CN115830508A - 5G message content detection method - Google Patents
5G message content detection method Download PDFInfo
- Publication number
- CN115830508A CN115830508A CN202211600092.8A CN202211600092A CN115830508A CN 115830508 A CN115830508 A CN 115830508A CN 202211600092 A CN202211600092 A CN 202211600092A CN 115830508 A CN115830508 A CN 115830508A
- Authority
- CN
- China
- Prior art keywords
- frame
- violation
- video
- frames
- characteristic
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 13
- 230000011218 segmentation Effects 0.000 claims abstract description 83
- 238000000034 method Methods 0.000 claims abstract description 29
- 238000012216 screening Methods 0.000 claims abstract description 10
- 230000002159 abnormal effect Effects 0.000 claims description 22
- 238000004364 calculation method Methods 0.000 claims description 21
- 238000005192 partition Methods 0.000 claims description 19
- 230000003044 adaptive effect Effects 0.000 claims description 14
- 238000004891 communication Methods 0.000 abstract description 7
- 238000012544 monitoring process Methods 0.000 abstract description 4
- XOFYZVNMUHMLCC-ZPOLXVRWSA-N prednisone Chemical compound O=C1C=C[C@]2(C)[C@H]3C(=O)C[C@](C)([C@@](CC4)(O)C(=O)CO)[C@@H]4[C@@H]3CCC2=C1 XOFYZVNMUHMLCC-ZPOLXVRWSA-N 0.000 description 5
- 102000003712 Complement factor B Human genes 0.000 description 2
- 108090000056 Complement factor B Proteins 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the technical field of communication network monitoring, in particular to a 5G message content detection method. Firstly, acquiring video information in a 5G message; screening out characteristic frames in the video image; matching each frame of video image in the initially screened violation video with the characteristic frame to obtain a multi-frame violation frame; when the feature frame is matched with different violation frames, self-adaptive segmentation is carried out to obtain the self-adaptive segmentation size, the feature frame and the violation frames are segmented to obtain segmented areas, and area common factors corresponding to the segmented areas in the feature frame are calculated; obtaining an integral common factor according to the region common factor of each divided region; calculating violation stability according to the fluctuation degree of the overall common factor; and identifying the violation condition of the video information according to the violation stability. According to the method, the content of the self-adaptive area is identified and matched by the characteristic frame and the violation frame in the violation video database, and violation identification can be carried out after the violation video is edited.
Description
Technical Field
The invention relates to the technical field of communication network monitoring, in particular to a 5G message content detection method.
Background
With the development and the upgrade of communication technology, the former short message communication service is upgraded to 5G message and is fully commercial. Compared with the prior short message communication, the 5G message is more diversified and supports the communication of various types of contents. The message brings convenience to people and brings certain challenges, such as monitoring and treatment of bad messages, different from the previous text message, which only needs to monitor a small amount of text information, detection of various types of content of the 5G message is more complicated than the previous detection of the text information, especially, the identification of the content of the video in the 5G message is more difficult, and the provision of the high-efficiency, low-delay, comprehensive and accurate bad message identification is a key for guaranteeing the user experience for improving the user experience.
When a video message in a 5G message is identified, in order to meet the requirement of low detection delay, a currently universal detection method for video content is to match hash values of videos, that is, a hash algorithm is used for generating corresponding hash values for the video content in the 5G message, and then the illegal video is identified by comparing the hash values with hash values of illegal videos in an illegal video library, which are already stored, but the hash values are extremely sensitive to the content of the videos, that is, after the same illegal video is subjected to certain content clipping, the hash values of the videos are different, so that the success rate of illegal video identification is reduced, and the illegal videos after clipping are difficult to identify.
Disclosure of Invention
In order to solve the above technical problem, an object of the present invention is to provide a method for detecting content of a 5G message, which adopts the following technical solutions:
acquiring video information in the 5G message; preliminarily matching the video information with the violation videos in the violation video database to obtain preliminarily screened violation videos;
screening out characteristic frames in the video images according to the information entropy of each frame of video image in the video information; matching each frame of video image in the initially screened violation video with the characteristic frame to obtain a plurality of frames of violation frames; performing self-adaptive segmentation when the feature frame is matched with different violation frames to obtain the self-adaptive segmentation size;
segmenting the feature frame and the violation frame based on the self-adaptive segmentation size to obtain at least two segmentation areas, and obtaining an area common factor corresponding to each segmentation area in the feature frame according to the similarity between each segmentation area in the feature frame and each segmentation area in the violation frame; obtaining an integral common factor according to the region common factor of each partition region for the feature frame and any corresponding violation frame;
calculating violation stability according to the fluctuation degree of the overall common factor; and identifying the violation condition of the video information in the 5G message according to the violation stability.
Preferably, the calculation formula of the adaptive segmentation size is as follows:
wherein, X m,j The adaptive segmentation size is obtained when the characteristic frame of the mth frame is matched with the corresponding violation frame of the jth frame; exp is an exponential function with a natural constant as a base; d j The Hamming distance of the hash value of the mth frame characteristic frame and the jth frame violation frame is obtained; AVG (D) m,J ) The mean value of the Hamming distances of the hash values of the mth frame characteristic frame and all corresponding violation frames is obtained; NG m The number of pixel points in the mth frame of the feature frame is the number of the pixel points in the mth frame of the feature frame; NG' j The number of pixel points in the violation frame of the jth frame is calculated;
is rounding up the symbol.
