Interest-Oriented Video Summarization with Keyframe Extraction

To address the requirements of summarising large video files, we introduce a novel approach incorporating the interest of the user to advance the current approaches to develop a well-defined single summary for a single video. Our approach has the capability of generating multiple summaries for a single video catering to different user interests, thus advancing the current state of the art. The proposed method takes the video to be summarized as well as the user interested feature(s) which we define as Object of Interest, as inputs. Inputs are processed through two pipelines. The first pipeline takes the video frames and divide the video into video shots based on the colour features of the frames. Then a generalized keyframe based summary is generated by extracting static and dynamic features of the video from the defined shots. The second pipeline compares each frame with the Object of Interest and selects the set of frames, which have features similar to the Object of Interest. In order to generate the summary of the video, the proposed solution takes the intersection operation of both pipeline-outputs. Evaluation of the system is conducted against benchmark datasets to demonstrate the effectiveness and validity of the proposed approach.