Abstract
Data fusion has always been a hot research topic in human-centric computing and extended with the development of artificial intelligence. Generally, the coupled data fusion algorithm usually utilizes the information from one data set to improve the estimation accuracy and explain related latent variables of other coupled datasets. This paper proposes several kinds of coupled images decomposition algorithms based on the coupled matrix and tensor factorization-optimization (CMTF-OPT) algorithm and the flexible coupling algorithm, which are termed the coupled images factorization-optimization (CIF-OPT) algorithm and the modified flexible coupling algorithm respectively. The theory and experiments show that the effect of the CIF-OPT algorithm is robust under the influence of different noises. Particularly, the CIF-OPT algorithm can accurately restore an image with missing some data elements. Moreover, the flexible coupling model has better estimation performance than a hard coupling. For high-dimensional images, this paper adopts the compressed data decomposition algorithm that not only works better than uncoupled ALS algorithm as the image noise level increases, but saves time and cost compared to the uncompressed algorithm.
Highlights
Image data fusion has been a hot research topic in neuroscience, metabonomics and other fields, and has been widely used in real life
We can see that the compression decomposition algorithm obviously saves time and cost compared to the uncompressed algorithm which may be rather slow
For the proposed coupled images factorizationoptimization (CIF-OPT) algorithm, the corresponding experiments show that the effect of the coupled image decomposition under the influence of different noise is robust, and the fusion effect is better than the CMTF-OPT algorithm, which shows that the coupled images decomposition algorithm is feasible
Summary
Image data fusion has been a hot research topic in neuroscience, metabonomics and other fields, and has been widely used in real life. The coupled data fusion algorithm usually utilizes the information of one data set to improve the estimation accuracy and the interpretation of related potential variables of other data sets. With the development of electronic and imaging technology, it is difficult to find accurate data for digital images for human beings, such as medical science [1], information remote sensing and so on. In this situation, people hope to primitively analyze mass images and select the information quickly and effectively by more convenient calculation way. We apply the tensor structures to represent massive data to solve above problems in this paper because of its multi-dimensional property
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have