Shot based keyframe extraction using edge-LBP approach
| Autor: | B. S. Rashmi, H.K. Chethan, H.M. Nandini |
|---|---|
| Rok vydání: | 2022 |
| Předmět: |
General Computer Science
Local binary patterns Computer science business.industry Frame (networking) ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION 020206 networking & telecommunications Sobel operator Pattern recognition 02 engineering and technology TRECVID Standard deviation Euclidean distance Position (vector) Histogram 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Artificial intelligence business |
| Zdroj: | Journal of King Saud University - Computer and Information Sciences. 34:4537-4545 |
| ISSN: | 1319-1578 |
| DOI: | 10.1016/j.jksuci.2020.10.031 |
| Popis: | Advancement in technology has led to tremendous increase in the online video content that requires efficient and effective content based video analysis approaches. In this regard, efficient approach for abrupt Shot Boundary Detection (SBD) and keyframe extraction has been presented. The proposed method detects abrupt shots by extracting binarized edge information from frames for texture characterisation using Local Binary Pattern (LBP) method. Further, Euclidean distance has been applied on the histogram features constructed and an adaptive threshold is used to detect abrupt shots. During keyframe extraction phase, magnitude gradient using Sobel operator has been extracted from each frame of the segmented shot. Subsequently, magnitude values are transformed into Z-score which describes the position of each pixel in terms of its distance from the mean, when measured in standard deviation units of every frame. Finally, Co-efficient of variation is computed for each frame and the frame possessing the highest value is selected as a keyframe from every shot. Experiments were conducted on TRECVID 2001 dataset to analyze and validate the proposed approach. Experimental result manifest that the proposed SBD and keyframe extraction method outperforms some of the state-of-the-art algorithms with average F1-score of 98.15% and average fidelity measure of 90% respectively. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|