Video Content-aware Dynamic Subtitles

Imposing user's tags on image and video, such as subtitle, icon and other symbols, is helpful to understand the contents of the image and video. It is one of basic operations in image and video processing. The operation is widely applied in video display, remote education, medical image processing, astronomical observation, digital photographing, copyright protection, internet multimedia, and etc. Among them, subtitle for video is the most frequent case.In general, the subtitles are imposed at central bottom part of video frame. It is a simple and efficient solution. Due to diversities of video contents, the subtitles imposed may occlude some interested or critical contents. Since there are coherence among the successive video images, the content occlusion may be continuous, which will influence visual experiences of audiences. In this paper, a novel video content aware dynamic subtitle method is proposed. An energy function, which takes the video content and coherence into account, is defined and optimized to find an unimportant region in the video image, on which the subtitles are imposed. Unlike the traditional subtitles rasterized and filled with a fixed color, a dynamic subtitle color is proposed according to the average color of video contents. It can increase the contrast between subtitle and video image. The proposed algorithm is. local. There is no coupling in the numerical optimization. Thus it can be accelerated via GPU so that we can impose the dynamic subtitles on the high definition video in real-time. Experimental results and users'investigations show that the video content aware dynamic subtitles can reduce the critical content occlusion and improve the display effects.