CN108921130B - Video key frame extraction method based on saliency region - Google Patents

Video key frame extraction method based on saliency region Download PDF

Info

Publication number
CN108921130B
CN108921130B CN201810836824.0A CN201810836824A CN108921130B CN 108921130 B CN108921130 B CN 108921130B CN 201810836824 A CN201810836824 A CN 201810836824A CN 108921130 B CN108921130 B CN 108921130B
Authority
CN
China
Prior art keywords
similarity
frame
video
image
frame image
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.)
Active
Application number
CN201810836824.0A
Other languages
Chinese (zh)
Other versions
CN108921130A (en
Inventor
冯德瀛
张来刚
赵颖
楚晓华
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.)
Liaocheng University
Original Assignee
Liaocheng University
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 Liaocheng University filed Critical Liaocheng University
Priority to CN201810836824.0A priority Critical patent/CN108921130B/en
Publication of CN108921130A publication Critical patent/CN108921130A/en
Application granted granted Critical
Publication of CN108921130B publication Critical patent/CN108921130B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a video key frame extraction method based on a significance region in the technical field of computer image processing and pattern recognition, which comprises the steps of firstly carrying out sampling conversion on video data, converting a video into a continuous frame image sequence, then extracting the significance region in each frame image by using a spectrum residual model, then sequencing the significance region according to the area of the significance region in the image, carrying out color vectorization on the significance region sequenced at the front, and finally carrying out similarity measurement between the front frame image and the rear frame image according to the vectorized significance region and determining a video key frame according to the size of the similarity. The key frame image sequence extracted by the invention can effectively reserve the main content of the video.

