Image Compression Based on Truncated HOSVD

Abstract

The paper presents a novel image compression technique. Inspired by High Order Singular Value Decomposition (HOSVD), the technique respects each image both gray image and color image as a high order tensor and discards singular sub-tensor corresponding smaller singular value to implement image compression. The paper also proves that HOSVD is naturally the extension of matrix SVD. The comparison experiments show that the image compression technique based on truncated HOSVD has better performance than JPEG.