Network sensitive video detection method

Abstract

The invention discloses a novel network sensitive video detection method which comprises the steps that network video is collected, peripheral texts of the video are extracted, then video features and text features in the network video are extracted, the video features comprise audio features and visual features, and the video features and the text features form the feature set of the network video; whether the video is sensitive or not is demarcated manually in sequence; the quality factors of the video-audio features are considered, the extracted feature set is used for computing content rich similarity between words, the text features established before are added, then a classifier core is established jointly, through the obtained classifier core, a network sensitive video classifier is trained with an improved support vector machine algorithm, and finally during classifying, only the text features of a tested sample are extracted to be used as prediction input data. The method can be used for harmful video filtering in the Internet, and the content health and the safety of a computer network can be effectively maintained.

Description

technical field [0001] The invention belongs to the technical field of pattern recognition, in particular to a new method for detecting network-sensitive video. Background technique [0002] With the development of the times, information also explodes. The emergence of the Internet now provides extremely convenient conditions for people to obtain information, but everything has two sides. The Internet age provides us with convenience but also brings many negative effects. For example, the Internet is full of pornography, terror As well as sensitive videos such as violence, pornography and horror videos are known to have a bad impact on people's health, and there are many ways to detect and stop them; but people ignore the dangers of violent videos to people's mental health, especially children. Existing methods rarely detect violent videos, and even if they exist, they are not used in practice for many reasons. [0003] As early as the last century, there were studies on t...

AI Technical Summary

Problems solved by technology

A large number of experimental data show that text-based classifiers are efficient and fast, but the text features in online videos are sparse and cannot represent the content of the video well; classifiers based on visual content can classify videos well, but Its disadvantage is that it is time-consuming and not very efficient. In the field of information fusion, there is a consensus that multi-modal classification results are generally better than single-modal recognition results. Experimental data also prove this point. That is, the classifier based on the fusion of text and visual information is better than the previous text-based and visual-based classification.

As we all know, extracting the text features of an online video is very efficient and takes very little time. No matter how much it contributes to the final classification effect, it does not actually have much impact on the final classification efficiency. However, it takes a lot of time to extract the visual information of the video. The time is relatively long. In reality, the quality of videos on video sharing websites varies, and it is impossible to guarantee that the quality of video features and audio features in any video is very good. Any visual feature that people want to extract It is helpful for the final classification, so it is necessary to evaluate the visual features of the network video and evaluate its quality, so it is necessary to introduce a quality factor to ensure the reliability of the classification results and achieve a more reasonable recognition function

Images