Description

Video key frame extraction method based on saliency region
Technical Field
The invention mainly relates to the technical field of computer image processing and pattern recognition, in particular to a video key frame extraction method based on a saliency region.
Background
With the large-scale popularization of surveillance cameras in real life, surveillance videos are widely applied, and the data volume of the surveillance videos also shows exponential growth, so that huge challenges are provided for storage, organization and query of the videos. In the field of security monitoring, how to effectively organize, manage and query massive videos becomes a hot problem in current attention and research.
Video is made up of successive frames, one image each representing a segment or content of the video. Due to the temporal and spatial continuity of adjacent frames, there is a large amount of redundant information in the sequence of frame images, which is not conducive to efficient classification and retrieval in video. In response to the above problem, the video key frame extraction technology extracts a plurality of key frames from a video, and the key frames represent main content contained in the video. Conventional video key frame extraction methods are classified into a shot boundary-based method, an image content-based method, a motion analysis-based method, a video clustering-based method, a compressed domain-based method, and the like.
Through the literature search of the prior art, Xiaodi Hou et al adopts a Fourier transform method to obtain frequency domain information of an image in the literature 'Saliency Detection: A Spectral Residual Approach', and then utilizes a spectrum Residual method to detect a salient region in the image in a time domain, but the method is not used for extracting a video key frame. In the patent of 'a video key frame extraction method' (application number: CN201711165320.2, published: 2018, 3, month and 27), Royuan et al detect a moving target by using a Vibe algorithm and an inter-frame difference method, then determine global similarity by using a global feature peak signal-to-noise ratio, then judge local similarity by using SURF features, and finally synthesize the two similarities to obtain a target key frame sequence. The patent mainly extracts a key frame sequence from the perspective of an interframe difference method, and does not relate to the distinguishing of foreground objects and background noise. Similarly, Qian et al in the patent "a video key frame extraction algorithm" (application No.: CN201711047162.0, published: 2018, 3, 23) used an interframe difference method to extract key frames. The method comprises the steps of firstly calculating the size of an effective area of a certain frame of image, then detecting the characteristic information of the area, comparing the characteristic information with previous and next frames, and finally extracting a key frame by calculating the similarity between frames. Although the patent detects the effective image area, only the progressive scanning method is adopted, and the potential foreground object and the interfering background noise are not distinguished.
Disclosure of Invention
The invention provides a video key frame extraction method based on a significance region, which aims at the defects in the prior art, detects the significance region in each frame of image by utilizing a spectrum residual model in a video frame image sequence, determines a potential foreground target, avoids the influence of irrelevant background noise, filters video frame images containing similar contents by judging the color similarity of the significance regions in the previous and next frame images, and determines key frame images containing main contents, thereby laying a foundation for content-based video retrieval.
The invention is realized by the following technical scheme, and the invention specifically comprises the following steps:
firstly, sampling and converting video data, and converting a video into a continuous frame image sequence;
then extracting a salient region in each frame of image by using a spectrum residual error model;
then, sorting the salient regions according to the areas of the salient regions in the image, and carrying out color vectorization on the salient regions which are sorted in front;
and finally, carrying out similarity measurement between the front frame image and the back frame image according to the vectorized salient region, and determining the video key frame according to the size of the similarity.
The sampling conversion of the video data is as follows: the video is composed of continuous images of a frame, the sampling frequency of the video is set according to the total frame number and the frame rate contained in the video, and the video is converted into a group of continuous frame image sequences according to the sampling frequency.
In the video data, the total number of frames is NTFrame rate of nfSampling frequency of nsThen the sampled continuous frame image sequence is
Figure BDA0001744664250000021
Wherein
Figure BDA0001744664250000022
The extraction of the salient region by using the spectrum residual error model in each frame of image refers to the following steps: fourier transformation is carried out on each frame of image in a frequency domain, spectrum residual errors are calculated, and then salient regions in the images are extracted in a time domain according to inverse Fourier transformation.
Further, the step of extracting the significant region by using the spectrum residual model includes:
1) for the ith frame image I in the frequency domainiFourier transform is carried out to obtain a transformed amplitude spectrum A (f) and a phase spectrum P (f), wherein
Figure BDA0001744664250000031
Further, a logarithmic magnitude spectrum l (f) is calculated, wherein l (f) is log (a (f));
2) setting a local mean filter h of n x nn(f) Convolving with a logarithmic magnitude spectrum L (f), and calculating a spectrum residual R (f), wherein R (f) L (f) -hn(f)*L(f)；
