect is bad when, user can use other method to scan for, carry The high accuracy of search platform；
It is an advantage of the current invention that:
1) video content features is utilized to be indexed, owing to index data amount only has the one thousandth size of original video, It is possible to carry out video search efficiently；
2) monitoring device and search platform are docked so that user can directly search monitor video.
3) compared to manual search, accuracy and search efficiency are greatly improved.
Accompanying drawing explanation
Fig. 1 is the structure composition schematic diagram of the present invention.
Fig. 2 is the color layout feature extraction algorithm flow chart of the present invention.
Fig. 3 is that the Z of the present invention scans schematic diagram.
Fig. 4 is the top left region in the 8x8 pixel fritter of the present invention, right regions, lower left region, lower right area four The average gray of 4x4 block of pixels calculates schematic diagram.
Detailed description of the invention
Below in conjunction with concrete drawings and Examples, the invention will be further described.
As it is shown in figure 1, the present invention proposes a kind of image/video search platform based on content, including: video storage clothes Business device, feature extraction server, search server, and video index data base；
Video storage server, for storing the video of recording；
Feature extraction server, is used for:
1) to the video extraction key frame of video recorded；
2) by the feature extraction algorithm preset, key frame of video is carried out image characteristics extraction, and the image that will extract Feature is sent to video index data base and preserves；
3) by the feature extraction algorithm preset, the search graph picture of user is carried out image characteristics extraction, and to search clothes Business device returns the characteristics of image of search graph picture；
Search server, for providing the network interface searching video with image；And the search graph picture of user's input is sent Image characteristics extraction is carried out to feature extraction server；Look in video index data base after obtaining the characteristics of image of search graph picture Look for, and the characteristics of image of the video preserved in video index data base does and mates, if finding occurrence, video index service Device returns corresponding list of videos；
Video index data base, for preserving the characteristics of image of video；And return Search Results to search server, will Return closest to the image place list of videos of characteristics of image with the characteristics of image of search graph picture.
The photographic head of monitoring device is accessed this image/video search platform, by camera collection video, when there being thing In the case of body moves, or other needs video recording, photographic head is recorded and uploaded videos stores to video storage server, video Server is for storing the video of recording；
In order to reduce the feature extraction and calculation amount of feature extraction server, key frame of video extraction first can be carried out；Video Key-frame extraction is exactly that a few width images extracting key from one section of video are to represent this section of video.Due to the most most in video The image of number is similar the most identical, so the key frame that one section of video actually has only to minority can replace；The most detailed Carefully describe the Key-frame Extraction Algorithm of video:
In H264 and H265 video, frame of video is divided into I, P, B frame, I frame be key frame of video it contain present frame Complete information, P frame be video estimation frame it need I frame could represent present frame together, B frame is bi-directional predicted frames, need I with And P frame could represent present frame.
Can retain during video is decoded and I frame can be fully retained, the meansigma methods of P frame sign can be added up simultaneously, If the size of P frame is more than the 120% of meansigma methods, current P frame can be retained and otherwise abandon current P frame (the biggest district with I frame of P frame The biggest, illustrate that this frame is that the probability of key frame is the biggest.)；Can abandon completely for B frame, because the quantity of information that B frame comprises Considerably less.
After above-mentioned steps, key frame of video will be reduced to 1/the tens of frame of video, greatly reducing The quantity of key frame, is greatly improved the extraction rate of key frame；Also greatly reduce the amount of calculation of feature extraction simultaneously.
The image characteristics extraction of key frame can be carried out after extracting key frame of video, the invention provides four kinds of default spies Levying extraction algorithm, be color layout feature extraction algorithm respectively, contour feature extraction algorithm, outline color compound characteristics extracts to be calculated Method, gradient orientation histogram feature extraction algorithm；Owing to different algorithms adapts to different scenes, so when a kind of algorithm effect The when of bad, user can use other method to scan for, and improves the accuracy of search platform；