Preferably, the obtaining of the region common factor corresponding to each segmented region in the feature frame according to the similarity between each segmented region in the feature frame and each segmented region in the violation frame includes:
selecting any segmentation area in the characteristic frame as a target segmentation area;
calculating the similarity between the target segmentation area and each segmentation area in the violation frame; selecting the maximum similarity as a region common factor corresponding to the target segmentation region;
the calculation formula of the similarity is as follows:
wherein, XS mk,j′k Similarity of the kth segmentation region in the feature frame of the mth frame and the kth segmentation region in the violation frame of the jth frame is obtained; x m,j The adaptive segmentation size is obtained when the characteristic frame of the mth frame is matched with the corresponding violation frame of the jth frame; gmk i The gray value of the ith pixel point in the kth partition area in the mth frame of feature frame; gj' k i The gray value of the ith pixel point in the kth partition area in the jth frame violation frame is obtained; gmk i′ The gray value of the ith pixel point in the kth partition area in the mth frame of feature frame corresponding to the ith' pixel point in the eight neighborhoods; gj' k i′ The gray value of the ith pixel point in the kth partition area in the jth frame violation frame corresponding to the ith' pixel point in the eight neighborhoods is obtained; exp is an exponential function with a natural constant as the base.
Preferably, the preliminarily matching the video information and the violation video in the violation video database to obtain a preliminarily screened violation video includes:
acquiring a hash value of the video information as a first hash value; acquiring a hash value of any illegal video in an illegal video database as a second hash value; and calculating Hamming distances between the first Hash value and the second Hash value, and selecting the violation video with the Hamming distance smaller than a preset first threshold value as an initially screened violation video.
Preferably, the screening out the feature frames in the video image according to the information entropy of each frame of video image in the video information includes:
selecting any video image as a target image, acquiring information entropy corresponding to the target image, and calculating abnormal values of the target image relative to the front and rear frames of video images according to the information entropy corresponding to the target image;
and when the abnormal value is greater than or equal to a preset second threshold value, taking the target image as a characteristic frame in the video image.
Preferably, the calculating an abnormal value of the target image relative to the two previous and next frames of video images according to the information entropy corresponding to the target image includes:
acquiring information entropies of front and back frame video images corresponding to a target image; calculating the average value of the information entropy of the target image and the front and rear frames of video images as the average value of the information entropy; calculating the absolute value of the difference value between the information entropy of the target image and the information entropy mean value; and the ratio of the absolute value of the difference value to the information entropy of the target image is an abnormal value corresponding to the target image.
Preferably, the matching of each frame of video image in the initially screened violation video and the feature frame to obtain a multi-frame violation frame includes:
acquiring a hash value of any characteristic frame as a third hash value; selecting any initially screened violation video as a target video, and acquiring a hash value of any video image in the target video as a fourth hash value; calculating Hamming distances of the third Hash value and the fourth Hash value, and taking a video image in the target video corresponding to the minimum Hamming distance as an illegal frame;
and for any characteristic frame, obtaining multiple violation frames corresponding to the initially screened violation videos, and obtaining multiple violation frames.
Preferably, the calculating the violation stability according to the fluctuation degree of the overall common factor includes:
and for any feature frame, acquiring the overall common factor of each violation frame corresponding to the feature frame, calculating the difference value of the overall common factors corresponding to the adjacent violation frames, and taking the sum of the difference values as violation stability.
Preferably, the identifying the violation condition of the video information in the 5G message according to the violation stability includes:
when the violation stability corresponding to any feature frame is greater than or equal to a preset third threshold, the corresponding video information is a violation video; otherwise, the corresponding video information is a normal video.
Preferably, the overall common factor is: the mean of the area common factors of the respective divided areas.
The embodiment of the invention at least has the following beneficial effects:
firstly, acquiring video information in a 5G message; preliminarily matching the video information with the violation videos in the violation video database to obtain preliminarily screened violation videos; although the whole video of the illegal video is changed from the original illegal video before clipping after a certain clipping, the illegal video still has a certain relation, so that the illegal video in the illegal video database is preliminarily matched. The feature frames in the video images are screened out, and the feature frames are extracted to be used for being matched with the illegal video in the illegal video database, so that the calculation amount of video matching can be reduced, and the matching accuracy can be improved. Matching each frame of video image in the initially screened violation video with the characteristic frame to obtain a plurality of frames of violation frames; based on the hamming distance between frames, when the feature frame is matched with different violation frames, the self-adaptive segmentation is carried out to obtain the self-adaptive segmentation size, and the feature frame and the violation frames are segmented to obtain the segmentation area. Obtaining a region common factor corresponding to each segmentation region in the feature frame according to the similarity between each segmentation region in the feature frame and each segmentation region in the violation frame; obtaining an integral common factor according to the region common factor of each partition region for the feature frame and any corresponding violation frame; calculating violation stability according to the fluctuation degree of the overall common factor; and identifying the violation condition of the video information in the 5G message according to the violation stability. The method extracts the characteristic frames of the video information in the 5G message, and performs adaptive area size content identification and matching on the characteristic frames and the violation frames in the violation video database, so that the violation identification can be performed after the violation video is clipped compared with the content identification of the video information in the existing 5G message, and the method has higher force for the violation video identification.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a flowchart of a method for detecting content of a 5G message according to an embodiment of the present invention.
