Integrated Image Enhancement and Artifacts Reduction Based on Frequency Bands Manipulation

The spectrum of an image can be split into a number of frequency bands by taking the differences of the low-pass filtered versions of the original image. The visual quality of an image can be improved by manipulating different frequency bands of the image. Amplifying certain frequency bands would enhance the sharpness/contrast of the image; and suppressing certain frequency bands would reduce noise/coding artifacts. In this paper, a least mean square (LMS) training algorithm is proposed to optimally manipulate the frequency bands of an image for simultaneously enhancing object details and reducing coding artifacts. Experimental results show that the proposed algorithm outperforms other classification-based LMS approaches substantially.