Infrared image super resolution with structure prior from uncooled infrared readout circuit

Area of Science:

• Computer Vision
• Image Processing
• Infrared Technology

Background:

• Infrared images offer valuable temperature data but are limited by low resolution in uncooled detector arrays.
• Existing infrared image super-resolution (SR) methods face challenges in efficiency and performance.
• The need for high-resolution infrared imaging drives the development of advanced SR algorithms.

Purpose of the Study:

• To develop an efficient infrared image super-resolution (SR) algorithm for uncooled infrared detector arrays.
• To improve the reconstruction details and capture spatio-temporal correlations in infrared images.
• To achieve a balance between performance and computational efficiency in infrared SR.

Main Methods:

• Proposed an efficient Row-Column Transformer Block (RCTB) inspired by Uncooled Infrared Focal Plane Array (IRFPA) readout circuits.
• Developed a Compact Convolution Block (CCB) with a U-shape Spatial Channel Attention Block (USCAB) for local feature extraction.
• Incorporated a regularization control term in the loss function to enhance reconstruction details.
• Introduced a hybrid network, Efficient Infrared Image Super-Resolution (EIRSR), and its variants (EIRSRs).

Main Results:

• The proposed EIRSR network achieved a superior balance between performance and efficiency (parameters and computational cost).
• EIRSR demonstrated competitive performance compared to existing state-of-the-art methods.
• The model variants (EIRSRs) offer scalable solutions for infrared image super-resolution.

Conclusions:

• The novel RCTB and CCB effectively capture spatio-temporal correlations and local features for infrared SR.
• EIRSR provides an efficient and effective solution for enhancing the resolution of infrared images.
• The proposed method contributes to advancing the capabilities of uncooled infrared imaging systems.