Text-Guided Image Selection Through Score-Distribution Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image selection systems based on text information often result in the selection of highly similar images, particularly in time series, which can lead to imbalanced and less varied training data for deep learning models, affecting recognition accuracy.
Innovation Solution
An image selection device and method that calculates a score representing the degree of match between text information and images, and selects images by sampling based on the distribution of these scores to diversify the selected images, reducing the bias towards similar images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If images are selected based on the goodness of fit between text information and each candidate image, then the selection accuracy is improved, but the diversity of selected images deteriorates
Solution Approach 1:
The patent changes the selection parameter from deterministic (selecting only images above a threshold score) to probabilistic (sampling based on score distribution). By using the score as a probability weight rather than a binary threshold, the system maintains selection accuracy while enabling diversity through stochastic sampling.
Solution Approach 2:
The patent introduces dynamic sampling where the selection process adapts to the score distribution. Instead of static threshold-based selection, the system dynamically samples images according to their relative scores, allowing the selection to balance accuracy and diversity based on the inherent variability in the score distribution.
2Stability of the object's composition
If successive images from a moving image are selected, then the temporal continuity is improved, but the variety of selected images deteriorates
Solution Approach 1:
The patent applies periodic sampling to the temporal sequence of images. By sampling images at periodic intervals based on their score distribution rather than selecting all successive images, the system maintains temporal continuity for relevant scenes while preventing over-representation of similar successive frames.
Solution Approach 2:
The patent changes the temporal selection parameter from inclusive (selecting all images above threshold) to probabilistic sampling. This allows the system to maintain temporal continuity when scores indicate relevant content while naturally skipping redundant successive frames through the random sampling process.
Data Source
AI summary
The image selection device 1X includes a text information acquisition means 30X, a score calculation means 34X, and an image selection means 35X. The text information acquisition means 30X is configured to acquire text information specifying an image to be acquired from an image group. The score calculation means 34X is configured to calculate a score which represents a degree of match between each image of the image group and the text information. The image selection means 35X is configured to select images from the image group by sampling based on a distribution of the score.