Superpixels using random walker

We present a novel image superpixel approach using random walker in this paper. Our method begins with downsampling the input image with a sampling factor, and then we initialize the seed positions and run the random walker algorithm on the low-resolution image to reduce the computational complexity. Finally, we upsample the probabilities to obtain the superpixels boundaries of the full resolution image. The experimental results show that the superpixels by our method achieve the better performance than the result by the previous superpixel segmentation method.