Edge-Based Defocus Blur Estimation With Adaptive Scale Selection.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Defocus blur occurs in digital images when objects are not at the camera's focal distance.
• Accurate blur estimation is crucial for image deblurring and restoration tasks.

Purpose of the Study:

• To develop a novel edge-based method for estimating spatially varying defocus blur from a single image.
• To improve the accuracy and efficiency of defocus blur estimation compared to existing techniques.

Main Methods:

• The method utilizes reblurred gradient magnitudes for blur estimation.
• It involves computing a scale-consistent edge map and selecting local reblurring scales.
• A connected edge filter and a guided filter are employed for smoothing and propagating the blur map.

Main Results:

• The proposed method achieves a good balance between estimation error and computational time.
• It demonstrates superior performance compared to current state-of-the-art methods.
• Evaluation in image deblurring suggests that standard blur estimation metrics may not always correlate with visual quality.

Conclusions:

• The presented edge-based approach offers an effective solution for spatially varying defocus blur estimation.
• The findings highlight the importance of considering visual quality beyond traditional metrics in deblurring applications.