A Novel Super-Resolution Algorithm for Hyperspectral Images
In 2012, we developed a novel algorithm, which fuses a high resolution color image with a low resolution hyperspectral image to yield a high resolution hyperspectral image. AVIRIS data (15 m resolution) was used in our experiments. The AVIRIS data has 213 bands with wavelengths range from 380 nm to 2500 nm. The image is downsampled to 60 m resolution. In the experiment, we pick only R, G, B bands from original high resolution hyperspectral image for fusion. The bicubic method in the following plots is implemented by upsampling the low resolution image using bicubic interpolation. The results of bicubic method are used as baseline for comparison study. To demonstrate our algorithm, we performed material classification studies using bicubic and our approach. From the figure below, it can be seen that the material classification using our super-resolution image gave results very close to the ground truth whereas bicubic missed a lot of the fine details in material distribution.
Classification Map based on Bicubic Interpolation Image
Classification Map based on Our Super Resolution Image
Classification Map based on Our Ground Truth Image
Comparison of material classification results. The dotted circled area was missing in the bicubic results whereas our results can recover the line inside the dotted circle.