3) Inverse Fourier transform of the spectral residual R (f) in the time domain and smoothing with a Gaussian filter g (x) to obtain an image IiCorresponding region of significance SiWherein
Figure BDA0001744664250000032
The sorting according to the areas of the saliency regions in the image and performing color vectorization on the saliency region ranked in the front refers to: each frame of image comprises a plurality of salient regions, and for a salient region with a larger area, more potential foreground objects are contained, and the influence of background noise is inhibited, so that according to the area of the salient region, the salient regions are sequenced in the image from large to small, the similarity of the front and rear frame images can be judged by utilizing the plurality of salient regions sequenced at the front, and for this reason, color vectorization is respectively carried out on R, G, B three channels for each salient region, so as to generate a corresponding color vector.
Further, the step of ordering the salient regions according to their areas in the image and performing color vectorization on the top-ordered salient regions includes:
a) in the ith frame image IiIn (1), the extracted significant region is expressed as
Figure BDA0001744664250000033
Wherein
Figure BDA0001744664250000034
Is an image IiThe r-th salient region. For significant region SiSorting according to the order of the areas from large to small, and taking out the front z displays in the front of the sortingCopy region S'iIs shown as
Figure BDA0001744664250000035
b) To significant region S'iMiddle (r) significant region
Figure BDA0001744664250000036
Histogram statistics is carried out in the interval of R, G, B three channels and 0-255 gray values respectively to generate corresponding color vectors
Figure BDA0001744664250000037
And
Figure BDA0001744664250000038
further generating the r significant region
Figure BDA0001744664250000039
Corresponding color vector
Figure BDA00017446642500000310
Accordingly, significance region S'iCorresponding color vector
Figure BDA0001744664250000041
The similarity measurement is carried out between the previous frame image and the next frame image according to the vectorized significant region, and the determination of the video key frame according to the similarity refers to: in order to determine a key frame in a frame image sequence, similarity measurement is performed between front and rear frame images through color vectors corresponding to salient regions, if the similarity is small, the difference of contents contained in the front and rear frame images is large, and the front and rear frame images can be determined as the key frame, otherwise, if the similarity is large, the difference of contents contained in the front and rear frame images is small, the front and rear frame images can be removed from the frame image sequence, and after the similarity measurement is completed in all the frame images, the remaining frame images are the key frame sequence corresponding to the video.
Further, the step of performing similarity measurement between the previous and subsequent frame images according to the vectorized salient region and determining the video key frame according to the size of the similarity includes:
to judge the ith frame image IiAnd the (I + 1) th frame image Ii+1In their corresponding color vector ViAnd Vi+1Cosine similarity measure is performed between them. Due to the vector ViFrom z component vectors
Figure BDA0001744664250000042
Formed, and then vector ViAnd Vi+1The similarity measure between the two is converted into z corresponding component vectors
Figure BDA0001744664250000043
And
Figure BDA0001744664250000044
the similarity measure between them can be expressed as
Figure BDA0001744664250000045
② in z corresponding component vectors
Figure BDA0001744664250000046
And
Figure BDA0001744664250000047
after the similarity measurement is finished, z similarity measurement values can be obtained, and if the similarity is larger, the significant regions in the front and rear frame images are illustrated
Figure BDA0001744664250000048
And
Figure BDA0001744664250000049
the contents contained are similar, if the similarity is smaller, the description is given
Figure BDA00017446642500000410
And
Figure BDA00017446642500000411
the content involved is very different. For frame image IiAnd Ii+1The overall similarity between the z similarity values, which is the minimum value of the z similarity values, can be expressed as
Figure BDA00017446642500000412
Therefore, the difference of the content of the previous and next frame images can be reflected;
calculating frame image IiAnd Ii+1After the overall similarity between the two, a similarity threshold T is set, if sim (I)i,Ii+1) If T is less than or equal to T, then I is indicatediAnd Ii+1The content contained is greatly different, IiAnd Ii+1All remain as key frame images if sim (I)i,Ii+1) If > T, then I is indicatedi+1And IiHigh similarity of contained content, IiRemains as a key frame image, and Ii+1And removing from the frame image sequence. And finally determining the key frame image sequence of the video after traversing all the frame images in sequence.
The invention has the beneficial effects that: the method utilizes the spectrum residual model to extract the salient region from the frame image sequence, detects the potential foreground target, avoids the interference of background noise and is beneficial to judging the content similarity between the previous frame image and the next frame image. By ordering and representing the salient regions in the image as independent color vectors, not only is the color information of each salient region effectively represented, but also the mutual influence among a plurality of salient regions is avoided. The similarity between the images of the previous frame and the image of the next frame is judged through the saliency region after the color vectorization, the maximum difference between the previous frame and the next frame can be reflected, and the extraction of the video key frame is facilitated. Compared with the prior art, the key frame image sequence extracted by the method can effectively reserve the main content of the video.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The invention is further described with reference to the accompanying drawings and specific embodiments. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Further, it should be understood that various changes or modifications of the present invention may be made by those skilled in the art after reading the teaching of the present invention, and these equivalents also fall within the scope of the present application.