For key frame of video, or by the image that search server inputs, use any of the above-described kind of identical feature The characteristics of image that extraction algorithm extracts, it is possible to compare, thus find and the figure of search graph picture video index data base As feature is closest to the video of characteristics of image；User can take pictures, or uses any image, is sent to search server Carry out content relevant inside search video.
(1) color layout feature extraction algorithm, for extracting key frame of video and the color characteristic of user's search graph picture Value；
As in figure 2 it is shown, the key step of this algorithm includes:
Step 1.1, is divided into the block of 8x8 pixel by input picture (Y, Cb, Cr space)；Input picture herein can be to regard Frequently the search graph picture of key frame images or user；
Step 1.2, (meansigma methods here is meant that 64 pixels of block are each to average the block of each 8x8 pixel The value of color channel is averaged respectively), replace current block, then one new image of composition, new images by this meansigma methods Wide height be the little figure of artwork image width high 1/8；This little figure is zoomed to 80x40 pixel by bilinear interpolation；
Step 1.3, the Y to the image of this 80x40 pixel size, it is (discrete that Cb, Cr do 8x8 block of pixels dct transform respectively Cosine transform), quantify and Z scanning；Color feature value will be obtained；
Step 1.4, is stored in video index data base by the color feature value of image；
The method that dct transform (discrete cosine transform) is presented herein below:
Carry out calculating in units of 8x8 block of pixels by discrete cosine transform (80x40 pixel image being divided into non-overlapping copies The block of 8x8, does dct transform to every piece respectively)；The block of pixels of 8x8 is carried out dct transform, alternative approach such as following formula (1) institute line by line Show,
f m = Σ k = 0 n - 1 x k c o s [ π n m ( k + 1 2 ) ] - - - ( 1 )
Wherein n=8, xkPixel value for a line kth pixel；fmIt is the result of calculation of this row m-th pixel, the value of m Scope is 0～7；
After carrying out dct transform line by line, the block of a new 8x8 pixel can be obtained, to the block of this 8x8 pixel again Entering dct transform by column, computing formula is identical with the method calculated line by line, simply xkFor the pixel value of the kth pixel of string, fmBeing the result of calculation of this row m-th pixel, the span of m is similarly 0～7；Rank transformation can obtain one newly after completing The block of 8x8 pixel, this block is exactly the final calculation result of dct transform；
Computational methods that dct transform result quantified are presented herein below:
The result of calculation of this 8x8 pixel being quantified, following table one is the quantization table of 8x8；Respectively the calculating of DCT is tied The respective pixel of fruit, divided by the respective value of quantization table, obtains a new 8x8 result of calculation；
Table one
The method that Z scanning is presented herein below:
To the result quantified, take front 15 values in proper order according to Fig. 3 arrow indication, abandon 49 values below；
The block of each 8x8 pixel of the image (Y, Cb, Cr) of this 80x40 pixel size is carried out dct transform, amount Changing, Z scans, using all values that obtains as the color feature value of image；
In this algorithm, the method for Image Feature Matching is:
A) current algorithm is used to obtain the color feature value of the search graph picture that user inputs；
B) by the color feature value of search graph picture and the color feature value side of doing of existing video in video index data base Difference, what variance was minimum is exactly as immediate image with search graph, and returns immediate image place list of videos；
Lower formula (2) is formula of variance, DY, DCb, DCr and DY ', and DCb ', DCr ' are respectively search graph picture and data base The color feature value of storage, D is the variance calculated.(DY represents brightness value, DCb and DCr represents chromaticity Value)；ωyi、ωbi、ωriRepresent weight coefficient；
D = Σ i w y i ( DY i - DY i ′ ) 2 + Σ i w b i ( DCb i - DCb i ′ ) 2 + Σ i w r i ( DCr i - DCr i ′ ) 2 - - - ( 2 )
(2) contour feature extraction algorithm, for extracting key frame of video and the contour feature of user's search graph picture；
Step 2.1, is converted into gray-scale map by input picture (Y, Cb, Cr) image, and conversion method is to abandon Cb, and Cr becomes code insurance Stay Y composition；
Step 2.2, is divided into the fritter of the 8x8 pixel of non-overlapping copies by gray-scale map；
Step 2.3, calculates the top left region in 8x8 pixel fritter, right regions, lower left region, lower right area four respectively The average gray of individual 4x4 block of pixels, obtains the image block of a 2x2 pixel；As shown in Figure 4；
Step 2.4, calculates the marginal information of each 2x2 block of pixels respectively；Marginal information is divided into 5 kinds, respectively: 0 degree of limit Edge, 45 degree of edges, 90 degree of edges, 135 degree of edges and directionless edge；
The computational methods at 0 degree of edge are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by 0 degree of wave filter every respectively Individual value, then sues for peace；As shown in formula (3),
δ h ( i , j ) = | Σ k = 0 3 a k ( i , j ) f h ( k ) | - - - ( 3 )
Wherein akRepresent the value of each pixel of 2x2 block of pixels, fhRepresent each value of 0 degree of wave filter；0 degree of wave filter Each value, each value of 45 degree of wave filter, each value of 90 degree of wave filter, each value of 135 degree of wave filter, directionless wave filter Each value be shown in Table two:
