The impact of colour visual attention for video summarization

Abstract

A High Definition visual attention based video summarization algorithm is proposed to extract feature frames and create a video summary. Specifically, the proposed framework is used as the basis for establishing whether or not there is a measurable impact on summaries constructed when choosing to incorporate visual attention mechanisms into the processing pipeline. The algorithm was assessed against manual human generated key-frame summaries presented with tested datasets from the Open Video Dataset (www.open-video.org). Of the frames selected by the algorithm, up to 68.1% were in agreement with the manual frame summaries depending on the category and length of the video. Specifically, a clear impact of agreement rate with the ground truth is demonstrated when including colour-attention models (in general) into the summarization framework, with the proposed colour-attention model achieving stronger agreement with human selected summaries, than other models from the literature.