Superpixels partition an image into homogeneous regions with regular shapes and adherence to object edges. A critical factor of superpixels generation is the distance measure of pixels. Euclidean distance is usually utilized to compute the spatial and color difference between pixels. As a pre-processing procedure, superpixels should provide image information as much as possible for further analysis. But image content is ignored as Euclidean distance can hardly deal with. In this paper, we focus on the distance computation of superpixel methods. We propose the adaptive k distance method for image content analysis. We design a computational method of adaptive k distance which is applied to images specially. Combining Euclidean distance and the proposed distance, we present a superpixel method which provides tunable parameters to compromise between superpixel compactness and image edge adherence. We evaluate our method on BSD500 dataset comparing with several state of the art superpixel methods. We discuss the experimental results qualitatively and quantitatively, and the referred evaluation criteria that have not been analyzed in detail. Experimental results show that our method achieves favorable performance against the state of the art.