Depth Hypotheses Fusion through 3D Weighted Least Squares in Shape from Focus

Area of Science:

• Computer Vision
• 3D Reconstruction
• Computational Imaging

Background:

• Traditional shape-from-focus (SFF) methods often use a single focus measure, limiting depth accuracy for varied object shapes.
• Existing SFF techniques frequently neglect structural or prior information, resulting in incomplete object shape details.

Purpose of the Study:

• To develop an improved shape-from-focus method addressing the limitations of single focus measures and lack of prior information.
• To enhance the accuracy and detail of 3D shape reconstruction by incorporating structural priors.

Main Methods:

• Obtained depth hypotheses by applying multiple focus operators to image sequences.
• Extracted structural prior information (guidance volume) from focus measure volumes.
• Applied 3D weighted least squares optimization, using weights derived from the guidance volume, to refine the depth hypothesis volume.

Main Results:

• The proposed method successfully combined depth hypotheses and structural priors for more accurate 3D shape computation.
• Experimental results on synthetic and real microscopic objects demonstrated significant improvements in depth map quality.
• Comparative analysis confirmed the effectiveness of the enhanced SFF technique.

Conclusions:

• Integrating multiple focus operators and structural priors via 3D weighted least squares significantly enhances shape-from-focus accuracy.
• The proposed method overcomes limitations of single-measure operators and provides more detailed 3D shape reconstructions.
• This approach offers a robust solution for accurate 3D shape recovery from image sequences.