Joint demosaicing and denoising via learned nonparametric random fields.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Demosaicing reconstructs color images from incomplete sensor data.
• Effective demosaicing requires modeling natural image statistics and handling noise.
• Objective assessment of demosaicing methods is limited by data availability.

Purpose of the Study:

• Introduce a machine learning approach for demosaicing.
• Address challenges of natural image statistics and noise in reconstruction.
• Provide a public dataset for evaluating demosaicing and denoising techniques.

Main Methods:

• Developed a machine learning model trained on a large dataset of natural images.
• The model learns statistical properties of images and noise for simultaneous demosaicing and denoising.
• Utilized a public ground truth dataset for training and evaluation.

Main Results:

• The machine learning approach optimizes user-specified performance measures like PSNR and structural similarity.
• The model can be retrained for novel color filter array layouts.
• Achieved state-of-the-art performance, improving PSNR by up to 0.7 dB and reconstructing details accurately.

Conclusions:

• Machine learning offers a powerful approach to image demosaicing and denoising.
• Discriminative models provide peak performance with reduced engineering effort.
• The proposed method advances the field of computational imaging.