Super-Resolution Reconstruction From Multiple Defocused Infrared Images of Stationary Scene

Infrared image has poor visual effect for its low resolution. Super-resolution reconstruction (SRR) is an effective means to address this problem. Existing SRR algorithms use well-focused images and ignore the value of defocused images generated by the infrared imaging system during focusing. The basic idea of the present study is to treat a defocused infrared image as distribution and accumulation of scene information among different pixels of the infrared detector, as well as a valid observation of the imaged subject; defocused images are the result of blurring a corresponding high resolution (HR) image using a point spread function (PSF) followed by downsampling. From this idea, we used multiple defocused images to build an observation model for HR images and propose an SRR algorithm to approach the HR images. We have developed an image degradation model by analyzing optical lens imaging, using the particle swarm optimization algorithm to estimate the PSF of the HR image, and using compressed sensing theory to implement SRR based on the noncoherent characteristics of the defocused infrared images. Experiments demonstrate that our method can be used to obtain more information about details of a scene and improve the visual effect without adding any hardware facilities, improving the recognition and interpretation of the image subject.