An auto-adapted weight filtering method is put forward. The validity of the pixel is analyzed with a Rank-order Absolute Difference (ROAD) method

if the pixel is noise

the pixel will be reconstructed by the neighbor pixels with different weights

and the weight of the neighbor pixel is correlative with its validity. At last

not only the impulse noise is restrained

but also the detail of the image is reserved. Experiments show that when using the auto-adapted weight filtering to get rid of the impulse noise

an evident improvement is found compared with the general method

both in visual image quality and quantitative measures of signal restoration. When the impulse noise occupies 35% in image

the effect of data recovery by this method can be increased by 10 dB compared with conventional median filter.