The embodiment adopts a video key frame extraction method based on a saliency region, and the specific implementation steps are as follows:
1. and performing sampling conversion on the video data, and converting the video into a continuous frame image sequence.
In the video data, the total number of frames is NTDue to the frame rate nf24, so the sampling frequency nsThe same is set to 24, i.e. the video is sampled every 1s, and the sequence of successive frame images after sampling is
Figure BDA0001744664250000051
Wherein
Figure BDA0001744664250000052
2. And extracting a salient region in each frame of image by using a spectrum residual error model.
For a sequence of frame images
Figure BDA0001744664250000053
Middle image IiFirst, fourier transform is performed in the frequency domain to calculate an amplitude spectrum a (f), a phase spectrum p (f), and a logarithmic amplitude spectrum l (f). Then, a 3 x 3 local mean filter h is providedn(f) (n-3), convolved with the log-amplitude spectrum l (f), and the spectrum residual r (f) is calculated. Finally, Fourier inversion is carried out on the spectrum residual error R (f) in a time domain, and a Gaussian filter g (x) is adopted for smoothing, so as to extract an image IiSignificant region S in (1)i。
3. And sorting according to the areas of the salient regions in the images, and carrying out color vectorization on the salient regions which are sorted at the front.
Firstly in the image IiMiddle pair salient region
Figure BDA0001744664250000061
Arranging according to the order of the areas from large to small, and taking out the first 3 significant areas which are ranked in the front
Figure BDA0001744664250000062
Then, for the 3 saliency areas, color vectorization is performed sequentially in R, G, B three channels, each channel can generate a 256-dimensional color vector, and each saliency area corresponds to a 768-dimensional color vector. Last S'iCorresponding to 3 768-dimensional color vectors, expressed as
Figure BDA0001744664250000063
4. And performing similarity measurement between the front frame image and the back frame image according to the vectorized salient region, and determining the video key frame according to the size of the similarity.
Firstly, in the ith frame image IiAnd the (I + 1) th frame image Ii+1Corresponding color vector ViAnd Vi+1Cosine similarity measure is performed between them. ViFrom 3 component vectors
Figure BDA0001744664250000066
Is composed of, and then ViAnd Vi+1The similarity measure between them is converted into 3 corresponding component vectors
Figure BDA0001744664250000064
And
Figure BDA0001744664250000065
a measure of similarity between them. Then, the minimum value of the 3 similarity degrees obtained by calculation is taken, and the minimum value is the image IiAnd Ii+1Content similarity between them. Finally, setting a similarity threshold T equal to 0.8 if IiAnd Ii+1Similarity between them sim (I)i,Ii+1) Less than or equal to 0.8, then IiAnd Ii+1All remain as key frame images, otherwise, if IiAnd Ii+1Similarity between them sim (I)i,Ii+1) If greater than 0.8, then IiReserved as key frame image, Ii+1And removing from the frame image sequence. And traversing all the frame images in sequence to obtain a final key frame image sequence.
The simulation experiment of the method of the invention is as follows:
in the experiment, 5 monitoring cameras are selected, 4 sections of videos are recorded by each monitoring camera, 20 sections of monitoring videos are selected in total, and performance test is performed on the video key frame extraction method based on the saliency region. For the 20 segments of monitoring videos, the monitoring duration, the total video frame number, the video sampling frame number, the key frame number and the ratio of the sampling frame number to the key frame number are respectively given. Table 1 gives detailed test data for 20 segments of surveillance video. As can be seen from table 1, the number of key frames of 20 segments of video is reduced to some extent compared to the number of frames after video sampling. Because the scenes recorded by the 5 monitoring cameras are different, the ratio of the frame number of the video recorded by the 5 monitoring cameras to the key frame number after sampling has a certain difference. However, the key frame sequence extracted from the video may represent the main content of the video.
Key frame extraction performance contrast analysis of table 120 segment monitoring video
Serial number Camera head Monitoring duration Total frame number Number of frames after sampling Number of key frames Ratio of
1 Camera 1 05:55 8527 355 251 1.4
2 Camera 1 18:07 26104 1088 710 1.5
3 Camera 1 04:57 7138 297 129 2.3
4 Camera 1 19:05 27493 1146 639 1.8
5 Camera 2 00:32 771 32 17 1.9
6 Camera 2 23:21 33636 1402 1050 1.3
7 Camera 2 13:45 19801 825 574 1.4
8 Camera 2 05:50 8402 350 130 2.7
9 Camera 3 35:12 50690 2112 1045 2.0
10 Camera 3 35:18 50835 2118 811 2.6
11 Camera 3 35:16 50786 2116 1285 1.6
12 Camera 3 29:10 42019 1751 1045 1.7
13 Camera 4 23:36 33994 1416 982 1.4
14 Camera 4 09:16 13367 557 390 1.4
15 Camera 4 09:42 13971 582 460 1.3
16 Camera 4 23:29 33833 1410 958 1.5
17 Camera 5 23:29 33819 1409 786 1.8
18 Camera 5 23:29 33816 1409 842 1.7
19 Camera 5 15:01 21640 902 506 1.8
20 Camera 5 15:29 22310 930 419 2.2

