Saliency-Adaptive Snapshot Compressive Imaging for Efficient Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional snapshot compressive imaging (SCI) systems disregard high-level image information, leading to inefficient resource allocation and low sampling efficiency due to the independence between computer visions, resulting in suboptimal computation and reconstruction quality.
Innovation Solution
A saliency-aware and self-adaptive snapshot compressive imaging system that incorporates saliency detection to dynamically update coding masks based on saliency maps, assigning higher sampling probabilities to salient regions and lower probabilities to non-salient regions, optimizing the sampling process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If random binary masks are used for all pixels in traditional SCI systems, then the sampling process is simple and uniform, but high-level information such as objects and saliency is completely disregarded leading to inefficient resource allocation
Solution Approach 1:
The patent applies local quality by differentiating the sampling process across different image regions. Instead of uniform random sampling, the system generates adaptive coding masks where sampling probability varies locally based on saliency detection results. Salient regions receive higher sampling probabilities while non-salient regions receive lower probabilities, optimizing resource allocation according to local information importance.
Solution Approach 2:
The system changes the sampling parameter (sampling probability) dynamically based on image content. By introducing saliency detection, the sampling probability parameter is adjusted from a fixed uniform value to a variable value that reflects the importance of different image regions, thereby preserving high-level information while maintaining sampling efficiency.
2Device complexity
If traditional SCI systems process each frame independently, then the processing pipeline is simple, but sampling efficiency is hampered due to independence between computer vision tasks
Solution Approach 1:
The patent implements feedback by using saliency detection results from previously reconstructed frames to guide the sampling process of subsequent frames. The saliency maps generated from reconstruction outcomes feed back into the adaptive coding mask generation, creating a closed-loop system that continuously optimizes sampling efficiency based on actual image content and reconstruction quality.
Solution Approach 2:
The system performs preliminary saliency detection and adaptive coding mask generation before the actual compression sampling of each frame. By preparing the sampling strategy in advance based on previous frame information, the system optimizes the sampling process beforehand, improving efficiency without adding complexity to the core sampling mechanism.
3Ease of operation
If uniform sampling is applied to all image regions, then the sampling method is straightforward, but resource allocation is wasteful for content-irrelevant computation
Solution Approach 1:
The patent applies local quality by differentiating the sampling process across different image regions. Instead of uniform random sampling, the system generates adaptive coding masks where sampling probability varies locally based on saliency detection results. Salient regions receive higher sampling probabilities while non-salient regions receive lower probabilities, optimizing resource allocation according to local information importance.
Solution Approach 2:
The system applies partial action by concentrating sampling resources on salient regions rather than uniformly sampling all regions. By assigning higher sampling probabilities only to important regions identified through saliency detection, the system performs sampling action where it is most needed, reducing wasteful resource allocation to content-irrelevant areas while maintaining overall reconstruction quality.
Data Source
AI summary
The present invention provides a self-adaptive and saliency-aware snapshot compressive imaging (SCI) system. In comparison with the existing SCI systems, the self-adaptive and saliency-aware SCI system of the present invention integrates saliency detection, which feedbacks to the coding masks with a calculated sampling efficiency for updating the coding masks. As such, self-adaptation of the system is achieved, and high-level information such as saliency is also obtained and given consideration, thereby producing reconstruction results with better quality, lower power cost and higher efficiency.