Image denoising with geometrical thresholds

Observation on the interpolation effect and selectivity of human visual perception can be used to construct a criterion to find and preserve geometrical details while denoising an image. An approach to denoise an image with such a criterion is introduced, through which the pointwise thresholding of wavelet coefficients is turned into a bandwise geometrical thresholding. Interestingly this method can make up for the loss of details in a pointwise thesholding procedure. Experimental results validate its good performance, particularly on images with dominant high frequency contents and faint directional details such as Barbara.