Coupled Deep Autoencoder for Single Image Super-Resolution.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Sparse coding methods for single image super-resolution (SR) often produce artifacts due to rigid assumptions about low-resolution (LR) and high-resolution (HR) image patch representations.
• Existing approaches struggle with ringing, jaggy, and blurring due to limitations in representation mapping.

Purpose of the Study:

• To develop a novel deep learning model for single image super-resolution (SR) that overcomes the limitations of traditional sparse coding methods.
• To introduce a data-driven approach that learns intrinsic representations and precise mappings between LR and HR image patches.

Main Methods:

• A Coupled Deep Autoencoder (CDA) model was developed, featuring a new deep architecture with high representational capacity.
• CDA simultaneously learns distinct representations for LR and HR image patches.
• A big-data-driven function precisely maps learned LR representations to their corresponding HR representations.

Main Results:

• The CDA model demonstrated superior effectiveness in single image super-resolution compared to state-of-the-art methods.
• Experiments on Set5 and Set14 datasets confirmed the efficiency and enhanced performance of CDA.
• The proposed method significantly reduces common artifacts like ringing and jagging in super-resolved images.

Conclusions:

• The Coupled Deep Autoencoder (CDA) offers a robust and effective deep learning solution for single image super-resolution.
• CDA's data-driven approach and novel architecture successfully address the artifact issues prevalent in previous SR techniques.
• This method represents a significant advancement in achieving high-quality image super-resolution through learned representations.