Table two
In like manner,
The computational methods at 45 degree of edges are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by 45 degree of wave filter respectively Each value, then sues for peace；
The computational methods at 90 degree of edges are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by 90 degree of wave filter respectively Each value, then sues for peace；
The computational methods at 135 degree of edges are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by 135 degree of wave filter respectively Each value, then sue for peace；
The most extrorse computational methods are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by directionless filtering respectively Each value of device, then sues for peace；
In 5 results calculated, maximum is exactly the current direction character of 2x2 block of pixels；Such as calculate 90 degree Marginal value is maximum, then this 2x2 block of pixels is exactly 90 degree of edge block；
Step 2.5, calculates the block of image all 2x2 pixel, adds up 0 degree of edge, 45 degree of edges, 90 degree of edges, 135 degree of limits The number of times that edge and these 5 kinds of edges, directionless edge occur, forms one 5 dimension rectangular histogram as contour feature；
Step 2.6, is stored in video index data base by contour feature；
In this algorithm, the method for Image Feature Matching is:
The contour feature 5 of search graph picture is tieed up rectangular histogram and the contour feature 5 of existing video in video index data base Dimension rectangular histogram does variance, and what variance was minimum be exactly as immediate image with search graph, and returns to immediate image place and regard Frequently list；
(3) outline color compound characteristics extraction algorithm, for extracting key frame of video and the profile face of user's search graph picture Color compound characteristics；Specific algorithm is as follows:
Step 3.1, color feature extracted: the rgb value of original image is converted to HSV form；HSV (tone, saturation, Angle)；
The image making HSV form was obscured by 10-bins blur filter (10 frequency blur filter) and 24-bins Filter (24 frequency blur filter), it is thus achieved that 10-bins color characteristic histogram and 24-bins color characteristic histogram；
Step 3.2, contour feature extracts: from rgb space, original image is transformed into YIQ space, only retains Y and becomes Point, use method identical in algorithm (two), obtain contour feature 5 and tie up rectangular histogram；
Step 3.3, ties up 24-bins color characteristic histogram and contour feature 5 rectangular histogram and generates the Nogata of one 120 dimension Figure, final acquisition outline color compound characteristics i.e. CEDD eigenvalue；
In this algorithm, the method for Image Feature Matching is:
The CEDD eigenvalue of search graph picture is done variance with the CEDD eigenvalue of existing video in video index data base, What variance was minimum is exactly as immediate image with search graph, and returns immediate image place list of videos；
(4) gradient orientation histogram feature extraction algorithm, for extracting key frame of video and the image of user's search graph picture Textural characteristics；Specific algorithm is as follows:
Step 4.1, is converted into gray-scale map by input picture (Y, Cb, Cr) image, and conversion method is to abandon Cb, and Cr becomes code insurance Stay Y composition；
Step 4.2, zooms to 80x40 pixel by gray-scale map by bilinear interpolation；
Step 4.3, calculates the Grad of each pixel respectively, shown in computational methods equation below (4), (5),
Ix=f (x+1, y)-f (x-1, y) (4)
Iy=f (x, y+1)-f (x, y-1) (5)
Wherein Ix and Iy represents the Grad both horizontally and vertically gone up, and (x y) represents x, the pixel value of y-coordinate to f；
Step 4.4, calculates gradient direction and range value, and (x, y) represents the range value of gradient to M, and (x y) represents gradient to θ Direction (span 0～360)；Shown in computational methods equation below (6), (7),
M ( x , y ) = I x 2 + I y 2 - - - ( 6 )
Step 4.5, normalizes to 0 by the gradient information that 0 to 360 spend, and 20,40,60,80,100,120,140,160, 180,200,220,240,260,280,300,320,340 these 18 directions；Computational methods are as follows:
A (xy)=[(θ (x, y)+10)/18] * 18 (8)
F (a (xy))=f (a (xy))+M (x, y) (9)
The wherein rectangular histogram that result is 18 dimensions of f (a (xy)), is the image texture characteristic value extracted；
In this algorithm, the method for Image Feature Matching is:
The rectangular histogram that the image texture characteristic 18 of search graph picture is tieed up and the figure of existing video in video index data base As the rectangular histogram of textural characteristics 18 dimension does variance, what variance was minimum be exactly and search graph is as immediate image, and return connects most Near image place list of videos；
Finally, the use of image/video search platform:
For a user,
1. user selects a pictures or takes a picture；
2. by mobile phone A PP, the image such as picture or photo is uploaded to search server；
3. search server sends images to feature extraction server；
4. feature extraction server returns image feature value；
5. search in video index data base after search server takes image feature value, it is thus achieved that most like image institute List at video returns to user APP；
6. user can click on the list of videos inside APP to check associated video.