Claims (3)

1. The video key frame extraction method based on the saliency region is characterized by specifically comprising the following steps of:
step one, carrying out sampling conversion on video data, and converting a video into a continuous frame image sequence;
secondly, extracting a salient region in each frame of image by using a spectrum residual error model;
thirdly, sorting the salient regions according to the areas of the salient regions in the images, and carrying out color vectorization on the salient regions which are sorted in front;
according to the area of the saliency region, the saliency regions are sequenced from large to small in the image, the similarity of the previous and next frame images can be judged by utilizing a plurality of saliency regions sequenced at the front, and color vectorization is respectively carried out on R, G, B channels aiming at each saliency region to generate a corresponding color vector, and the method specifically comprises the following steps:
1) in the ith frame image IiIn (1), the extracted significant region is expressed as
Figure FDA0003466812080000011
Wherein
Figure FDA0003466812080000012
Is an image IiMiddle r significant region, for significant region SiSorting is carried out according to the sequence of the area from large to small, and the front z significant areas S 'in the front sorting are taken out'iIs shown as
Figure FDA0003466812080000013
2) To significant region S'iMiddle (r) significant region
Figure FDA0003466812080000014
Histogram statistics is carried out in the interval of R, G, B three channels and 0-255 gray values respectively to generate corresponding color vectors
Figure FDA0003466812080000015
And
Figure FDA0003466812080000016
further generating the r significant region
Figure FDA0003466812080000017
Corresponding color vector
Figure FDA0003466812080000018
Accordingly, significance region S'iCorresponding color vector
Figure FDA0003466812080000019
Fourthly, similarity measurement is carried out between the front frame image and the back frame image according to the vectorized saliency region, and a video key frame is determined according to the size of the similarity;
similarity measurement is carried out between front and back frame images through color vectors corresponding to the salient regions, if the similarity is smaller, the difference of contents contained in the front and back frame images is larger, the front and back frame images can be determined as a key frame, otherwise, if the similarity is larger, the difference of contents contained in the front and back frame images is smaller, the front and back frame images are removed from a frame image sequence, and after the similarity measurement is completed in all the frame images, the rest frame images are the key frame sequence corresponding to the video, and the specific steps comprise:
1) to judge the ith frame image IiAnd the (I + 1) th frame image Ii+1In their corresponding color vector ViAnd Vi+1Cosine similarity measure between them, due to vector ViFrom z component vectors
Figure FDA0003466812080000021
Formed, and then vector ViAnd Vi+1The similarity measure between the two is converted into z corresponding component vectors
Figure FDA0003466812080000022
And
Figure FDA0003466812080000023
the similarity measure between them can be expressed as
Figure FDA0003466812080000024
2) At z corresponding component vectors
Figure FDA0003466812080000025
And
Figure FDA0003466812080000026
after the similarity measurement is finished, z similarity measurement values can be obtained, and if the similarity is larger, the significant areas in the front and rear frame images are indicated
Figure FDA0003466812080000027
And
Figure FDA0003466812080000028
the contained contents are similar, if the similarity is smaller, the contents are similarDescription of the invention
Figure FDA0003466812080000029
And
Figure FDA00034668120800000210
the content contained is greatly different, and the frame image IiAnd Ii+1The overall similarity between the z similarity values, which is the minimum value of the z similarity values, can be expressed as
Figure FDA00034668120800000211
Therefore, the difference of the content of the previous and next frame images can be reflected;
3) in calculating frame image IiAnd Ii+1After the overall similarity between the two, a similarity threshold T is set, if sim (I)i,Ii+1) If T is less than or equal to T, then I is indicatediAnd Ii+1The content contained is greatly different, IiAnd Ii+1All remain as key frame images if sim (I)i,Ii+1)>T, then, illustrate Ii+1And IiHigh similarity of contained content, IiRemains as a key frame image, and Ii+1And removing the key frame image sequence from the frame image sequence, and finally determining the key frame image sequence of the video after traversing all the frame images in sequence.
2. The method for extracting key frames from video based on salient regions as claimed in claim 1, wherein in step one, the video sampling frequency is set according to the total frame number and frame rate contained in the video, and the video is converted into a group of continuous frame image sequences according to the sampling frequency; in the video data, the total number of frames is NTFrame rate of nfSampling frequency of nsThen the sampled continuous frame image sequence is
Figure FDA00034668120800000212
Wherein
Figure FDA00034668120800000213
3. The method as claimed in claim 1, wherein in the second step, fourier transform is performed on each frame of image in the frequency domain, spectral residuals are calculated, and then the salient regions in the image are extracted in the time domain according to inverse fourier transform, and the specific steps include:
1) for the ith frame image I in the frequency domainiFourier transform is carried out to obtain a transformed amplitude spectrum A (f) and a phase spectrum P (f), wherein
Figure FDA0003466812080000031
Further, a logarithmic magnitude spectrum l (f) is calculated, wherein l (f) is log (a (f));
2) setting a local mean filter h of n x nn(f) Convolving with a logarithmic magnitude spectrum L (f), and calculating a spectrum residual R (f), wherein R (f) L (f) -hn(f)*L(f)；
3) Inverse Fourier transform of the spectral residual R (f) in the time domain and smoothing with a Gaussian filter g (x) to obtain an image IiCorresponding region of significance SiWherein
Figure FDA0003466812080000032
CN201810836824.0A 2018-07-26 2018-07-26 Video key frame extraction method based on saliency region Active CN108921130B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810836824.0A CN108921130B (en) 2018-07-26 2018-07-26 Video key frame extraction method based on saliency region

