Salient Region Guided Blind Image Sharpness Assessment
Siqi Liu1,2, Shaode Yu1,2, Yanming Zhao1,2
1Key Laboratory of Convergent Media and Intelligent Technology (Communication University of China), Ministry of Education, Beijing 100024, China.
Sensors (Basel, Switzerland)
|July 2, 2021
Summary
Salient region detection improves blind image sharpness assessment (BISA). Using salient regions as input for BISA models leads to comparable or better performance than using whole images, especially with the SORBD method.
Area of Science:
- Computer Vision
- Image Processing
- Machine Learning
Background:
- Salient regions are crucial for human scene understanding.
- The utility of salient regions in blind image blur estimation remains unexplored.
Purpose of the Study:
- To propose a salient region guided blind image sharpness assessment (BISA) framework.
- To investigate the impact of detected salient regions on BISA performance.
Main Methods:
- A hybrid framework integrating three salient region detection (SRD) methods with ten BISA models.
- Re-organizing saliency maps as input for BISA models to quantify performance changes.
- Experiments conducted on three Gaussian blurred image databases.
Main Results:
- Salient region input achieved performance comparable or superior to whole image input for BISA models.
- The Saliency Optimization from Robust Background Detection (SORBD) method consistently improved score prediction.
- The proposed framework demonstrates the benefit of saliency detection for image blur estimation.
Conclusions:
- Salient region detection positively impacts blind image sharpness assessment.
- Further research will focus on optimal integration of SRD and BISA models for enhanced score prediction.