Claims (6)

1. an image/video search platform based on content, it is characterised in that including:
Video storage server, for storing the video of recording；
Feature extraction server, is used for:
1) to the video extraction key frame of video recorded；
2) by the feature extraction algorithm preset, key frame of video is carried out image characteristics extraction, and the characteristics of image that will extract It is sent to video index data base preserve；
3) by the feature extraction algorithm preset, the search graph picture of user is carried out image characteristics extraction, and to search server Return the characteristics of image of search graph picture；
Search server, for providing the network interface searching video with image；And the search graph picture that user inputs is sent to spy Levy extraction server and carry out image characteristics extraction；Search in video index data base after obtaining the characteristics of image of search graph picture, Doing with the characteristics of image of the video preserved in video index data base and mate, if finding occurrence, video index server returns Return corresponding list of videos；
Video index data base, for preserving the characteristics of image of video；And to search server return Search Results, will with search The characteristics of image of rope image returns closest to the image place list of videos of characteristics of image.
2. image/video search platform based on content as claimed in claim 1, it is characterised in that
The described video extraction key frame of video to recording, specifically includes:
In H264 and H265 video, frame of video is divided into I, P, B frame, and I frame is fully retained；The meansigma methods of statistics P frame sign, as Really the size of P frame is more than the 120% of meansigma methods, retains current P frame and otherwise abandons current P frame；B frame is abandoned completely.
3. image/video search platform based on content as claimed in claim 1 or 2, it is characterised in that
Described feature extraction algorithm includes: color layout feature extraction algorithm, is used for extracting key frame of video and user's search graph The color feature value of picture；
Step 1.1, is divided into the block of 8x8 pixel by input picture (Y, Cb, Cr space)；
Step 1.2, averages to the block of each 8x8 pixel, replaces current block by this meansigma methods, and then composition one is new Image, the wide height of new images is the little figure of artwork image width high 1/8；This little figure is zoomed to 80x40 picture by bilinear interpolation Element；
Step 1.3, the Y to the image of this 80x40 pixel size, Cb, Cr do respectively 8x8 block of pixels dct transform, quantization and Z scans；Color feature value will be obtained；
In this algorithm, the method for Image Feature Matching is:
Current algorithm is used to obtain the color feature value of the search graph picture that user inputs；
The color feature value of search graph picture is done variance, variance with the color feature value of existing video in video index data base Minimum is exactly as immediate image with search graph, and returns immediate image place list of videos.
4. image/video search platform based on content as claimed in claim 1 or 2, it is characterised in that
Described feature extraction algorithm includes: contour feature extraction algorithm, for extract key frame of video and user's search graph as Contour feature；
Step 2.1, is converted into gray-scale map by input picture (Y, Cb, Cr) image；
Step 2.2, is divided into the fritter of the 8x8 pixel of non-overlapping copies by gray-scale map；
Step 2.3, calculates the top left region in 8x8 pixel fritter, right regions, lower left region, four 4x4 of lower right area respectively The average gray of block of pixels, obtains the image block of a 2x2 pixel；
Step 2.4, calculates the marginal information of each 2x2 block of pixels respectively；Marginal information is divided into 5 kinds, respectively: 0 degree of edge, and 45 Degree edge, 90 degree of edges, 135 degree of edges and directionless edge；
The computational methods at 0 degree of edge are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by each value of 0 degree of wave filter respectively, Then sue for peace；
The computational methods at 45 degree of edges are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by 45 degree of wave filter each respectively Value, then sues for peace；
The computational methods at 90 degree of edges are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by 90 degree of wave filter each respectively Value, then sues for peace；