Detailed Description
To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description will be given for the 5G message content detection method according to the present invention, and the specific implementation, structure, features and effects thereof with reference to the accompanying drawings and the preferred embodiments. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The embodiment of the invention provides a specific implementation method of a 5G message content detection method, which is suitable for an illegal identification scene of video information in a 5G message. In order to solve the problem that the hash value is extremely sensitive to the content of a video, the hash values of the videos are different after the same illegal video is edited with a certain content, so that the success rate of illegal video identification is reduced, and the illegal video after being edited is difficult to identify. According to the method, the characteristic frame of the video information in the 5G message is extracted, and the content identification and matching of the size of the self-adaptive area are carried out on the characteristic frame and the illegal frame in the illegal video database.
The following describes a specific scheme of the 5G message content detection method provided by the present invention in detail with reference to the accompanying drawings.
Referring to fig. 1, a flowchart illustrating steps of a method for detecting content of a 5G message according to an embodiment of the present invention is shown, where the method includes the following steps:
step S100, acquiring video information in the 5G message; and preliminarily matching the video information with the violation videos in the violation video database to obtain preliminarily screened violation videos.
In order to achieve the purpose of reducing workload when identifying the violation condition of the video information in the 5G message, firstly, the video information in the 5G message is preliminarily matched with the violation video in the violation video database to obtain a preliminarily screened violation video corresponding to the current video information. In the embodiment of the invention, the violation video database is manually identified and screened to contain non-repetitive violation contents.
The primary matching method comprises the following steps: acquiring a hash value of video information as a first hash value; acquiring a hash value of any illegal video in an illegal video database as a second hash value; and calculating Hamming distances between the first Hash value and the second Hash value, and selecting the violation video with the Hamming distance smaller than a preset first threshold value as an initially screened violation video. In the embodiment of the present invention, the value of the first threshold is preset to be 10, and in other embodiments, an implementer may adjust the value according to an actual situation. Namely, a hash value is generated by video information in the 5G message of the content to be identified, then Hamming distances are calculated by the hash value with all illegal videos in the illegal video database, and then the illegal video with the smaller Hamming distance is selected as a preliminarily screened illegal video of the 5G message to be identified. The purpose of the preliminary matching is that although the hash value corresponding to the whole video of the illegal video is changed from the hash value of the illegal video which is not edited originally after the illegal video is edited to a certain extent, the illegal video still has a certain relationship, so that the 5G message to be identified is preliminarily screened by using the Hamming distance of the hash value, and the preliminarily screened illegal video corresponding to the video information in the 5G message is obtained.
S200, screening out characteristic frames in the video images according to the information entropy of each frame of video image in the video information; matching each frame of video image in the initially screened violation video with the characteristic frame to obtain a plurality of frames of violation frames; and when the characteristic frame is matched with different violation frames, performing adaptive segmentation to obtain the adaptive segmentation size.
The process of identifying the violation content of the video message in the 5G message is to extract a feature frame by using the video message in the 5G message, obtain violation frames by using a hash value and a hamming distance of a video image of the violation video in a violation video database of each extracted feature frame, perform adaptive-size region division on each violation video in each feature frame and a corresponding violation frame, further calculate the similarity between each division region of the feature frame and each division region of the violation frame in a corresponding violation frame set to obtain a region common factor, obtain an overall common factor of each feature frame and the corresponding violation frame by using the region common factor, and finally identify whether the current feature frame is violation or not by using the stability of the overall common factor, thereby identifying whether the video in the 5G message is the violation video.
Firstly, according to the information entropy of each frame of video image in the video information, screening out the characteristic frame in the video image, namely extracting the characteristic frame by using the video information in the 5G message.
The characteristic frames of the video information in the 5G message refer to video images with large differences with other frame video images in all frame video images of the video information in the 5G message, the video images are often not strong in relation to the whole video, and the characteristic frames are often omitted when ordinary clipping software clips illegal videos in a database, so that the characteristic frames are extracted to be matched with the illegal videos in the illegal video database, the calculation amount of video matching can be reduced, and the accuracy of matching can be improved. Taking video information in a certain 5G message as an example, the extraction mode of the characteristic frame is to select any video image as a target image, obtain the information entropy corresponding to the target image, and calculate the abnormal value of the target image relative to the front and back two frames of video images according to the information entropy corresponding to the target image; and when the abnormal value is greater than or equal to a preset second threshold value, taking the target image as a characteristic frame in the video image. In the embodiment of the present invention, the value of the preset second threshold is 0.45, and in other embodiments, an implementer may adjust the value according to an actual situation. The specific steps for obtaining the feature frame are as follows:
firstly, the overall information amount of each frame of video image is calculated for the video information, namely, the information entropy of each frame of video image is calculated. Take the nth frame video image in the video information as an example, the corresponding information entropy E n The calculation method is as follows:
wherein p is g The frequency of occurrence of pixel points with the gray value of g in the nth frame of video image is set; log is a logarithmic function. The frequency of occurrence is calculated by dividing the number of occurrences of g pixels in the nth frame of video image by the number of all pixels in the nth frame, wherein g belongs to [0,255 ]]. It should be noted that the formula for calculating the entropy is well known to those skilled in the art, and will not be described herein.
Further, difference value calculation based on continuous frame information quantity is carried out on continuous frames to screen the characteristic frames in the multiple video images, namely, abnormal values of the target image relative to the two frames of video images before and after are calculated according to the information entropy corresponding to the target image; and when the abnormal value is greater than or equal to a preset second threshold value, taking the target image as a characteristic frame in the video image.
The formula for the outlier is:
wherein, delta E n An abnormal value of the nth frame video image; e n The information entropy of the nth frame video image is obtained;
the average value of the information entropy of the nth frame video image and the two previous and next frame video images is obtained.
