Compressed sensing is an arisen and significant theory, which has been widely used in infrared image reconstruction and many methods based on compressed sensing have been proposed. However, the existing methods can hardly accurately reconstruct infrared images. Considering that the sparsity of an infrared image plays a crucial role in compressed sensing to accurately reconstruct image, this paper presents a new sparse representation (MBFSF) that integrates the multiscale bilateral filter with shearing filter to overcome the above disadvantage. Firstly, one approximation subband image and a series of detail subband images at different scales and directions are obtained by the MBFSF. Then, in view of the feature that the most information is preserved in the approximation subband image, the proposed method only measures the detail subband images and preserves the approximation subband image. Subsequently, a very sparse random measurement matrix is used for the measurement at the detail subband images to reduce the computation cost and storage of large random measurement matrices in compressed sensing. Finally, an accelerated iterative hard thresholding algorithm is employed to reconstruct the infrared image. Experimental results show that the proposed method has superior performance in terms of reconstruction accuracy and compares favorably with existing compressed sensing methods, which is an effective method in high-resolution infrared imaging based on compressed sensing.
Due to the limitations of the manufacturing technology, the response rates to the same infrared radiation intensity in each infrared detector unit are not identical. As a result, the non-uniformity of infrared focal plane array, also known as fixed pattern noise (FPN), is generated. To solve this problem, correcting the non-uniformity in infrared image is a promising approach, and many non-uniformity correction (NUC) methods have been proposed. However, they have some defects such as slow convergence, ghosting and scene degradation. To overcome these defects, a novel non-uniformity correction method based on locally adaptive regression filter is proposed. First, locally adaptive regression method is used to separate the infrared image into base layer containing main scene information and the detail layer containing detailed scene with FPN. Then, the detail layer sequence is filtered by non-linear temporal filter to obtain the non-uniformity. Finally, the high quality infrared image is obtained by subtracting non-uniformity component from original image. The experimental results show that the proposed method can significantly eliminate the ghosting and the scene degradation. The results of correction are superior to the THPF-NUC and NN-NUC in the aspects of subjective visual and objective evaluation index.