CN110738128A - repeated video detection method based on deep learning - Google Patents
repeated video detection method based on deep learning Download PDFInfo
- Publication number
- CN110738128A CN110738128A CN201910888907.9A CN201910888907A CN110738128A CN 110738128 A CN110738128 A CN 110738128A CN 201910888907 A CN201910888907 A CN 201910888907A CN 110738128 A CN110738128 A CN 110738128A
- Authority
- CN
- China
- Prior art keywords
- video
- feature
- layer
- features
- library
- 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
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 17
- 238000013135 deep learning Methods 0.000 title claims abstract description 16
- 238000013528 artificial neural network Methods 0.000 claims abstract description 20
- 230000003252 repetitive effect Effects 0.000 claims abstract description 6
- 238000005259 measurement Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 32
- 238000000034 method Methods 0.000 claims description 13
- 238000010586 diagram Methods 0.000 claims description 8
- 238000013527 convolutional neural network Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 5
- 239000010410 layer Substances 0.000 description 37
- 238000004364 calculation method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000011229 interlayer Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses repetitive video detection methods based on deep learning, which comprises the steps of extracting features of an existing video by using a neural network, establishing a video feature library, extracting the features of the video to be detected, calculating Euclidean distances between the features of the video and the features in the library as similarity measurement, and marking as a repetitive video when the distances are smaller than a set threshold value.
Description
Technical Field
The invention belongs to the technical field of computer vision, digital image processing and deep learning, and particularly relates to repeated video detection methods based on a deep learning technology.
Background
With the advent of the internet era, video production and distribution are more and more convenient, video data is growing in a large scale, for example general use of short video applications, video shooting is gradually becoming a way for many people to share life content, but a large amount of repeated videos are inevitably generated at the same time, video content has an economic value of , many actions such as making profit by using pirated videos exist today, videos which are protected by copyright are modified and uploaded to video websites which are not authorized by video producers, copyright problems are generated, interests of video producers are damaged, the video websites face legal risks of , the repeated videos also increase bandwidth and storage cost of the video websites.
The common repeated video mainly comprises format conversion of the video, addition of subtitles and watermarks in the video, video compression, rotation, clipping and the like, the traditional video file hash detection can generate the same hash value for the video files with the same content, and whether the video files belong to the same video is judged by comparing the hash value with the hash value of , but the method can only detect the video with complete content, the changed video files have larger hash value and cannot be used for repeated video detection, so that the digital image processing technology is required to automatically detect the video content and judge the similarity of the video[1]，SURF[2]Etc., but this approach is less robust.
The use of convolutional neural networks in general for computer vision tasks in recent years by has facilitated an increase in the accuracy of the series of vision tasks, visual features extracted by convolutional neural networks are generally more robust, but there are fewer ways to solve using neural networks for repeated video detection.
[ reference documents ]
[1]Liu H,Lu H,Xue X.A Segmentation and Graph-Based Video SequenceMatching Method for Video Copy Detection[J].IEEE Transactions on Knowledgeand Data Engineering,2013,25(8):1706-1718.
[2]Yang G，Chen N，Jiang Q.A robust hashing algorithm based on SURF forvideo copy detection[J].Computers&Security，2012，31(1)：33-39.
[3]Simonyan K，Zisserman A.Very Deep Convolutional Networks for Large-Scale ImageRecognition[J].Computer Science，2014.
Disclosure of Invention
Firstly, carrying out feature extraction on video frames by using a neural network, using the intermediate layer output of the neural network as the feature representation of an image, then fusing the features of all the video frames as the feature representation of the video, and finally measuring the similarity of different videos by using the distance between the video features.
In order to solve the technical problem, the repetitive video detection methods based on deep learning provided by the invention use a neural network to extract features from an existing video, establish a video feature library, extract features from a video to be detected, calculate the Euclidean distance between the video features and the features in the library as similarity measurement, and mark the video as a repetitive video when the distance is smaller than a set threshold.