Wherein the information entropy reflects the whole information quantity of the video image; n is an element of [2,N]And N is the maximum frame number corresponding to the current video information. The logic of the formula for calculating the outlier is: firstly, taking the nth frame of video image as a basis, calculating an average value of the overall information content of the nth frame of video image and the two consecutive frames of video images, and then calculating a difference with the overall information content of the nth frame of video image, wherein the larger the difference is, the more abnormal the frame of video image is reflected, the larger the abnormal value of the corresponding video image is; conversely, the smaller the difference is, the more normal the frame video image is reflected, and the smaller the abnormal value of the corresponding video image is. In the video information, generally, each frame of video has little difference in the videos of the consecutive frames, so the difference between the average value of the total information amount corresponding to the video images of the preceding and following consecutive frames corresponding to the video image of the nth frame and the average value of the total information amount corresponding to the video image of the nth frame is correspondingly small. However, when a special frame is encountered, the difference between the special frame and the previous and next continuous frame video images is strong, so that the difference between the average value of the whole information quantity of the corresponding continuous frame video and the average value of the whole information quantity of the nth frame video image is very large, the special frame video is quantized, and then the abnormal value delta E of the nth frame video image is represented by calculating the ratio of the average value of the whole information quantity of the corresponding continuous frame video and the average value of the whole information quantity of the nth frame video image n Abnormal value Δ E n The larger the abnormal value is, the smaller the relation between the video image of the nth frame and the video images of the previous and subsequent frames is reflected, and conversely, the abnormal value delta E is n The smaller the size, the greater the relationship between the video image of the nth frame and the video images of the preceding and succeeding frames. When abnormal value Δ E n And if the size is larger, the nth frame video image can be considered as a characteristic frame. That is, when the abnormal value is greater than or equal to the preset second threshold, the nth frame video image is considered as the characteristic frame.
Screening all frame video images of video information in the 5G message to obtain all M characteristic frames M in the video images in all the 5G messages, wherein the M belongs to the 1,N; and N is the total number of the video images corresponding to the video information.
After the characteristic frames in the video images are screened out, matching each frame of video image in the initially screened violation video with the characteristic frame to obtain a multi-frame violation frame.
Firstly, all frames in all initially screened violation videos in an initially screened violation video database and M characteristic frames in extracted 5G messages are subjected to Hash value acquisition. It should be noted that the manner of obtaining the hash value is a known technique of those skilled in the art, and is not described herein.
And then, calculating Hamming distances by using the hash values of any one of the M characteristic frames and all frame video images of the initially screened violation videos in the initially screened violation video library, and selecting one violation frame with the smallest Hamming distance from each initially screened violation video as a violation frame set of each characteristic frame, namely, each characteristic frame has the violation frame violation set composed of the violation frames with the smallest Hamming distances from different initially screened violation videos.
Obtaining a plurality of violation frames corresponding to each feature frame is also: acquiring a hash value of any characteristic frame as a third hash value; selecting any initially screened violation video as a target video, and acquiring a hash value of any video image in the target video as a fourth hash value; calculating Hamming distances of the third Hash value and the fourth Hash value, and taking a video image in the target video corresponding to the minimum Hamming distance as an illegal frame; and for any characteristic frame, obtaining multiple violation frames corresponding to the initially screened violation videos, and obtaining multiple violation frames.
After the violation frame corresponding to each feature frame is obtained, the segmentation sizes of the feature frames and the matched violation frames are self-adapted according to the hamming distance between the frames, namely, the feature frames are self-adapted to be segmented when matched with different violation frames based on the hamming distance between the frames, so that the self-adapted segmentation sizes are obtained.
A violation frame set which is similar to each frame of the M characteristic frames of the 5G message in comparison is obtained by using the Hamming distance, each violation frame in the violation frame set corresponding to each characteristic frame is likely to have part of the same content, but the same content is likely to be not in the same position, so that the characteristic frames and the violation frames in the violation frame set corresponding to each characteristic frame are subjected to self-adaptive segmentation based on the Hamming distance, each characteristic frame and the violation frames in the violation frame set corresponding to each characteristic frame are segmented by using the self-adaptive segmentation to obtain at least two segmentation areas, and then the different areas in the characteristic frames and the violation frames are subjected to content identification and matching.
Specifically, taking the mth feature frame in the feature frames as an example, the calculation formula of the adaptive segmentation size of the mth feature frame in the feature frame set and the jth violation frame in the corresponding violation frame set is as follows:
wherein, X m,j The adaptive segmentation size is obtained when the characteristic frame of the mth frame is matched with the corresponding violation frame of the jth frame; exp is an exponential function with a natural constant as a base; d j The Hamming distance of the hash value of the mth frame characteristic frame and the jth frame violation frame is obtained; AVG (D) m,J ) The mean value of the Hamming distances of the hash values of the mth frame characteristic frame and all corresponding violation frames is obtained; NG m The number of pixel points in the mth frame of the feature frame is the number of the pixel points in the mth frame of the feature frame; NG' j The number of pixel points in the violation frame of the jth frame is calculated;
is rounding up the symbol.
Wherein X m,j The side length of each divided region when the m-th feature frame and the j-th illegal frame in the corresponding illegal frame set are subjected to content identification and matching, namely the size of the divided region is X m,j ×X m,j ，j∈[1,J]Wherein J is all violation frames in the violation frame set corresponding to the mth feature frameThe total number of the cells.
