Tensor Approximation Algorithm based on High Order Singular Value Decomposition

Abstract

To solve some problems which JPEG compression obtains results of poor reconstruction quality and high computational complexity for image containing more high frequency information, a novel tensor approximation algorithm based on high order singular value decomposition has been proposed. The new algorithm respects each image both gray image and color image as a high order tensor. It transforms the image into singular value matrix that contains nonzero singular values to implement image compress and discards singular sub-tensor corresponding smaller singular values in tensor decomposition to reduce the calculation work. The experiment result compared with JPEG shows that the algorithm has better performance than JPEG for color images. It is easy to apply the algorithm to any high order tensor in computer vision.