Automatic Generation of Preview Images Based on Video Sequence Analysis Using Computer Vision

The article is devoted to the problems of video analysis, selecting keyframes from it, searching for the most attractive thumbnails in it, as well as analyzing existing approaches to solving these problems. The article offers a new method for selecting attractive frames based on selecting frames with the largest number of attractive objects. This results in a frame that is visually appealing and relevant to the video. The solution also considers the selection of the most attractive thumbnails based on the analysis of various visual qualities and aesthetic indicators of video frames, and then performs cluster analysis to ensure that the resulting images match the video content. The research also considers the method of selecting the frames with the greatest activity through the emotional analysis of people in the video. If necessary, a style is applied to the final image to get a more interesting result.