Multiobjective Image Compression based on Tensor Decomposition

Most of the traditional image compression methods vectorize the data contained in the image and then compress it. However, this approach does not take into account the high-dimensional information inside the image. To solve this problem, this paper regards the color image as a third-order tensor, and proposes a method of color image compression based on Tucker decomposition and multiobjective optimization. The tensor size compression ratio and Hu invariant moment similarity are proposed to measure the image compression quality. And to more comprehensively consider the sensitivity of human visual system to different visual signals, the five-objective optimization model of image compression is constructed. The five-objective optimization model includes: the above two indexes, information content weighted structure similarity index, color image feature similarity and information fidelity criterion. In addition, an angle-aware opposition-based learning strategy is proposed to improve the reference vector guided selection strategy of RVEA*. In the experiments, this method could effectively solve the problem of color image compression.

Automated summary

This paper proposes a color image compression method based on Tucker decomposition and multi-objective optimization. The color image is regarded as a third-order tensor, and the method compresses the image by decomposing and optimizing the tensor. The evaluation methods for image compression quality include tensor size compression ratio and Hu invariant moment similarity. In addition, a five-objective optimization model is constructed, which considers multiple aspects of image compression. Experimental results show that the proposed method effectively solves the problem of color image compression.