Noise Remove With A Notch Filter. The following figures demonstrate an FFT-based analysis of image noise patterns and use of a notch filter to remove the noise. The raw image shows the US submarine Greeneville with scars after a collision with a Japanese fishing research vessel. The image is the red band captured from the video originally taken by KHNL and published in internet news media. Significant periodic noise pattern appears across the image in all three RGB bands. The source of the noise is not clear. After FFT, the frequency domain image shows concentration of the noise pattern as two symmetric clusters. The centers of the two cluster are (173,261) and (173, 251) and the shape of the clusters is a rectangle segment of 4 x 32. A notch filter is applied to the FFT image to take out the clusters of the noise and the resulting image is converted back to the spatial domain with IFFT. In the filtered image, the noise is completely depressed and the image quality is greatly improved. The notch filtering process can be done with a single XBit command: rimage gfftran -in inputImage -out outImage -op XY -fp notch,4,32,4,1,,173,261
|