The repeated video detection method comprises the following steps:
step 1: acquiring video frames from an existing video set to obtain a set of all the video frames;
step 2: extracting characteristics of the video frame by adopting a convolutional neural network intermediate layer; the convolutional neural network is a network structure of vgg 16;
firstly, for videos, acquiring a video frame set S, wherein each frame in the set is scaled to be a 3-channel image with the size of 224 multiplied by 224 and used as an input of a neural network, wherein the neural network intermediate layer output is used as a video characteristic, and a feature map of conv2_1, conv2_2, conv3_1, conv3_2, conv3_3, conv4_1, conv4_2, conv4_3, conv5_1, conv5_2 and conv5_3 layers is taken for a network structure of vgg16, wherein 11 layers are all convolution layer outputs, the convolution kernel size of the layers is 3 multiplied by 3, all 0 is filled, and the convolution kernels move by 1 pixel at a time when sliding;
combining the feature output of the middle layer to finally obtain unique 4096 feature vectors for each video;
and step 3: extracting characteristics from the video library V to obtain a video characteristic library Fv;
and 4, step 4: in the retrieval stage, extracting features of a video v to be retrieved;
and 5: comparing the characteristics of the video characteristic library by using the characteristics of the video v to be retrieved, if the conditions are met, determining the video is a repeated video, and setting the comparison mode and the conditions as follows: during retrieval, a distance formula d between different video features is calculated, the distance between videos i and j is d, a threshold value t is set, after the distance d is obtained, when d is smaller than t, a similar video is judged, and otherwise, the similar video is not judged.
Further , in step 1, the video frames are obtained from the existing video set, and the set of all the video frames is obtained as follows:
V＝(S(1)，S(2)，…S(n))，S＝(P(1)，P(2)，…P(n)) (ii) a V, S represents a set of frames of a single video, S(n)Set of frames, P, representing the nth video(n)The nth video frame representing the video.
In step 2, the process of finally obtaining unique 4096 feature vectors for each video by combining the feature output of the middle layer is that the feature graph dimension of each layer output is as follows:
F(k)＝d(W(k)×W(k)×C(k))，k＝1，2，...，11 (1)
equation (1) represents that the dimension of the characteristic diagram output of the k-th layer is W(k)×W(k)×C(k)，W(k)×W(k)Is the dimension of the characteristic diagram of the k layer，C(k)Is the number of channels of the k-th layer feature map;
dimension of the compressed feature map:
FM(k)，＝max(F(k))，k＝1，2，...，11 (2)
equation (2) shows a characteristic diagram F for the k-th layer(k)Taking the maximum value of each channel to obtain C(k)Vector representation FM of dimensions(k)Dimension is , vector length is C(k)；
Connecting the feature representations of all layers yields a feature representation of the entire video frame: FPnThe number of output channels of each convolutional layer is respectively as follows: 128, 128, 256, 256, 256, 512, 512, 512, 512, 512; the final feature dimension is the sum of the dimensions of the respective layers:
128+128+256+256+256+512+512+512+512+512+512＝4094 (3)
i.e. for video frames P(n)The extracted vector size is 4096, and S is equal to (P) for all video frames of each video(1)，P(2)，…P(n)) N 4096-dimensional vectors, and the n vectors are averaged to obtain 4096-dimensional vectors T, which are then classified into to obtain the whole video V(n)Is characterized by F (V)(n))；
The formula return to is:
in the formula (4), μ is the mean value of the vector T, σ is the variance, Tv is the final video feature vector, and finally, for each video, there are obtained unique 4096-dimensional vector tables TvnAnd represents the characteristics of the nth video.
In step 3, extracting features from the video library V by adopting the mode of step 2 to obtain a video feature library Fv; and 4, extracting features of the video v to be retrieved by adopting the mode of the step 2.
In step 5, the calculation formula of the distance between the videos i, j is as follows:
compared with the prior art, the invention has the beneficial effects that:
in the repeated video detection method, the video characteristics to be detected are obtained without manually judging whether the video is repeated with the existing video library, and the video characteristics extracted by the deep neural network are directly compared with the existing video characteristics to judge whether the video is repeated. The method is different from the traditional method of extracting a plurality of frames from the video, and a feature library is established by selecting a frame sequence. Instead, a single profile is generated for each video by merging the intermediate layer outputs of the neural networks. The scheme provided by the embodiment of the invention utilizes a deep learning method to detect the repeated video, the video can be better represented through the features extracted by the deep neural network, and compared with the traditional image feature extraction operator, the accuracy of detecting the repeated video is higher.
Drawings
FIG. 1 is a flow chart of a method for detecting repeated video based on deep learning according to the present invention;
fig. 2 is a schematic diagram of video frame feature extraction in the present invention.
Detailed Description
The invention is further illustrated in the following description with reference to the figures and the examples, which are not intended to limit the invention in any way.
