Video fingerprint extraction method based on sparse coding

Abstract

The invention discloses a video fingerprint extraction method based on sparse coding. The method comprises the following steps of firstly extracting preselected video frame image features through an SURF algorithm to obtain a training feature point set; performing sparse processing on the training feature point set to further obtain a sparse dictionary; then, performing sparse coding on a video to be processed through the SURF algorithm so that visual dictionary is built and the word frequency information is counted; finally, converting each frame of word frequency statistic result of the video into a series of hexadecimal sequences by using a similarity saving hash algorithm to thus obtain simplest fingerprints of each video frame; and then, connecting the simplest fingerprints in series according to the time sequence of the video frame in the video to obtain the video fingerprints.

Description

technical field

[0001] The invention belongs to the technical field of video fingerprint and data retrieval, and more specifically, relates to a method for extracting video fingerprint based on sparse coding. Background technique

[0002] The image content contained in the video itself has unique temporal and spatial characteristics, and there are large differences between different videos. However, when the same video is subjected to different attacks, its temporal and spatial characteristics are less affected. This uniqueness is similar to The fingerprint characteristics of human beings, we call this feature of video "video fingerprint". As an emerging video image processing technology, video fingerprint extracts and processes video content features, and then achieves the purpose of uniquely representing the video. By comparing the "fingerprint sequences" of two videos, the similarity and dissimilarity between the videos can be found, and then the copyright protection and...

Examples

Embodiment

[0050] figure 1 It is a flow chart of the method for extracting video fingerprints based on sparse coding in the present invention.

[0051] In this example, if figure 1 Shown, a kind of video fingerprint extraction method based on sparse coding of the present invention comprises the following steps:

[0052] S1. Build a sparse dictionary

[0053] S1.1. The sparse dictionary is constructed by alternately updating the base vector and sparse. In this embodiment, the image library of Fergus is selected as the training set, and the SURF algorithm is used to obtain the feature point set of the training image;

[0054] Use the SURF algorithm to extract feature points from the pre-selected dictionary training video library to obtain the training feature point set Wherein, m represents the number of videos in the dictionary training video storehouse; Represents the feature point set of the i-th video in the dictionary training video library, each feature point set represents ...

example

[0095] In this embodiment, a promotional video is selected from Youku.com for simulation, and the extracted SURF feature points of 17 key frames are sparsely encoded, and then the corresponding visual words are searched for these sparse results.

[0096] For a sparse feature point to match its own visual word in the dictionary D of size 1024, the average time consumed is recorded as t sc , and for a non-sparse feature point in the same size D', the average consumption time of matching visual words is recorded as t surf . For a certain frame in this video, which contains an average of 300 feature points, then the word search time before and after the feature points of this video are sparse is shown in Table 2.

[0097] Table 2 is the visual word lookup overhead table before and after sparse coding;

[0098]

[0099] Table 2

[0100] As can be seen from Table 2, in terms of storage, the result of sparse coding can greatly reduce the computational and storage overhead. For...

technical field

[0001] The invention belongs to the technical field of video fingerprint and data retrieval, and more specifically, relates to a method for extracting video fingerprint based on sparse coding. Background technique

[0002] The image content contained in the video itself has unique temporal and spatial characteristics, and there are large differences between different videos. However, when the same video is subjected to different attacks, its temporal and spatial characteristics are less affected. This uniqueness is similar to The fingerprint characteristics of human beings, we call this feature of video "video fingerprint". As an emerging video image processing technology, video fingerprint extracts and processes video content features, and then achieves the purpose of uniquely representing the video. By comparing the "fingerprint sequences" of two videos, the similarity and dissimilarity between the videos can be found, and then the copyright protection and...

Examples

Embodiment

[0050] figure 1 It is a flow chart of the method for extracting video fingerprints based on sparse coding in the present invention.

[0051] In this example, if figure 1 Shown, a kind of video fingerprint extraction method based on sparse coding of the present invention comprises the following steps:

[0052] S1. Build a sparse dictionary

[0053] S1.1. The sparse dictionary is constructed by alternately updating the base vector and sparse. In this embodiment, the image library of Fergus is selected as the training set, and the SURF algorithm is used to obtain the feature point set of the training image;

[0054] Use the SURF algorithm to extract feature points from the pre-selected dictionary training video library to obtain the training feature point set Wherein, m represents the number of videos in the dictionary training video storehouse; Represents the feature point set of the i-th video in the dictionary training video library, each feature point set represents ...

example

[0095] In this embodiment, a promotional video is selected from Youku.com for simulation, and the extracted SURF feature points of 17 key frames are sparsely encoded, and then the corresponding visual words are searched for these sparse results.

[0096] For a sparse feature point to match its own visual word in the dictionary D of size 1024, the average time consumed is recorded as t sc , and for a non-sparse feature point in the same size D', the average consumption time of matching visual words is recorded as t surf . For a certain frame in this video, which contains an average of 300 feature points, then the word search time before and after the feature points of this video are sparse is shown in Table 2.

[0097] Table 2 is the visual word lookup overhead table before and after sparse coding;

[0098]

[0099] Table 2

[0100] As can be seen from Table 2, in terms of storage, the result of sparse coding can greatly reduce the computational and storage overhead. For...

Images