Still Image Extraction from Video Using Adaptive Feature Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for extracting still images from moving image data face challenges in balancing calculation accuracy and processing load, particularly in detecting and extracting frames with specific features like faces, which are often spread across multiple frames rather than appearing in a single frame.
Innovation Solution
The image processing apparatus employs a method that samples moving image data at predetermined intervals, calculates feature values for target frames and adjacent frames, and extracts still image data based on these values, using techniques like OpenCV and machine learning to identify and group frames with feature values exceeding predetermined thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature values are calculated for all frame images to ensure accurate detection, then measurement precision is improved, but processing load increases
Solution Approach 1:
The patent divides the frame images into two groups: sampling target frame images (first group) and non-sampling target frame images (second group). Feature values are calculated only for the first group at predetermined sampling periods, reducing the number of calculations while maintaining detection accuracy through strategic sampling.
Solution Approach 2:
The patent performs preliminary sampling and feature value calculation on a subset of frame images before final extraction. By pre-identifying candidate frames with interesting features through sampling, the system avoids exhaustive analysis of all frames, thus reducing processing load while ensuring accurate detection.
2Productivity
If sampling period is increased to reduce processing load, then productivity is improved, but measurement precision deteriorates
Solution Approach 1:
The patent dynamically adjusts the sampling period based on the detected feature values. When interesting features are detected in sampled frames, the system reduces the sampling period (increases sampling frequency) to capture more frames around the feature occurrence, ensuring accurate detection while maintaining overall processing efficiency.
Solution Approach 2:
The system uses feedback from detected feature values to adjust the sampling strategy. When features are detected, the sampling period is modified to capture additional relevant frames, ensuring that detection accuracy is maintained even with variable sampling rates that optimize processing speed.
3Quantity of substance
If all frame images are extracted to ensure complete capture of features, then quantity of substance is improved, but loss of energy increases
Solution Approach 1:
The patent extracts only the necessary frame images that contain interesting features rather than all frames. By using feature value calculations and threshold comparisons on sampled frames, the system identifies and extracts only the relevant frames, reducing the total number of extracted images and associated computational energy.
Solution Approach 2:
The system discards frame images that do not contain interesting features by not extracting them, while recovering and preserving only the frames that meet the extraction criteria. This selective approach reduces energy consumption associated with processing and storing unnecessary frames.
Data Source
AI summary
An image processing apparatus includes a still image data creation unit and a still image data extraction unit. The still image data extraction unit calculates a first feature value of the sampling target still image data, determines whether the sampling target still image data is first extraction target still image data as an extraction target based on the first feature value, and calculates a second feature value of adjacent still image data. The adjacent still image data are a plurality of still image data in a range preset in time series of the first extraction target still image data. The still image data extraction unit determines whether the adjacent still image data is second extraction target still image data as an extraction target or not based on the second feature value and extracts the first extraction target still image data and the second extraction target still image data.