Index-Based Ranking for Large Multimedia Archives
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multimedia archive searching systems face inefficiencies due to the vast amount of irrelevant data and the slow application of discriminative classifiers on large datasets, particularly when features do not cleanly separate classes, leading to inaccurate nearest neighbor-type searching.
Innovation Solution
The implementation of a metric-based indexing system that organizes features using pivot points and a priority queue to accelerate the evaluation of discriminative classifiers, allowing for rapid exploration of relevant parts of the feature space and prioritizing features likely to be positively identified, thereby reducing the time needed to retrieve top representatives of a certain class.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If nearest neighbor-type searching is used to search multimedia archives, then search speed is improved, but classification accuracy deteriorates when features do not cleanly separate classes
Solution Approach 1:
The patent segments the feature space into multiple classes using discriminative classifiers (e.g., SVMs) that probabilistically label features. Instead of relying on a single nearest neighbor search, the system divides the search space into distinct class regions, allowing for more accurate classification while maintaining search efficiency through indexed retrieval of features in each class region.
2Measurement precision
If discriminative classifiers are applied to large datasets to improve classification accuracy, then accuracy is improved, but processing speed deteriorates
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing class probability estimates for features in an index structure before actual search queries are executed. During search, these pre-computed probabilities are retrieved and used to rank features, eliminating the need to re-evaluate the discriminative classifier on the entire large dataset for each query, thus maintaining high accuracy while improving processing speed.
3Reliability
If all features in the multimedia archive are searched to ensure comprehensive results, then completeness is improved, but time consumption increases
Solution Approach 1:
The patent applies local quality by focusing the search on specific regions of the feature space that are most likely to contain relevant data. Using the index structure with class probability estimates, the system prioritizes features with higher probabilities of being relevant, allowing the search to concentrate computational resources on promising regions rather than uniformly examining all features, thus reducing time consumption while maintaining reliability.
Data Source
AI summary
Devices, systems, and methods of performing feature detection on a set of multimedia files are disclosed. One method of organization includes identifying a feature from each multimedia file within the set of multimedia files wherein each file has one feature, organizing the features based on their similarities wherein similar features are grouped based upon a proximity in a feature space and a representative feature is identified for each group, receiving a detection model having one or more detection criteria the detection model having previously been trained for detection using the organized features, and using the representative features to apply the detection model in a decreasing order of detection probability in order to detect the files satisfying the detection criteria within the set of multimedia files.