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810836824.0A CN108921130B (en) 2018-07-26 2018-07-26 Video key frame extraction method based on saliency region

Publications (2)

Publication Number Publication Date
CN108921130A CN108921130A (en) 2018-11-30
CN108921130B true CN108921130B (en) 2022-03-01

Family

ID=64418225

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810836824.0A Active CN108921130B (en) 2018-07-26 2018-07-26 Video key frame extraction method based on saliency region

Country Status (1)

Country Link
CN (1) CN108921130B (en)

Families Citing this family (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111597980B (en) * 2018-12-17 2023-04-28 北京嘀嘀无限科技发展有限公司 Target object clustering method and device
CN109815852A (en) * 2019-01-03 2019-05-28 深圳壹账通智能科技有限公司 Smart city event management method, device, computer equipment and storage medium
WO2020199198A1 (en) * 2019-04-04 2020-10-08 深圳市大疆创新科技有限公司 Image capture control method, image capture control apparatus, and movable platform
CN110290426B (en) * 2019-06-24 2022-04-19 腾讯科技（深圳）有限公司 Method, device and equipment for displaying resources and storage medium
CN110267041B (en) * 2019-06-28 2021-11-09 Oppo广东移动通信有限公司 Image encoding method, image encoding device, electronic device, and computer-readable storage medium
CN110399847B (en) * 2019-07-30 2021-11-09 北京字节跳动网络技术有限公司 Key frame extraction method and device and electronic equipment
CN111400528B (en) * 2020-03-16 2023-09-01 南方科技大学 Image compression method, device, server and storage medium
CN111444826B (en) * 2020-03-25 2023-09-29 腾讯科技（深圳）有限公司 Video detection method, device, storage medium and computer equipment
CN111639601B (en) * 2020-05-31 2022-05-13 石家庄铁道大学 Video key frame extraction method based on frequency domain characteristics
CN111738117B (en) * 2020-06-12 2023-12-19 鞍钢集团矿业有限公司 Deep learning-based detection method for electric bucket tooth video key frame
CN112949560B (en) * 2021-03-24 2022-05-24 四川大学华西医院 Method for identifying continuous expression change of long video expression interval under two-channel feature fusion
CN114897762B (en) * 2022-02-18 2023-04-07 众信方智(苏州)智能技术有限公司 Automatic positioning method and device for coal mining machine on coal mine working face
CN114422807B (en) * 2022-03-28 2022-10-21 麒麟软件有限公司 Transmission optimization method based on Spice protocol
CN114727021B (en) * 2022-04-19 2023-09-15 柳州康云互联科技有限公司 Cloud in-vitro diagnosis image data processing method based on video analysis
CN117475381B (en) * 2023-12-22 2024-03-29 济宁久邦工程机械设备有限公司 Real-time monitoring system for operation state of aerial working platform

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103761738A (en) * 2014-01-22 2014-04-30 杭州匡伦科技有限公司 Method for extracting video sequence key frame in three-dimensional reconstruction
CN104754403A (en) * 2013-12-27 2015-07-01 Tcl集团股份有限公司 Method and system for video sequential alignment
CN104954791A (en) * 2015-07-01 2015-09-30 中国矿业大学 Method for selecting key frame from wireless distributed video coding for mine in real time
CN105100688A (en) * 2014-05-12 2015-11-25 索尼公司 Image processing method, image processing device and monitoring system
CN105574063A (en) * 2015-08-24 2016-05-11 西安电子科技大学 Image retrieval method based on visual saliency