The invention provides a repeated video detection method based on deep learning, which mainly comprises the following steps: extracting features of an existing video by using a neural network, establishing a video feature library, extracting the features of the video to be detected, calculating Euclidean distance between the video features and the features in the library to serve as similarity measurement, and marking the video to be repeated when the distance is smaller than a set threshold value. The invention obtains the feature representation of low-level to high-level semantic features of the video by extracting the feature output of different levels of the neural network, and can obtain more accurate video feature representation by combining the features of different levels.
As shown in fig. 1, the duplicate video detection method includes the following steps:
step 1: acquiring video frames from an existing video set to obtain a set of all video frames: v, S represent a set of frames of a single video.
V＝(S(1)，S(2)，…S(n))，S(n)A set of frames representing the nth video.
S＝(P(1)，P(2)，…P(n))，P(n)The nth video frame representing the video.
Step 2: using convolutional neural network vgg16[3]The middle layer extracts features from the video frame, vgg16 has a network structure as shown in the following table:
layer represents the different layers of the neural network, Output Shape represents the Output vector dimension per layers, and Param represents the number of parameters per layers.
Using neural network interlayer outputs as video features, we take the feature maps of conv2_1, conv2_2, conv3_1, conv3_2, conv3_3, conv4_1, conv4_2, conv4_3, conv5_1, conv5_2, conv5_3 layers for vgg16, for a total of 11 layers, all being convolutional layer outputs, the convolutional kernel size of these layers is 3 × 3, all 0 padding, 1 pixel at a time of convolutional kernel sliding.
The feature map dimensions output by each layer are as follows:
F(k)＝d(W(k)×W(k)×C(k))，k＝1，2，...，11 (1)
the above formula represents that the dimension of the characteristic diagram output of the k layer is W(k)×W(k)×C(k)，W(k)×W(k)Is the dimension of the characteristic map of the k-th layer, C(k)Is the number of channels in the k-th layer profile.
And step 3: dimension of the compressed feature map:
FM(k)，＝max(F(k))，k＝1，2，...，11 (2)
the above formula shows the characteristic diagram F of the k layer(k)Taking the maximum value of each channel to obtain C(k)Vector representation FM of dimensions(k)Dimension is , vector length is C(k)。
Then the feature representations of all layers are concatenated, as shown in fig. 2, resulting in a feature representation of the entire video frame: FPnAs can be seen from the table in step 2, the number of output channels of each convolutional layer is: 128, 128, 256, 256, 256, 512, 512, 512, 512, 512, 512. The final feature dimension is thus the sum of the dimensions of the respective layers:
128+128+256+256+256+512+512+512+512+512+512＝4094 (3)
i.e. for video frames P(n)The extracted vector size is 4096, and S is equal to (P) for all video frames of each video(1)，P(2)，…P(n)) N 4096-dimensional vectors, and the n vectors are averaged to obtain 4096-dimensional vectors T, which are then classified into to obtain the whole video V(n)Is characterized by F (V)(n))
The formula return to is:
mu is the mean of the vector T, sigma is the variance, Tv is the final video feature vector, with a dimension of 4096 for a set of video frames, unique tables of 4096-dimensional vectors Tv are finally obtained for each videonAnd represents the characteristics of the nth video.
And step 3: and (5) extracting the characteristics of the video library V in the mode of the step 2 to obtain a video characteristic library Fv.
And 4, step 4: and in the retrieval stage, for the video v to be retrieved, extracting the features Tv in the mode of the step 2.
And 5: and comparing the video feature library features by using the Tv, if the conditions are met, the video is a repeated video, and the comparison mode and the conditions are set as follows:
at the time of retrieval, calculationDistance formula d between different video features, distance between videos i, j is d (Tv)i，Tvj) The calculation formula is:
and setting a threshold value t, judging as a repeated video when d is smaller than t after the distance d is obtained, otherwise, judging as the repeated video.
In summary, the repeated video detection method based on deep learning of the present invention can be summarized as including a feature library establishment stage and a discrimination stage.
1. The characteristic library establishing stage, extracting the characteristics of the existing videos through a neural network, outputting the characteristics as the characteristic representation of the videos through a middle layer of a combined neural network, wherein each video corresponds to characteristic files with 4096 dimensions (fig. 2 shows the characteristic extraction principle of a single video frame), and storing the characteristic files into a database to obtain the video characteristic library.
2. And a judging stage, namely extracting features of the judged video by using the same method, calculating Euclidean distance of the existing video features in the library, judging the video to be repeated if the distance is less than a preset threshold value t (the threshold value is 0.3 as set by ), otherwise not judging the video to be repeated.
