Video Abstraction Method Using Color Moment and Density-Based Clustering Algorithm

Abstract

The video summarization abstracts the essential information of video content. This paper proposed a new video abstraction method to extract meaningful video frames. The proposed method encompasses the following stages: the extracted frames are converted to grayscale images. The quality of the frames is assessed using the NIQE method. The feature vector for each frame is obtained using the kurtosis moment, and the difference between two consecutive feature vectors is calculated. The DBSCAN algorithm is applied to classify these difference values, recording any temporal transitions when a difference value is identified as an outlier. Finally, the frame with the highest NIQE value in each segment is compiled into the video abstraction. The results demonstrated excellent performance of the proposed method, which achieved 100% Accuracy and F- Score. The average time of the abstraction videos is 2.962 seconds. Various factors were analyzed for their impact on the method, revealing that using Euclid’s metric and setting epsilon to 15 yielded the best results for DBSCAN-based temporal segmentation.