A linear image filter and the corresponding filtering method is disclosed. In conventional linear image filters, only one pixel can be processed during a single cycle because it is applied to an odd-order filter matrix. Consequently, the present linear image filter and the corresponding method expand the original odd-order filter matrix into an even-order matrix by adding a row and a column of zeros at the topmost/bottommost and the rightmost/leftmost of the original filter matrix. This not only increases the filtering speed of the linear image filter because four pixels can be simultaneously reconstructed, but also fully utilizes current 32-bit processors because a pixel is currently represented only as an 8-bit value.