The Research Of Super-resolution Reconstruction Algorithm For Infrared Image

In recent years, with the rapid development of infrared technology, the price of infrared devices keeps reducing.Infrared image has applicated in civil, industrial and military fields widely. On one hand, the characteristics of the thermal infrared imager and some of the inherent flaws of the thermal infrared imager make the quality of infrared image not high. On the other hand, the demand for high resolution infrared images is growing. Therefore, the algorithm of adaptive steering kernel regression based on structure tensor and adaptive regression order as well as the algorithm based on self-similarity and the selection of adaptive sparse dictionary is used to complete the super resolution reconstruction of the single image in this paper.Kernel function is used as a weight to estimate the unknown pixels with weighting and averaging the pixel values of the local block in classical kernel regression.And it's effect is better than bicubic interpolation algorithm.Adaptive steering kernel regression is better than the classic kernel regression algorithm by introducing gray value and structure of the information can not be effectively extracted with the adaptive steering kernel regression algorithm, and the fixed regression order is not suitable to describe the structural characteristics of different image blocks. The reconstruction algorithm of adaptive steering kernel regression based on structure tensor and adaptive regression order in this paper,the structure tensor is used to describe the structure information of the image with the good job in isotropic area, and the order is chosen by the structural characteristics. So the two questions can be solved well.The reconstruction learning algorithm is a prevailing research direction in recent years.Generally, super resolution reconstruction algorithms based on sparse representation are using a common global over-complete dictionary for image reconstruction. However,a lot of high-resolution image library is required to do dictionary training in this method. It not only costs a long time, but also is not optimal for the image block of different characteristics. The reconstruction algorithm based on self-similarity and the selection of adaptive sparse dictionary in this paper, the high frequency characteristic information which in high resolution visible gray image is richer than that in infrared image, and the priorinformation of multi-scale self-similarity is used to reconstruct. It. effectively reduces the training time, meanwhile the double sparse dictionary is adaptively chosen by the structural characteristics of the image. It has achieved good results. Finally, a simulation is used to demonstrate it.