Abstract In this article, we propose an edge-directed switching median filter that considers the local correlation of pixels and edge directions for impulse noise reduction. The proposed algorithm consists of two main steps, detection and correction. In the first step, the impulse noise is detected using minimum and maximum values in a scalable mask. In the second step, a corrupted pixel is corrected using the local correlation between the uncorrupted pixels in the mask. This step specifically performs edge-directed filtering using principal component analysis to preserve the edge and detail information in the highly corrupted image. The experimental results showed that the proposed method can reduce impulse noise significantly and preserve more edge information than the existing state-of-the-art methods. In addition, the proposed method outperforms existing methods for highly corrupted images by an average of 6.47 dB. Therefore, we believe that the proposed method can be a useful tool for removing impulse noise in the field of infrared related devices and digital cameras.