An adaptive image restoration technique is presented. This is an extended version of the spatially invariant image restoration technique using constrained deconvolution. The aim of the extension is to improve the characteristics of the original filter. The proposed filter partitions the image into three types of regions. For each region this filter has a different impulse response. The most distinctive feature of this filter is that the form of the impulse response varies according to the direction of the edges in the image. It becomes an ordinary inverse filter irrespective of each region, however, when additive noise does not exist in the degraded image. Simulation results show that the observation noise is suppressed, while the edges are preserved.