Method and apparatus for providing video cover

Abstract

The embodiment of the invention discloses a method and apparatus for providing a video cover. The method comprises: a video file uploaded by a user is received, a scene changing key frame is determined according to a changing situation of contents of adjacent frames in the video file, and a picture corresponding to the scene changing key frame is cut out; on the basis of a pre-trained machine learning model for picture classification, the cut pictures are graded and ranked; according to the ranks, the preset number of the pictures having the high scores are used candidate pictures for a video cover and are provided for a user, so that the user can select one from the candidate pictures for the video cover. Therefore, missing of all important scenes in the video file can be avoided; picture repetition in the provided candidate pictures for the video cover can be reduced; the qualities of the candidate pictures are improved; and the user can select a proper video cover conveniently.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a method and device for providing a video cover. Background technique [0002] When we watch a video on a video website, we will see that each video has a video cover on the relevant webpage. The quality of the picture corresponding to the video cover is an important factor to attract users to click on the video, especially for the popular short videos. In other words, the quality of the picture corresponding to the video cover is particularly important. [0003] The selection scheme of the existing video cover is usually according to a fixed time point (for example, a video is divided into several sub-videos on average according to the duration, and the time point when each sub-video starts playing is used as a fixed time point, etc.), from the video However, the video cover pictures obtained in this way often have problems such as blurring and defocusing, or the pi...

AI Technical Summary

Problems solved by technology

[0006] First of all, directly use the pictures uploaded by users as the "high-quality" training set, and use the pictures intercepted from the video at fixed time points as the "low-quality" training set, which will introduce a lot of "dirty data". There may be many pictures with poor quality in the picture, and there may also be many pictures with good quality in the pictures taken from the video at a fixed time point. Therefore, this training set containing "dirty data" will directly lead to training The resulting machine learning model cannot achieve a good classification effect;

[0007] Secondly, when the video file is long, this screenshot method will make the captured pictures more repetitive, and the video cover picture finally provided to the user is likely to be some pictures with relatively high repetition

Images