Video Scene Search Using Temporal Interval Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for searching similar scenes within video content databases face challenges due to high computational load and large data size, particularly when scenes are reordered or edited, leading to potential search failures.
Innovation Solution
An image processing device and method that acquire and calculate the similarity between appearance patterns of individuals in video content scenes, using a search pattern and a search target pattern to determine similar scenes based on calculated similarity, reducing data size and computational load by utilizing temporal information and noise removal techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If spatial information such as color space histogram or edge histogram is used as feature value, then search accuracy is improved, but computational load and data size increase significantly
Solution Approach 1:
The patent extracts only the necessary temporal information (time intervals between scene change points) from the video content, separating it from unnecessary spatial information. This extraction of essential temporal features reduces computational load and data size while maintaining search effectiveness for reordered or edited content.
Solution Approach 2:
The patent changes the feature value parameter from complex spatial information (color histograms, edge histograms) to simple temporal information (time intervals between scene changes). This parameter transformation reduces computational complexity while preserving the ability to identify similar scenes even when reordered or edited.
2Device complexity
If scene change point intervals are used as feature values, then computational load is reduced, but search reliability decreases when scenes are re-ordered or content is edited
Solution Approach 1:
The patent transitions from using absolute time positions of scene changes to using relative time intervals between consecutive scene changes. This dimensional transformation creates a feature representation that is invariant to reordering and editing operations, maintaining search reliability while keeping computational load low.
Solution Approach 2:
Instead of directly using the absolute timing of scene changes, the patent inverts the approach by using the intervals between scene changes as the feature. This inversion makes the feature representation robust against reordering and editing, as the interval pattern remains characteristic of the original scene sequence even when timestamps change.
3Measurement precision
If comprehensive spatial and temporal feature values are extracted, then search precision is improved, but data size becomes large
Solution Approach 1:
The patent extracts only the essential temporal component (time intervals between scene changes) from the video content, discarding redundant spatial information. This selective extraction maintains search precision for temporal patterns while significantly reducing the data size required for storage and processing.
Data Source
AI summary
An image processing device includes an evaluation unit which acquires a search pattern which is an appearance pattern of a person in a scene of a video content containing the scene, where a similar scene is to be searched, and a search target pattern which is an appearance pattern of a person in a video content which is to be a search target of the similar scene, calculates a degree of similarity between the search pattern and the search target pattern, and determines the similar scene on a basis of the calculated degree of similarity.