Scene Recognition Using Sensor and Histogram Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies fail to accurately determine whether a scene is High-Dynamic Range (HDR) or non-HDR, which is crucial for synthesizing high-quality images in HDR scenes.
Innovation Solution
A scene recognition method and apparatus that acquires image and sensor data, determines if the scene is non-HDR using sensor data, and if not, extracts image features like dark and bright area pixel ratios and histogram variations to accurately classify the scene as HDR, categorizing it into different types for precise HDR synthesis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If only sensor data is used for scene determination, then processing speed is improved, but recognition accuracy deteriorates when sensor data is insufficient
Solution Approach 1:
The scene determination process is segmented into two stages: first using sensor data for quick preliminary determination, then using image histogram features for accurate HDR scene recognition when needed. This segmentation allows the system to achieve both speed and accuracy by applying different methods at different stages.
Solution Approach 2:
The patent introduces image histogram features as an intermediary between sensor data and final scene determination. When sensor data alone is insufficient, the histogram features serve as additional evidence to accurately determine HDR scenes, bridging the gap between quick sensor-based determination and accurate scene recognition.
2Measurement precision
If image histogram features are extracted for all images, then recognition accuracy is improved, but processing time increases
Solution Approach 1:
The system performs preliminary scene determination using sensor data before extracting image histogram features. This preliminary action filters out images that can be determined from sensor data alone, avoiding unnecessary histogram extraction and processing time while maintaining accuracy for images that require detailed analysis.
Solution Approach 2:
Instead of extracting histogram features for all images, the system applies partial action by extracting features only for images where sensor data is insufficient for determination. This selective approach reduces overall processing time while maintaining recognition accuracy for the subset of images that need detailed analysis.
3Measurement precision
If multiple image features are extracted and multiple thresholds are applied, then determination accuracy is improved, but system complexity increases
Solution Approach 1:
The determination system is segmented into distinct modules: sensor data processing, histogram feature extraction, underexposure determination, and overexposure determination. Each module handles specific features and thresholds independently, making the complex system more manageable and maintainable while preserving high determination accuracy.
Solution Approach 2:
Different features and thresholds are applied locally to different aspects of scene determination: sensor data for preliminary determination, dark area histogram features for underexposure detection, and bright area histogram features for overexposure detection. This localized application of quality measures improves accuracy without requiring all features to be processed uniformly throughout the system.
Data Source
Figure 1
Figure 2~3
AI summary
The present invention relates to a scene recognition method and apparatus. The scene recognition method includes: acquiring an image and sensor data corresponding to the image; performing, according to the sensor data, non-high-dynamic range HDR scene determining on a scene of the image; extracting an image feature of the image when it is incapable of determining whether the scene of the image is a non-HDR scene; and determining, according to the image feature, whether the scene of the image is an HDR scene.