An Adaptive Image Frequency Subbanding Method for High-Frequency Pixel Estimation

In this paper, an adaptive image subbanding method is proposed for high-frequency pixel estimation. The similarity relation between a pixel and its top, down, right and left pixels is exploited to derive adaptive thresholds for classification of different high-frequency bands. Then, high-frequency pixel detection can be applied to estimate impulse-noise ratio in order to select the appropriate working windows for various high-frequency bands. In theory, the appropriate working window used for each band will substantially improve PSNR performance of noise-filtering. Experimental results show that PSNR performance of filtering on the subbanded image is significantly superior to that of non-subbanded image, for improvement of ranging from 0.5 dB to 5 dB in various corruption rates.