Video Sample Extraction via Structural Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video classification methods using deep learning require large datasets and often suffer from uneven sample quality, as the samples collected may not adequately cover the content information of the entire video, leading to suboptimal training of video classification models.
Innovation Solution
A sample extraction method that parses video data into consecutive single-frame images, extracts feature images by comparing structural similarities, and arranges them chronologically, ensuring that the samples summarize content information without redundant data, and optionally generates optical flow images to enhance training samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If consecutive video frames are taken out as samples, then a large number of training samples can be collected, but the sample quality becomes uneven and may not adequately cover the content information of the entire video
Solution Approach 1:
The patent extracts feature images from video frames by comparing structural similarities and removing redundant frames. Instead of using all consecutive frames as samples, it selectively extracts only the distinctive feature images that capture essential content information, thereby improving sample quality while maintaining sufficient quantity for training.
2Loss of information
If all consecutive video frames are used as samples, then comprehensive content coverage is achieved, but redundant image data increases sample size and reduces training efficiency
Solution Approach 1:
The patent extracts only the essential feature images from the video frames by comparing structural similarities. This extraction process removes redundant duplicate frames while preserving the key content information, achieving a balance between information coverage and data volume efficiency.
3Manufacturing precision
If feature images are extracted by comparing structural similarities, then redundant image data is removed and sample quality is improved, but the computational complexity increases
Solution Approach 1:
The patent extracts feature images by comparing structural similarities between frames and removing redundant ones. This extraction approach improves sample quality by eliminating duplicates while maintaining a manageable computational process focused on similarity comparison rather than processing all frames.
Data Source
AI summary
A sample extraction method targeting a video classification problem includes acquiring video data, parsing the video data to obtain a plurality of consecutive single-frame images corresponding to the video data, and extracting feature images from the plurality of consecutive single-frame images to compose samples, where the feature images are used to summarize content information of the video data, and the samples do not contain redundant image data.