The network described above uses the tensoflow deep learning framework to acquire a sequence of video frames via opencv and scale the sequence of video frames to a size that satisfies the neural network input. In this example, the feature representation of the entire video is obtained by merging vgg16 the feature maps of the middle layers.
While the present invention has been described with reference to the accompanying drawings, the present invention is not limited to the above-described embodiments, which are illustrative only and not restrictive, and various modifications which do not depart from the spirit of the present invention and which are intended to be covered by the claims of the present invention may be made by those skilled in the art.
Claims (6)
- A repetitive video detection method based on deep learning, which is characterized in that a neural network is used for extracting features of an existing video, a video feature library is established, then the features of the video to be detected are extracted, the Euclidean distance between the video features and the features in the library is calculated to be used as similarity measurement, and when the distance is smaller than a set threshold value, the video is marked as a repetitive video.
- 2. The method for detecting repeated video based on deep learning of claim 1, comprising the following steps:step 1: acquiring video frames from an existing video set to obtain a set of all the video frames;step 2: extracting characteristics of the video frame by adopting a convolutional neural network intermediate layer; the convolutional neural network is a network structure of vgg 16;firstly, for videos, acquiring a video frame set S, wherein each frame in the set is scaled to be a 3-channel image with the size of 224 multiplied by 224 and used as an input of a neural network, wherein the neural network intermediate layer output is used as a video characteristic, and a feature map of conv2_1, conv2_2, conv3_1, conv3_2, conv3_3, conv4_1, conv4_2, conv4_3, conv5_1, conv5_2 and conv5_3 layers is taken for a network structure of vgg16, wherein 11 layers are all convolution layer outputs, the convolution kernel size of the layers is 3 multiplied by 3, all 0 is filled, and the convolution kernels move by 1 pixel at a time when sliding;combining the feature output of the middle layer to finally obtain unique 4096 feature vectors for each video;and step 3: extracting characteristics from the video library V to obtain a video characteristic library Fv;and 4, step 4: in the retrieval stage, extracting features of a video v to be retrieved;and 5: comparing the characteristics of the video characteristic library by using the characteristics of the video v to be retrieved, if the conditions are met, determining the video is a repeated video, and setting the comparison mode and the conditions as follows: during retrieval, a distance formula d between different video features is calculated, the distance between videos i and j is d, a threshold value t is set, after the distance d is obtained, when d is smaller than t, a similar video is judged, and otherwise, the similar video is not judged.
- 3. The method according to claim 2, wherein the video frames are obtained from an existing video set, and the set of all video frames is obtained as follows:V＝(S(1),S(2),…S(n))，S＝(P(1),P(2),…P(n))；v, S represents a set of frames of a single video, S(n)Set of frames, P, representing the nth video(n)The nth video frame representing the video.
- 4. The method for detecting repeated videos based on deep learning of claim 2, wherein in step 2, the process of combining the feature output of the middle layer to finally obtain unique 4096 feature vectors for each video is:the feature map dimensions output by each layer are as follows:F(k)＝d(W(k)×W(k)×C(k))，k＝1,2,...,11 (1)equation (1) represents that the dimension of the characteristic diagram output of the k-th layer is W(k)×W(k)×C(k)，W(k)×W(k)Is the dimension of the characteristic map of the k-th layer, C(k)Is the number of channels of the k-th layer feature map;dimension of the compressed feature map:FM(k),＝max(F(k))，k＝1,2,...,11 (2)equation (2) shows a characteristic diagram F for the k-th layer(k)Taking the maximum value of each channel to obtain C(k)Vector representation FM of dimensions(k)Dimension is , vector length is C(k)；Connecting the feature representations of all layers yields a feature representation of the entire video frame: FPnThe number of output channels of each convolutional layer is respectively as follows: 128, 128, 256, 256, 256, 512, 512, 512, 512, 512; final characteristicsThe dimension is the sum of the dimensions of the respective layers:128+128+256+256+256+512+512+512+512+512+512＝4094 (3)i.e. for video frames P(n)The extracted vector size is 4096, and S is equal to (P) for all video frames of each video(1),P(2),…P(n)) N 4096-dimensional vectors, and the n vectors are averaged to obtain 4096-dimensional vectors T, which are then classified into to obtain the whole video V(n)Is characterized by F (V)(n))；The formula return to is:in the formula (4), μ is the mean value of the vector T, σ is the variance, Tv is the final video feature vector, and finally, for each video, there are obtained unique 4096-dimensional vector tables TvnAnd represents the characteristics of the nth video.