The adaptive partition size calculation formula logic: hamming distance D of hash values of characteristic frame of mth frame and violation frame of jth frame j The greater the Hamming distance relative to the entire violation frame set, then | D j -AVG(D m,J ) The larger the | is, the less the content of the m-th frame feature frame and the content of the j-th frame violation frame are compared with the rest violation frames, so that the segmentation area should be smaller when the m-th frame feature frame and the j-th frame violation frame are subjected to area division, and the matching result in subsequent matching is more accurate. Conversely, if the content of the m-th frame feature frame is the same as that of the j-th frame violation frame, then | D j -AVG(D m,J ) The smaller the | is, the same content can be easily recognized and matched even if the area is large, so that the divided area is enlarged to reduce the calculation amount in the subsequent matching. And then negating the natural constant-based negative exponential function to obtain a smaller parameter when the Hamming distance is larger and obtain a larger parameter when the Hamming distance is smaller, and then averaging all pixel points of the mth characteristic frame and the jth violation frame
And as a basis, multiplying the obtained parameter to obtain the adaptive segmentation size when the content matching and identification are carried out on the mth characteristic frame and the jth illegal frame in the follow-up process.
All the feature frames are calculated, and the adaptive segmentation size corresponding to all the feature frames and each frame violation frame in the violation frame set corresponding to all the feature frames can be obtained. The Hamming distance of the hash value of each characteristic frame and each violation frame in the corresponding violation frame set is utilized, and the self-adaptive segmentation size during subsequent content matching and identification is obtained.
Step S300, segmenting the feature frame and the violation frame based on the self-adaptive segmentation size to obtain at least two segmentation areas, and obtaining an area common factor corresponding to each segmentation area in the feature frame according to the similarity between each segmentation area in the feature frame and each segmentation area in the violation frame; and for the characteristic frame and any corresponding violation frame, obtaining an overall common factor according to the region common factor of each divided region.
In order to accurately identify illegal contents in the feature frames, each feature frame and the corresponding illegal frame are segmented by using the obtained self-adaptive segmentation size to obtain a plurality of at least two segmentation areas. And performing content matching calculation on the common factors of the regions by using each segmented region in the feature frame and each segmented region in the corresponding violation frame, namely obtaining the common factors of the regions corresponding to the segmented regions in the feature frame according to the similarity between each segmented region in the feature frame and each segmented region in the violation frame, and further calculating the overall common factor of each feature frame by using the common factors of the regions.
The method comprises the steps of segmenting a feature frame and an illegal frame based on self-adaptive segmentation size to obtain at least two segmentation areas, obtaining an area common factor corresponding to each segmentation area in the feature frame according to the similarity between each segmentation area in the feature frame and each segmentation area in the illegal frame, and taking an mth feature frame and a jth illegal frame as examples, wherein the area common factor B is j The calculation method is as follows:
firstly, the m-th characteristic frame and the j-th violation frame are subjected to area division according to the corresponding self-adaptive division size, so that m 'division areas corresponding to the m-th characteristic frame and j' division areas corresponding to the j-th violation frame after division can be obtained. And performing similarity calculation on each segmented region in the m-th characteristic frame and each segmented region in the j violation frame. Namely, any segmented region in the feature frame is selected as a target segmented region, and the similarity between the target segmented region and each segmented region in the violation frame is calculated.
The similarity calculation formula is as follows:
wherein XS mk,j′k Similarity of a kth segmented region in the m frame feature frame and a kth segmented region in the j frame violation frame; x m,j Self-adaption during matching of characteristic frame of mth frame and corresponding violation frame of jth frameDividing the sizes; gmk i The gray value of the ith pixel point in the kth segmentation region in the mth frame feature frame is obtained; gj' k i The gray value of the ith pixel point in the kth partition area in the jth frame violation frame is obtained; gmk i′ The gray value of the ith pixel point in the kth partition area in the mth frame of feature frame corresponding to the ith' pixel point in the eight neighborhoods; gj' k i′ The gray value of the ith pixel point in the kth partition area in the jth frame violation frame corresponding to the ith' pixel point in the eight neighborhoods is obtained; exp is an exponential function with a natural constant as the base.
Wherein, the gray value Gm1 i ∈[0,255]，i∈[1,X m,j ×X m,j ]，Gm1 i′ ∈[0,255]，i′∈[1,8]，Gj′1 i′ ∈[0,255]，Gj′1 i ∈[0,255]。
The logic of the similarity calculation formula is as follows: sharing X in the first segmentation area of the m-th frame characteristic frame and the j-th frame violation frame m,j ×X m,j And the pixel points at each position are utilized to calculate the difference value of the gray values of the pixel points at the corresponding positions so as to represent the similarity of the two segmentation areas, and if the pixel points of the two segmentation areas are more similar, the similarity value is smaller. However, there is a certain contingency in this case, although the pixel point at the ith position in the first region of the mth feature frame and the jth violation frame are the same, it is impossible to provide necessary contribution to the entire region, so that the average value calculation of the difference values is performed on the pixel points in the 8 neighborhoods around each position pixel point, so as to constrain the pixel point at each position, if the pixel points at the ith position are similar and the pixel points in the 8 neighborhoods around the ith position are similar in the two segmentation regions, it is indicated that the pixel point at the ith position is not only similar, and the contribution provided by the pixel point at the ith position when the similarity calculation is performed on the two regions is not accidental, and the specific constraint mode is to use the average value of the difference values of the pixel points in the 8 neighborhoods around the pixel point at the ith position as a specific constraint mode
With the ith of the two divided regionsThe average value of the differences of the pixel points of the positions is obtained, so that even if the pixel point of the ith position in the two partition areas shows very small accidental differences, the difference value of the pixel points of 8 adjacent areas around the pixel points is also large under the influence of the average value of the difference values. All xs in both partitions are then mapped in this way m,j ×X m,j And calculating each pixel point to obtain similarity, wherein the larger the value of the similarity is, the larger the similarity between the two segmentation areas is, and otherwise, the smaller the value of the similarity is, the smaller the similarity is.
