Introduction
A lot of application situation need to obtain high resolution video and image, such as astronomy, remote sensing, martial monitor and medical diagnosis (CT,MRI) etc. Due to imaging condition and imaging mode constrain the usual imaging system to obtain high resolution image. Therefore, Obtaining observation image require amplifying processing, Although people can directly obtain high resolution image by enhancing pixel number of the charge coupled device, this method only apply very limited situation. The main reason is high cost and physic difficult to achieve. Traditional single frame scaling algorithm like Nearest Neighbors and Bilinear Interpolation has realized amplifying for one Image. But this algorithm not providing more useful information and actually not improving the resolution of image.
Due to the continuous image frames including large number similar but not identical information, thus super-resolution image can be reconstructed according to image sequence. A group of low-resolution image sequence which has the same description of scene but each of containing different information is used to reconstruct high-resolution image. The technology is called super-resolution image reconstruction. In 1984 Tsa and Huang first presented super-resolution reconstruction base on the translation image sequence and given reconstruct method base on the frequency domain approximation. It resolved the problem of nonunique solution for the super-resolution image. So far variety algorithms include frequency, project on convex set (POCS), Maximum a Posteriori (MAP) etc have been presented. The study group of MDSP presented super-resolution reconstruction method based on the L1- norm and Bilateral Total Variation (BTV).
VIDEO DEMO
No comments:
Post a Comment