In this study, the authors propose a soft-switching filter to improve the performance of recent colour image smoothing filters when processing homogeneous image regions. The authors use a recent filter mixed with the classical arithmetic mean filter (AMF). The recent method is used to process image pixels close to edges, texture and details and the AMF is only used to process homogeneous regions. To this end, the authors propose a method based on the graph theory to distinguish image details and homogeneous regions and to perform a soft switching between the two filters. Experimental results show that the proposed method provides improved results which supports the appropriateness of the graph theory-based method and suggests that the same structure can be used to improve the performance of other non-linear colour image smoothing methods.