Object Masking Using Adaptive Thresholds Across Image Resolutions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods in video applications suffer from reduced precision in masking objects, leading to unnecessary masking and difficulty in identifying objects that should not be masked.
Innovation Solution
A method involving downscaling the input image for object detection, setting adaptive and fixed masking thresholds based on the ratio of output and object detection image resolutions, and using confidence scores to selectively mask objects, ensuring accurate masking and identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automatic object detection is used to mask objects in video images, then the masking process can be performed automatically in real time, but the precision of object detection is reduced leading to missed objects that should be masked
Solution Approach 1:
The image processing is segmented into multiple resolution levels. The input image is downscaled to create an object detection image at lower resolution for fast automatic detection, while the original high-resolution image is preserved for final masking application. This segmentation allows each processing stage to operate at appropriate resolution levels, maintaining both real-time performance and detection precision.
Solution Approach 2:
A downscaled object detection image serves as an intermediary representation between the original high-resolution input image and the final output image. This intermediary allows the object detection module to operate on reduced-resolution data for real-time performance, while confidence scores from this intermediary are used to guide masking decisions on the full-resolution image, preserving detection precision.
2Ease of operation
If a fixed masking threshold is used, then the masking process is simple and consistent, but objects at different distances and resolutions cannot be effectively distinguished
Solution Approach 1:
The masking threshold is transformed from a fixed value to a dynamic, adaptive threshold that changes based on the resolution ratio between the output image and the object detection image. The adaptive masking threshold is calculated as: adaptive masking threshold = fixed masking threshold × (object detection image resolution / output image resolution). This dynamic adjustment allows the system to maintain simplicity in implementation while achieving precision in object identification across varying distances and resolutions.
Solution Approach 2:
The masking threshold parameter is changed adaptively based on the resolution ratio. By modifying the threshold parameter according to the relationship between detection image resolution and output image resolution, the system can effectively distinguish objects at different distances and scales, improving identification accuracy while maintaining operational simplicity through a systematic parameter adjustment rule.
3Speed
If the object detection image resolution is reduced for faster processing, then real-time detection is achieved, but the ability to detect objects accurately is compromised
Solution Approach 1:
The processing is segmented into two resolution levels: a downscaled object detection image for fast real-time detection operations, and the original high-resolution input image for maintaining detection accuracy. This segmentation allows the system to achieve real-time processing speeds while preserving the ability to detect objects accurately by operating on appropriate resolution levels for each function.
Solution Approach 2:
A downscaled copy of the input image is created as an object detection image. This copy allows the object detection module to perform fast processing on reduced-resolution data, while the original high-resolution image remains intact and can be used with confidence scores from the detection copy to ensure accurate object identification in the final output.
Data Source
Figure 1
Figure 2
AI summary
An image processing device (200), a non-transitory computer readable medium, and a method (100) is provided for generating an output image with masking of objects of classes to be masked. An input image is downscaled (S110) to an object detection image having a resolution lower than a resolution of the input image and lower than a resolution of the output image. The object detection image is inputted (S120) to an object detection module and confidence scores for pixels or pixel areas of the object detection image are received (5130) from the object detection module. Each confidence score indicates a respective probability that the pixel or pixel area relates to an object of a class to be masked. Based on the input image an intermediate image is generated (S140) having a resolution higher than the object detection image resolution and an adaptive masking threshold is set (5150) such that the greater the ratio between the output image resolution and the object detection image resolution, the lower the masking threshold. The output image is then generated (S160) by masking pixels or pixel areas of the intermediate image corresponding to pixels or pixel areas of the object detection image having a confidence score higher than the adaptive masking threshold.