- 5. The repeated video detection method based on deep learning of claim 4, wherein in step 3, the video library V is subjected to feature extraction in the manner of step 2 to obtain a video feature library Fv; and 4, extracting features of the video v to be retrieved by adopting the mode of the step 2.
- 6. The method for detecting repeated video based on deep learning of claim 5, wherein in step 5, the distance between the videos i, j is calculated by the following formula:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910888907.9A CN110738128A (en) | 2019-09-19 | 2019-09-19 | repeated video detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910888907.9A CN110738128A (en) | 2019-09-19 | 2019-09-19 | repeated video detection method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110738128A true CN110738128A (en) | 2020-01-31 |
Family
ID=69268280
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910888907.9A Pending CN110738128A (en) | 2019-09-19 | 2019-09-19 | repeated video detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110738128A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111506773A (en) * | 2020-03-24 | 2020-08-07 | 中国科学院大学 | Video duplicate removal method based on unsupervised depth twin network |
| CN111723692A (en) * | 2020-06-03 | 2020-09-29 | 西安交通大学 | Near-repetitive video detection method based on label features of convolutional neural network semantic classification |
| CN112399236A (en) * | 2020-10-09 | 2021-02-23 | 北京达佳互联信息技术有限公司 | Video duplicate checking method and device and electronic equipment |
| CN112528856A (en) * | 2020-12-10 | 2021-03-19 | 天津大学 | Repeated video detection method based on characteristic frame |
| WO2022086767A1 (en) * | 2020-10-22 | 2022-04-28 | Micron Technology, Inc. | Accelerated video processing for feature recognition via an artificial neural network configured in a data storage device |
| US11599856B1 (en) | 2022-01-24 | 2023-03-07 | My Job Matcher, Inc. | Apparatuses and methods for parsing and comparing video resume duplications |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105701480A (en) * | 2016-02-26 | 2016-06-22 | 江苏科海智能系统有限公司 | Video semantic analysis method |
| CN105913456A (en) * | 2016-04-12 | 2016-08-31 | 西安电子科技大学 | Video significance detecting method based on area segmentation |
| CN106446015A (en) * | 2016-08-29 | 2017-02-22 | 北京工业大学 | Video content access prediction and recommendation method based on user behavior preference |
| CN106991373A (en) * | 2017-03-02 | 2017-07-28 | 中国人民解放军国防科学技术大学 | A kind of copy video detecting method based on deep learning and graph theory |
| CN108228915A (en) * | 2018-03-29 | 2018-06-29 | 华南理工大学 | A kind of video retrieval method based on deep learning |
| CN108764019A (en) * | 2018-04-03 | 2018-11-06 | 天津大学 | A kind of Video Events detection method based on multi-source deep learning |
| CN108848422A (en) * | 2018-04-19 | 2018-11-20 | 清华大学 | A kind of video abstraction generating method based on target detection |
| CN109341703A (en) * | 2018-09-18 | 2019-02-15 | 北京航空航天大学 | A kind of complete period uses the vision SLAM algorithm of CNNs feature detection |
| CN109766823A (en) * | 2019-01-07 | 2019-05-17 | 浙江大学 | A kind of high-definition remote sensing ship detecting method based on deep layer convolutional neural networks |
| CN109815364A (en) * | 2019-01-18 | 2019-05-28 | 上海极链网络科技有限公司 | A kind of massive video feature extraction, storage and search method and system |
-
2019
- 2019-09-19 CN CN201910888907.9A patent/CN110738128A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105701480A (en) * | 2016-02-26 | 2016-06-22 | 江苏科海智能系统有限公司 | Video semantic analysis method |
| CN105913456A (en) * | 2016-04-12 | 2016-08-31 | 西安电子科技大学 | Video significance detecting method based on area segmentation |
| CN106446015A (en) * | 2016-08-29 | 2017-02-22 | 北京工业大学 | Video content access prediction and recommendation method based on user behavior preference |
| CN106991373A (en) * | 2017-03-02 | 2017-07-28 | 中国人民解放军国防科学技术大学 | A kind of copy video detecting method based on deep learning and graph theory |
| CN108228915A (en) * | 2018-03-29 | 2018-06-29 | 华南理工大学 | A kind of video retrieval method based on deep learning |
| CN108764019A (en) * | 2018-04-03 | 2018-11-06 | 天津大学 | A kind of Video Events detection method based on multi-source deep learning |
| CN108848422A (en) * | 2018-04-19 | 2018-11-20 | 清华大学 | A kind of video abstraction generating method based on target detection |
