Reproducing kernel hilbert space based single infrared image super resolution

• Infrared image are analyzed by reflective, cooled emissive and uncooled emissive.
• Model by spline based reproducing kernel hilbert space and heaviside function.
• The robust model is applicable to all infrared images.
• The small root mean-square-error of reconstructed high resolution images.

The spatial resolution of Infrared (IR) images is limited by lens optical diffraction, sensor array pitch size and pixel dimension. In this work, a robust model is proposed to reconstruct high resolution infrared image via a single low resolution sampling, where the image features are discussed and classified as reflective, cooled emissive and uncooled emissive based on infrared irradiation source. A spline based reproducing kernel hilbert space and approximative heaviside function are deployed to model smooth part and edge component of image respectively. By adjusting the parameters of heaviside function, the proposed model can enhance distinct part of images. The experimental results show that the model is applicable on both reflective and emissive low resolution infrared images to improve thermal contrast. The overall outcome produces a high resolution IR image, which makes IR camera better measurement accuracy and observes more details at long distance.