By using the above manner to calculate the similarity between the first segmented region in the mth feature frame and each segmented region in the jth violation frame, the similarity between the first segmented region in the mth feature frame and each segmented region in the jth violation frame can be obtained. Then, the maximum similarity is selected, namely the maximum representative similarity is the region common factor B of the first segmentation region in the m-th characteristic frame 1 That is, after the similarity between the target segmentation region and each segmentation region in the violation frame is calculated, the maximum similarity is selected as the region common factor corresponding to the target segmentation region.
The region common factor of all the divided regions in the mth feature frame can be obtained by calculation.
Further, for the feature frame and any corresponding violation frame, an overall common factor is obtained according to the region common factor of each partition region, specifically: the average value of the region common factors of the respective divided regions is an overall common factor. For example, for each segmentation area obtained when the m-th frame feature frame is matched with the corresponding jth frame violation frame, the area common factor of each segmentation area is calculated, and further, the overall common factor when the m-th frame feature frame is matched with the jth frame violation frame is calculated.
The overall common factor is calculated by the formula:
wherein, bm j The global common factor is the global common factor when the m-th frame characteristic frame is matched with the j-th frame violation frame; m' is the number of the segmented regions of the characteristic frame of the mth frame; b is m″ And when the m < th > frame feature frame is matched with the jth frame violation frame, the area common factors of the m' area in the m < th > frame feature frame.
The logic of the calculation formula of the overall common factor is as follows: when the feature frame of the mth frame is matched with the violation frame of the jth frame, the average value of the common factors of the regions of all the segmented regions in the feature frame is obtained, the larger the average value is, the segmented regions with higher similarity are reflected in each segmented region in the mth feature frame and the violation frame of the jth frame in the matching process, and the higher the similarity degree is, otherwise, the smaller the average value is, the segmented regions with higher similarity are not reflected in each segmented region in the mth feature frame and the violation frame of the jth frame in the matching process, and the similarity degree between the segmented regions is relatively lower.
The overall common factor of each frame of the violation frame in the mth characteristic frame and the corresponding violation frame set is obtained through calculation, that is, how many violation frames exist in the violation frame set, how many overall common factors exist in the mth characteristic frame.
Step S400, calculating violation stability according to the fluctuation degree of the overall common factor; and identifying the violation condition of the video information in the 5G message according to the violation stability.
And after obtaining the overall common factor, detecting the illegal video information in the 5G message by using the overall common factor. The J overall common factors of the mth feature frame and the J violation frames in the violation frame set corresponding to the mth feature frame are obtained as described above, and then whether the mth feature frame is the violation frame or not is detected through the J overall common factors. Calculating violation stability according to fluctuation degrees of integral common factors corresponding to J violation frames, specifically: and for any feature frame, acquiring the overall common factor of each violation frame corresponding to the feature frame, calculating the difference value of the overall common factors corresponding to adjacent violation frames, and taking the sum of the difference values as the violation stability corresponding to the feature frame.
The calculation formula of the violation stability is as follows:
wherein, delta m The violation stability corresponding to the mth frame feature frame; bm j The overall common factor is the overall common factor when the m frame characteristic frame is matched with the j frame violation frame; bm j-1 The global common factor is the global common factor when the m frame characteristic frame is matched with the j-1 frame violation frame; j is the number of violation frames corresponding to the m frame feature frames.
If the mth characteristic frame is the violation frame, the mth characteristic frame is certainly similar to one of the J violation frames in the corresponding violation frame set, namely the mth characteristic frame is larger in overall common factor and smaller in similarity with the rest J-1 violation video frames, namely the mth characteristic frame is smaller in overall common factor, and the mth characteristic frame is smaller in overall common factor m The closer to 0, the less possibility that the m-th characteristic frame is the violation frame, and the stability delta when the violation is detected m The larger the probability that the mth feature frame is an offending frame.
And calculating the violation stability of the overall common factor for each frame of the M characteristic frames corresponding to the video information in the 5G message, so as to obtain the corresponding violation stability of all the M characteristic frames.
And whether the video information in the 5G message is the violation video is performed by using the corresponding violation stabilities of all M characteristic frames, namely. After the violation stability is obtained, identifying the violation condition of the video information in the 5G message according to the violation stability, specifically: when the violation stability corresponding to any feature frame is greater than or equal to a preset third threshold, the corresponding video information is a violation video; otherwise, the corresponding video information is a normal video. In the embodiment of the present invention, the value of the preset third threshold is 0.4, and in other embodiments, an implementer may adjust the value according to an actual situation.
And whether the video information in the 5G message violates the video identification is finished through the corresponding violation stabilities of the M characteristic frames. Further, the illegal video in the identified 5G message is processed, for example, the violation stability of the overall common factor of the feature frame is utilized to judge whether the video information content in the 5G message belongs to the illegal video, and subsequently, the 5G message containing the illegal video can be shielded, and the sender is identified according to the sending mobile phone number of the 5G message and is tracked.
