EANet: Depth Estimation Based on EPI of Light Field

Area of Science:

• Computer Vision
• Medical Imaging
• Computational Photography

Background:

• Light field imaging captures spatial information crucial for scene understanding.
• Accurate depth estimation is vital for applications like intelligent medical treatment.
• Existing methods may not fully leverage the rich spatial information in light fields.

Purpose of the Study:

• To develop a novel method for extracting depth information from light field images.
• To enhance the accuracy and reliability of depth maps for medical applications.
• To provide a foundation for advanced intelligent medical treatment systems.

Main Methods:

• Designed an attention module to extract salient features from light field images.
• Integrated an attention map with convolutional neural networks to weight subaperture viewpoints.
• Optimized initial depth estimation results for improved performance.

Main Results:

• The proposed method demonstrated significant improvements in depth map quality.
• Mean Squared Error (MSE) decreased by approximately 13%.
• Peak Signal-to-Noise Ratio (PSNR) increased by about 10 dB, and Structural Similarity Index Measure (SSIM) improved by 4% in specific scenarios.

Conclusions:

• The attention-based approach effectively extracts depth information from light fields.
• The method enhances depth map accuracy, making it suitable for medical applications.
• This work contributes to the advancement of intelligent medical treatment through improved depth estimation.