Research On The Super-resolution Reconstruction Algorithm Of Infrared Microscope Image

Infrared microscopy imaging system, which is widely used in the fields of industrialinspection, analytical chemistry, medical diagnosis and life science research, is capable ofobserving the details of temperature distribution of objects. Compared to the visible imagesensors such as CCD and CMOS, the size of pixels in the infrared detector is above20μmand the number of pixels is limited. Consequently, the spatial resolution of the images isrelatively low and the problems of under-sampling and aliasing may occur. Besides, theimages are blurred with rough edges due to the thermal diffusion effect. Despite theinfrared detector technology is growing, the drawbacks mentioned above can’t meet theneed of practical applications in the short term. The method, which utilizes thesuper-resolution image reconstruction technology to get high-resolution images based onthe existing hardware and micro-scanning system, has been one of the important ways todevelop the high-resolution infrared microscopy imaging system. Orienting therequirements of practical applications, this paper focused on the super-resolutionreconstruction algorithms under the conditions of controlled and uncontrolleddisplacements of images. Based on the micro-scanning infrared microscopy imagingsystem we developed, the simulated images and infrared microscope images were used todo experiments and the quality of reconstructed images was analyzed.Under the conditions of controlled displacements of images, the boundary backwardreconstruction algorithm was implemented and verified. The experiments of simulatedimages showed that the algorithm was able to remove the noise strip introduced by theapproximate boundary completely and reconstruct the original high-resolution image withhigh quality. The experiments of infrared microscope images showed that it had strictlimitations and was difficult to apply in practice because the practical imaging model oflow-resolution images couldn’t satisfy its precondition and there was a certain degree ofdisplacement error caused by the micro-scanning system which had serious impact on thequality of reconstructed images.Under the conditions of uncontrolled displacements of images, three super-resolutionreconstruction algorithms including Iterative Back Projection (IBP), improved IBP and the algorithm based on the set theory were implemented and verified. To further improve theaccuracy of image registration, an improved algorithm was proposed and implemented. Theimage registration experiments showed that when there were sub-pixel displacements, norotation or small rotation angles between the low-resolution images, the accuracy ofdisplacement estimation of the algorithm was higher than the Keren’s and Vandewalle’salgorithm. The experiments of infrared microscope images showed that the highestresolution of the original images was55.68μm, while the highest resolution of thereconstructed images reached39.37μm by three super-resolution algorithms employed inthis paper.In conclusion, the experiments showed that the super-resolution reconstructionalgorithms this paper utilized improved the image resolution significantly and made itpossible to observe targets whose size was above40μm. Furthermore, the quality ofreconstructed images met the need of biomedical detection and research.