Image denoising is a fundamental problem in image processing. The switching filtering is a popular approach to reduce the impulse noise. It faces two challenges including the impulse noise detection and filter design. The traditional detection methods based on single criterion or multiple criteria encounter uncertainty problems and produce many miss-detections and false alarms, especially when the image is severely corrupted. In this paper, the uncertainties encountered in the impulse noise detection are addressed using the theory of belief functions, and a multi-criteria detection strategy for the impulse noise based on evidential reasoning is proposed. Based on the pre-detection, an adaptive median filter is designed, which adaptively determines the size of the filtering window according to the estimated global noise density and the degree of local corruption. Experimental results and related analyses show that our proposed image denoising method for the impulse noise has superior performance compared with several state-of-the-art denoising methods.