Written in the course Computational Intelligence Lab at ETH Zürich

Authors: David Itten, Martin Lanter, Daniel Schweizer

Abstract: We present ScaleCompress, a three-step image compression algorithm. It first scales the input image to a smaller size, then reduces the number of bits representing a pixel, and additionally uses run-length encoding to further reduce the size of the compressed image. The evaluation shows that ScaleCompress performs clearly better than our PCA and SVD-based solutions from the assignments. ScaleCompress is very efficient for images with patches of similar colors or color gradients (e.g. photographs of landscapes or people), and performs sufficient on images with lots of fine-grained details (e.g. texture images).