The computational methods at 135 degree of edges are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by 135 degree of wave filter every respectively Individual value, then sues for peace；
The most extrorse computational methods are as follows, and the value of each pixel of 2x2 block of pixels is multiplied by directionless wave filter respectively Each value, then sues for peace；
In 5 results calculated, maximum is exactly the current direction character of 2x2 block of pixels；
Step 2.5, calculates the block of image all 2x2 pixel, adds up 0 degree of edge, 45 degree of edges, 90 degree of edges, 135 degree of edges with And the number of times that these 5 kinds of edges, directionless edge occur, form one 5 dimension rectangular histogram as contour feature；
In this algorithm, the method for Image Feature Matching is:
The contour feature 5 of search graph picture is tieed up rectangular histogram tie up directly with the contour feature 5 of existing video in video index data base Side's figure does variance, and what variance was minimum is exactly as immediate image with search graph, and returns immediate image place video row Table.
5. image/video search platform based on content as claimed in claim 4, it is characterised in that
Described feature extraction algorithm includes: outline color compound characteristics extraction algorithm, is used for extracting key frame of video and user searches The outline color compound characteristics of rope image；
Step 3.1, color feature extracted: the rgb value of original image is converted to HSV form；The image making HSV form passes through 10-bins blur filter and 24-bins blur filter, it is thus achieved that 10-bins color characteristic histogram and 24-bins color Feature histogram；
Step 3.2, contour feature extracts: from rgb space, original image is transformed into YIQ space, only retains Y composition, adopt By method identical in claim 4, obtain contour feature 5 and tie up rectangular histogram；
Step 3.3, ties up 24-bins color characteristic histogram and contour feature 5 rectangular histogram and generates the rectangular histogram of one 120 dimension, Final acquisition outline color compound characteristics i.e. CEDD eigenvalue；
In this algorithm, the method for Image Feature Matching is:
The CEDD eigenvalue of search graph picture is done variance, variance with the CEDD eigenvalue of existing video in video index data base Minimum is exactly as immediate image with search graph, and returns immediate image place list of videos.
6. image/video search platform based on content as claimed in claim 1 or 2, it is characterised in that
Described feature extraction algorithm includes: gradient orientation histogram feature extraction algorithm, is used for extracting key frame of video and user The image texture characteristic of search graph picture；Specific algorithm is as follows:
Step 4.1, is converted into gray-scale map by input picture (Y, Cb, Cr) image, and conversion method is to abandon Cb, and Cr composition retains Y Composition；
Step 4.2, zooms to 80x40 pixel by gray-scale map by bilinear interpolation；
Step 4.3, calculates the Grad of each pixel respectively, shown in computational methods equation below (4), (5),
Ix=f (x+1, y)-f (x-1, y) (4)
Iy=f (x, y+1)-f (x, y-1) (5)
Wherein Ix and Iy represents the Grad both horizontally and vertically gone up, and (x y) represents x, the pixel value of y-coordinate to f；
Step 4.4, calculates gradient direction and range value, and (x, y) represents the range value of gradient to M, and (x y) represents the direction of gradient to θ (span 0～360)；Shown in computational methods equation below (6), (7),
M ( x , y ) = I x 2 + I y 2 - - - ( 6 )
Step 4.5, normalizes to 0 by the gradient information that 0 to 360 spend, and 20,40,60,80,100,120,140,160,180, 200,220,240,260,280,300,320,340 these 18 directions；Computational methods are as follows:
A (xy)=[(θ (x, y)+10)/18] * 18 (8)
F (a (xy))=f (a (xy))+M (x, y) (9)
The wherein rectangular histogram that result is 18 dimensions of f (a (xy)), is the image texture characteristic value extracted；
In this algorithm, the method for Image Feature Matching is:
The rectangular histogram that the image texture characteristic 18 of search graph picture is tieed up and the image stricture of vagina of existing video in video index data base The rectangular histogram of reason feature 18 dimension does variance, and what variance was minimum is exactly as immediate image with search graph, and returns immediate Image place list of videos.
CN201610639960.1A 2016-08-05 2016-08-05 Image/video search platform based on content Pending CN106294690A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610639960.1A CN106294690A (en) 2016-08-05 2016-08-05 Image/video search platform based on content