| CN109341703A (en) * | 2018-09-18 | 2019-02-15 | 北京航空航天大学 | A kind of complete period uses the vision SLAM algorithm of CNNs feature detection |
| CN109766823A (en) * | 2019-01-07 | 2019-05-17 | 浙江大学 | A kind of high-definition remote sensing ship detecting method based on deep layer convolutional neural networks |
| CN109815364A (en) * | 2019-01-18 | 2019-05-28 | 上海极链网络科技有限公司 | A kind of massive video feature extraction, storage and search method and system |
Non-Patent Citations (2)
| Title |
|---|
| 汪冬冬: "基于深度模型的慢变视觉特征学习算法研究" * |
| 赵义萱: "基于短视频的近似重复视频检测" * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111506773A (en) * | 2020-03-24 | 2020-08-07 | 中国科学院大学 | Video duplicate removal method based on unsupervised depth twin network |
| CN111723692A (en) * | 2020-06-03 | 2020-09-29 | 西安交通大学 | Near-repetitive video detection method based on label features of convolutional neural network semantic classification |
| CN111723692B (en) * | 2020-06-03 | 2022-08-09 | 西安交通大学 | Near-repetitive video detection method based on label features of convolutional neural network semantic classification |
| CN112399236A (en) * | 2020-10-09 | 2021-02-23 | 北京达佳互联信息技术有限公司 | Video duplicate checking method and device and electronic equipment |
| WO2022086767A1 (en) * | 2020-10-22 | 2022-04-28 | Micron Technology, Inc. | Accelerated video processing for feature recognition via an artificial neural network configured in a data storage device |
| US11741710B2 (en) | 2020-10-22 | 2023-08-29 | Micron Technology, Inc. | Accelerated video processing for feature recognition via an artificial neural network configured in a data storage device |
| CN112528856A (en) * | 2020-12-10 | 2021-03-19 | 天津大学 | Repeated video detection method based on characteristic frame |
| CN112528856B (en) * | 2020-12-10 | 2022-04-15 | 天津大学 | Repeated video detection method based on characteristic frame |
| US11599856B1 (en) | 2022-01-24 | 2023-03-07 | My Job Matcher, Inc. | Apparatuses and methods for parsing and comparing video resume duplications |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110738128A (en) | repeated video detection method based on deep learning | |
| US8587668B2 (en) | Method and apparatus for detecting near duplicate videos using perceptual video signatures | |
| WO2021129435A1 (en) | Method for training video definition evaluation model, video recommendation method, and related device | |
| Lu et al. | Selection of image features for steganalysis based on the Fisher criterion | |
| Li et al. | Steganalysis over large-scale social networks with high-order joint features and clustering ensembles | |
| US8203554B2 (en) | Method and apparatus for identifying visual content foregrounds | |
| CN111325271B (en) | Image classification method and device | |
| Li et al. | Sparse representation-based image quality index with adaptive sub-dictionaries | |
| US20090263014A1 (en) | Content fingerprinting for video and/or image | |
| CN104661037B (en) | The detection method and system that compression image quantization table is distorted | |
| CN110457996B (en) | Video moving object tampering evidence obtaining method based on VGG-11 convolutional neural network | |
| Abdulrahman et al. | Color image stegananalysis using correlations between RGB channels | |
| Xie et al. | Bag-of-words feature representation for blind image quality assessment with local quantized pattern | |
| CN106503112B (en) | Video retrieval method and device | |
| Kang et al. | Color Image Steganalysis Based on Residuals of Channel Differences. | |
| CN111723692B (en) | Near-repetitive video detection method based on label features of convolutional neural network semantic classification | |
| CN107545570A (en) | A kind of reconstructed image quality evaluation method of half reference chart | |
| CN102547477B (en) | Video fingerprint method based on contourlet transformation model | |
| Li et al. | Quality evaluation for image retargeting with instance semantics | |
| CN110188625B (en) | Video fine structuring method based on multi-feature fusion | |
| Yang et al. | No‐reference image quality assessment via structural information fluctuation | |
| Nie et al. | Robust video hashing based on representative-dispersive frames | |
| CN110705499B (en) | Crowd counting method based on transfer learning | |
| CN102881008A (en) | Circular loop statistic characteristic-based anti-rotation image Hash method | |
| CN114005069A (en) | Video feature extraction and retrieval method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200131 |
|
| WD01 | Invention patent application deemed withdrawn after publication |