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6965645B2 (en) * 2001-09-25 2005-11-15 Microsoft Corporation Content-based characterization of video frame sequences
KR101537174B1 (en) * 2013-12-17 2015-07-15 가톨릭대학교 산학협력단 Method for extracting salient object from stereoscopic video
CN103747240B (en) * 2013-12-25 2015-10-21 浙江大学 The vision significance filtering method of Fusion of Color and movable information
US9922411B2 (en) * 2015-11-30 2018-03-20 Disney Enterprises, Inc. Saliency-weighted video quality assessment
CN106952286B (en) * 2017-03-21 2019-09-06 中国人民解放军火箭军工程大学 Dynamic background Target Segmentation method based on movement notable figure and light stream vector analysis

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104754403A (en) * 2013-12-27 2015-07-01 Tcl集团股份有限公司 Method and system for video sequential alignment
CN103761738A (en) * 2014-01-22 2014-04-30 杭州匡伦科技有限公司 Method for extracting video sequence key frame in three-dimensional reconstruction
CN105100688A (en) * 2014-05-12 2015-11-25 索尼公司 Image processing method, image processing device and monitoring system
CN104954791A (en) * 2015-07-01 2015-09-30 中国矿业大学 Method for selecting key frame from wireless distributed video coding for mine in real time
CN105574063A (en) * 2015-08-24 2016-05-11 西安电子科技大学 Image retrieval method based on visual saliency

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
图像技术在高清节目监测中的应用;管红等;《广播与电视技术》;20101015;154-155 *
基于人眼视觉特性的视频质量评价算法;谭晶晶;《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》;20150215(第2期);第5.1节,5.2.2节 *

Also Published As

Publication number Publication date
CN108921130A (en) 2018-11-30

Similar Documents

Publication Publication Date Title
CN108921130B (en) Video key frame extraction method based on saliency region
CN109919981B (en) Multi-feature fusion multi-target tracking method based on Kalman filtering assistance
CN104376003B (en) A kind of video retrieval method and device
CN105574515B (en) A kind of pedestrian recognition methods again under non-overlapping visual field
Omidyeganeh et al. Video keyframe analysis using a segment-based statistical metric in a visually sensitive parametric space
Priya et al. Edge strength extraction using orthogonal vectors for shot boundary detection
Huang et al. A novel method for video moving object detection using improved independent component analysis
CN110969101A (en) Face detection and tracking method based on HOG and feature descriptor
CN104504162B (en) A kind of video retrieval method based on robot vision platform
CN111242003B (en) Video salient object detection method based on multi-scale constrained self-attention mechanism
CN109271902B (en) Infrared weak and small target detection method based on time domain empirical mode decomposition under complex background
CN115393788B (en) Multi-scale monitoring pedestrian re-identification method based on global information attention enhancement
CN114821482A (en) Vector topology integrated passenger flow calculation method and system based on fisheye probe
Zhang et al. The target tracking method based on camshift algorithm combined with sift
Guangjing et al. Research on static image recognition of sports based on machine learning
CN110830734B (en) Abrupt change and gradual change lens switching identification method and system
Yi et al. Adaptive threshold based video shot boundary detection framework
Prabakaran et al. Key frame extraction analysis based on optimized convolution neural network (ocnn) using intensity feature selection (ifs)
CN113496159B (en) Multi-scale convolution and dynamic weight cost function smoke target segmentation method
Patil et al. Detection and tracking of moving object: A survey
CN111060887B (en) Gm-APD laser radar low signal-to-noise ratio echo data signal extraction method based on concave-convex search
CN113888428A (en) Infrared dim target detection method and device based on local contrast
Wang et al. Tracking salient keypoints for human action recognition
CN112926676B (en) False target identification method and device and computer equipment
Yunzuo et al. Key Frame Extraction Algorithm of Surveillance Video Based on Quaternion Fourier Significance Detection.

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
GR01 Patent grant
GR01 Patent grant