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610639960.1A CN106294690A (en) 2016-08-05 2016-08-05 Image/video search platform based on content

Publications (1)

Publication Number Publication Date
CN106294690A true CN106294690A (en) 2017-01-04

Family

ID=57665986

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610639960.1A Pending CN106294690A (en) 2016-08-05 2016-08-05 Image/video search platform based on content

Country Status (1)

Country Link
CN (1) CN106294690A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113014953A (en) * 2019-12-20 2021-06-22 山东云缦智能科技有限公司 Video tamper-proof detection method and video tamper-proof detection system
CN114741553A (en) * 2022-03-31 2022-07-12 慧之安信息技术股份有限公司 Image feature-based picture searching method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102254006A (en) * 2011-07-15 2011-11-23 上海交通大学 Method for retrieving Internet video based on contents
CN103500230A (en) * 2013-10-25 2014-01-08 苏州千视通信科技有限公司 Video retrieval method based on sample image
CN103942337A (en) * 2014-05-08 2014-07-23 北京航空航天大学 Video search system based on image recognition and matching
CN104216956A (en) * 2014-08-20 2014-12-17 北京奇艺世纪科技有限公司 Method and device for searching picture information

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102254006A (en) * 2011-07-15 2011-11-23 上海交通大学 Method for retrieving Internet video based on contents
CN103500230A (en) * 2013-10-25 2014-01-08 苏州千视通信科技有限公司 Video retrieval method based on sample image
CN103942337A (en) * 2014-05-08 2014-07-23 北京航空航天大学 Video search system based on image recognition and matching
CN104216956A (en) * 2014-08-20 2014-12-17 北京奇艺世纪科技有限公司 Method and device for searching picture information

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
李博 等: "一种新的基于梯度方向直方图的图像配准方法", 《计算机应用研究》 *
谢莉 等: "基于DCT系数和颜色直方图的图像检索", 《邵阳学院学报（自然科学版）》 *
邹煜: "基于轮廓提取和颜色直方图的图像检索", 《万方数据知识服务平台》 *
顾红飞 等: "基于颜色布局的压缩图像检索研究", 《合肥工业大学学报（自然科学版）》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113014953A (en) * 2019-12-20 2021-06-22 山东云缦智能科技有限公司 Video tamper-proof detection method and video tamper-proof detection system
CN114741553A (en) * 2022-03-31 2022-07-12 慧之安信息技术股份有限公司 Image feature-based picture searching method

Similar Documents

Publication Publication Date Title
Vu et al. ViS 3: An algorithm for video quality assessment via analysis of spatial and spatiotemporal slices
Vu et al. A spatiotemporal most-apparent-distortion model for video quality assessment
Fang et al. Objective quality assessment of screen content images by uncertainty weighting
Fang et al. Saliency detection in the compressed domain for adaptive image retargeting
Kim et al. Image contrast enhancement using entropy scaling in wavelet domain
Zhu et al. Image interpolation based on non-local geometric similarities and directional gradients
TWI475495B (en) Image identification method, electronic device, and computer program product
CN105049851B (en) General non-reference picture quality appraisement method based on Color perception
Malik et al. The statistical quantized histogram texture features analysis for image retrieval based on median and laplacian filters in the dct domain
CN111489346B (en) Full-reference image quality evaluation method and system
CN108830823B (en) Full-reference image quality evaluation method based on spatial domain combined frequency domain analysis
CN105469353B (en) The embedding grammar and device and extracting method and device of watermarking images
CN104036468A (en) Super-resolution reconstruction method for single-frame images on basis of pre-amplification non-negative neighbor embedding
Ma et al. Efficient saliency analysis based on wavelet transform and entropy theory
Lumb et al. Texture Feature Extraction of RGB, HSV, YIQ and Dithered Images using Wavelet and DCT Decomposition Techniques
CN114627154B (en) Target tracking method deployed in frequency domain, electronic equipment and storage medium
Fang et al. Saliency-based image retargeting in the compressed domain
CN106294690A (en) Image/video search platform based on content
CN108717690A (en) A kind of synthetic method of high dynamic range photo
Tang et al. Blind quality index for camera images with natural scene statistics and patch-based sharpness assessment
Au et al. Digital photo album compression based on global motion compensation and intra/inter prediction
CN110321452A (en) A kind of image search method based on direction selection mechanism
TW200910261A (en) Image processing methods and image processing apparatuses utilizing the same
Zengzhen Image quality assessment in multiband DCT domain based on SSIM
US9619864B2 (en) Image processing apparatus and method for increasing sharpness of images

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20170104

RJ01 Rejection of invention patent application after publication