In summary, the present invention relates to the field of communication network monitoring technology. The method comprises the steps of firstly obtaining video information in a 5G message; preliminarily matching the video information with the violation videos in the violation video database to obtain preliminarily screened violation videos; screening out characteristic frames in the video image; matching each frame of video image in the initially screened violation video with the characteristic frame to obtain a plurality of frames of violation frames; based on the Hamming distance between frames, carrying out self-adaptive segmentation when the feature frame is matched with different violation frames to obtain the self-adaptive segmentation size, segmenting the feature frame and the violation frames to obtain segmentation areas, and obtaining an area common factor corresponding to each segmentation area in the feature frame according to the similarity between each segmentation area in the feature frame and each segmentation area in the violation frames; obtaining an integral common factor according to the region common factor of each partition region for the characteristic frame and any corresponding violation frame; calculating violation stability according to the fluctuation degree of the overall common factor; and identifying the violation condition of the video information in the 5G message according to the violation stability. The method extracts the characteristic frames of the video information in the 5G message, and performs adaptive area size content identification and matching on the characteristic frames and the violation frames in the violation video database, so that the method is more accurate compared with the content identification of the video information in the existing 5G message, and has higher force for violation video identification.
It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. The processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments.
Claims (10)
1. A5G message content detection method is characterized by comprising the following steps:
acquiring video information in the 5G message; preliminarily matching the video information with the violation videos in the violation video database to obtain preliminarily screened violation videos;
screening out characteristic frames in the video images according to the information entropy of each frame of video image in the video information; matching each frame of video image in the initially screened violation video with the characteristic frame to obtain a plurality of frames of violation frames; performing self-adaptive segmentation when the feature frame is matched with different violation frames to obtain the self-adaptive segmentation size;
segmenting the feature frame and the violation frame based on the self-adaptive segmentation size to obtain at least two segmentation areas, and obtaining an area common factor corresponding to each segmentation area in the feature frame according to the similarity between each segmentation area in the feature frame and each segmentation area in the violation frame; obtaining an integral common factor according to the region common factor of each partition region for the feature frame and any corresponding violation frame;
calculating violation stability according to the fluctuation degree of the overall common factor; and identifying the violation condition of the video information in the 5G message according to the violation stability.
2. The method of claim 1, wherein the adaptive partition size is calculated according to the following formula:
wherein, X m,j Self-matching of characteristic frame of mth frame with corresponding violation frame of jth frameThe method is suitable for the size of the segmentation; exp is an exponential function with a natural constant as a base; d j The Hamming distance of the hash value of the mth frame characteristic frame and the jth frame violation frame is obtained; AVG (D) m,J ) The mean value of the Hamming distances of the hash values of the mth frame characteristic frame and all corresponding violation frames is obtained; NG m The number of pixel points in the mth frame feature frame is set; NG' j The number of pixel points in the violation frame of the jth frame is set;
is rounding up the symbol.
3. The method according to claim 1, wherein obtaining the region common factor corresponding to each segment in the feature frame according to the similarity between each segment in the feature frame and each segment in the violation frame comprises:
selecting any segmentation area in the characteristic frame as a target segmentation area;
calculating the similarity between the target segmentation area and each segmentation area in the violation frame; selecting the maximum similarity as a region common factor corresponding to the target segmentation region;
the calculation formula of the similarity is as follows:
wherein,
similarity of a kth segmented region in the m frame feature frame and a kth segmented region in the j frame violation frame; x m,j The adaptive segmentation size is obtained when the characteristic frame of the mth frame is matched with the corresponding violation frame of the jth frame; gmk i The gray value of the ith pixel point in the kth segmentation region in the mth frame feature frame is obtained; gj' k i The gray value of the ith pixel point in the kth partition area in the jth frame violation frame is obtained; gmk i′ For the kth partition in the m frame feature frameThe ith pixel point in the domain corresponds to the gray value of the ith' pixel point in the eight neighborhoods; gj' k i′ The gray value of the ith pixel point in the kth partition area in the jth frame violation frame corresponding to the ith' pixel point in the eight neighborhoods is obtained; exp is an exponential function with a natural constant as the base.
4. The method for detecting content of 5G messages according to claim 1, wherein the preliminary matching of the video information and the violation video in the violation video database to obtain a preliminary screening violation video comprises:
acquiring a hash value of the video information as a first hash value; acquiring a hash value of any illegal video in the illegal video database as a second hash value; and calculating Hamming distances between the first Hash value and the second Hash value, and selecting the violation video with the Hamming distance smaller than a preset first threshold value as an initially screened violation video.
5. The method as claimed in claim 1, wherein the screening out the characteristic frames in the video image according to the entropy of each frame of the video image in the video information comprises:
selecting any video image as a target image, acquiring information entropy corresponding to the target image, and calculating abnormal values of the target image relative to the front and rear frames of video images according to the information entropy corresponding to the target image;
and when the abnormal value is greater than or equal to a preset second threshold value, taking the target image as a characteristic frame in the video image.
6. The 5G message content detection method according to claim 5, wherein the calculating of the abnormal value of the target image relative to the two frames of video images according to the information entropy corresponding to the target image comprises:
acquiring information entropies of front and back frame video images corresponding to a target image; calculating the average value of the information entropy of the target image and the front and rear frames of video images as the average value of the information entropy; calculating the absolute value of the difference value between the information entropy of the target image and the average value of the information entropy; and the ratio of the absolute value of the difference value to the information entropy of the target image is an abnormal value corresponding to the target image.
