2005 年 60 巻 6 号 p. 1197-1200
We present a practical multidimensional framework for obtaining a super-resolution image from a subpixel shifted sequence through the reduction of noise followed by a recursive back-projection algorithm.
The noise-reduction stage, for thermal patterns, uses the multi-observation for identifying the fixed pattern noise. As an alternative, we present a neighborhood classification method that identifies the type and degree of the fixed noise that is subtracted from the image. In the case of radar and ultrasonic images, the multiplicative speckle noise is reduced with a filter that adaptively calculates the size of the filtering window and performs a median filter only considering the pixels with similar values inside the window.
The super-resolution technique performs the downsampling of the pre-processed image sequences followed by a back-projection algorithm. Different evaluation methods demonstrate the accuracy and quality of the results. Finally, we propose a multireflecting system adapted to the camera, provided with a rotational scanning as a hardware solution for the data acquisition stage.