7. The method for detecting content of 5G messages according to claim 1, wherein the step of matching each frame of video image in the initially screened violation video with the characteristic frame to obtain multiple frames of violation frames comprises the steps of:
acquiring a hash value of any characteristic frame as a third hash value; selecting any initially screened violation video as a target video, and acquiring a hash value of any video image in the target video as a fourth hash value; calculating Hamming distances of the third Hash value and the fourth Hash value, and taking a video image in the target video corresponding to the minimum Hamming distance as an illegal frame;
and for any characteristic frame, obtaining multiple violation frames corresponding to the initially screened violation videos, and obtaining multiple violation frames.
8. The method for detecting 5G message content according to claim 1, wherein the calculating the violation stability according to the fluctuation degree of the overall common factor comprises:
and for any feature frame, acquiring the overall common factor of each violation frame corresponding to the feature frame, calculating the difference value of the overall common factors corresponding to adjacent violation frames, and taking the sum of the difference values as violation stability.
9. The method according to claim 1, wherein the identifying the violation of the video information in the 5G message according to the violation stability comprises:
when the violation stability corresponding to any characteristic frame is greater than or equal to a preset third threshold value, the corresponding video information is a violation video; otherwise, the corresponding video information is a normal video.
10. The method of claim 1, wherein the overall common factor is: the mean of the area common factors for each divided area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211600092.8A CN115830508B (en) | 2022-12-13 | 2022-12-13 | 5G message content detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211600092.8A CN115830508B (en) | 2022-12-13 | 2022-12-13 | 5G message content detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115830508A true CN115830508A (en) | 2023-03-21 |
| CN115830508B CN115830508B (en) | 2024-03-08 |
Family
ID=85547012
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211600092.8A Active CN115830508B (en) | 2022-12-13 | 2022-12-13 | 5G message content detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115830508B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116866666A (en) * | 2023-09-05 | 2023-10-10 | 天津市北海通信技术有限公司 | Video stream picture processing method and device in rail transit environment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016082277A1 (en) * | 2014-11-28 | 2016-06-02 | 安科智慧城市技术（中国）有限公司 | Video authentication method and apparatus |
| CN109740424A (en) * | 2018-11-23 | 2019-05-10 | 深圳市华尊科技股份有限公司 | Traffic violations recognition methods and Related product |
| CN114581856A (en) * | 2022-05-05 | 2022-06-03 | 广东邦盛北斗科技股份公司 | Agricultural unit motion state identification method and system based on Beidou system and cloud platform |
-
2022
- 2022-12-13 CN CN202211600092.8A patent/CN115830508B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016082277A1 (en) * | 2014-11-28 | 2016-06-02 | 安科智慧城市技术（中国）有限公司 | Video authentication method and apparatus |
| CN109740424A (en) * | 2018-11-23 | 2019-05-10 | 深圳市华尊科技股份有限公司 | Traffic violations recognition methods and Related product |
| CN114581856A (en) * | 2022-05-05 | 2022-06-03 | 广东邦盛北斗科技股份公司 | Agricultural unit motion state identification method and system based on Beidou system and cloud platform |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116866666A (en) * | 2023-09-05 | 2023-10-10 | 天津市北海通信技术有限公司 | Video stream picture processing method and device in rail transit environment |
| CN116866666B (en) * | 2023-09-05 | 2023-12-08 | 天津市北海通信技术有限公司 | Video stream picture processing method and device in rail transit environment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115830508B (en) | 2024-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8358837B2 (en) | Apparatus and methods for detecting adult videos | |
| US7366330B2 (en) | Method, apparatus, and program for detecting faces | |
| US10169655B2 (en) | Detection of logos in a sequence of video frames | |
| US10489916B2 (en) | Method and apparatus for updating a background model | |
| CN111144366A (en) | Strange face clustering method based on joint face quality assessment | |
| CN115908154B (en) | Video later-stage particle noise removing method based on image processing | |
| CN115830508B (en) | 5G message content detection method | |
| US20060222217A1 (en) | Method, apparatus, and program for discriminating faces | |
| CN113657249B (en) | Training method, prediction method, device, electronic equipment and storage medium | |
| CN115083008A (en) | Moving object detection method, device, equipment and storage medium | |
| CN113657370B (en) | Character recognition method and related equipment thereof | |
| US7885474B2 (en) | Solution for detecting a streak in a compressed gray scale image | |
| CN110895811A (en) | Image tampering detection method and device | |
| CN113792661B (en) | Image detection method, device, electronic equipment and storage medium | |
| CN100433047C (en) | Device and method for detecting blurring of image | |
| CN113902899A (en) | Training method, target detection method, device, electronic device and storage medium | |
| CN116364064B (en) | Audio splicing method, electronic equipment and storage medium | |
| CN114283431B (en) | Text detection method based on differentiable binarization | |
| CN113965759A (en) | Video frame compression method and system | |
| CN114399497A (en) | Text image quality detection method and device, computer equipment and storage medium | |
| CN111291602A (en) | Video detection method and device, electronic equipment and computer readable storage medium | |
| CN113361426A (en) | Vehicle loss assessment image acquisition method, medium, device and electronic equipment | |
| CN114627340B (en) | Image steganography detection feature self-adaptive selection method based on triple measurement | |
| CN115082709B (en) | Remote sensing big data processing method, system and cloud platform | |
| Chun et al. | A method for original image recovery for caption areas in video |
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 |