A Visual Tensor Data Completion Method Based on Truncated Kernel Norm

Abstract

The present invention provides a visual tensor data completion method based on a truncated kernel norm, comprising the following steps: Step S1, storing a tensor to be completed as a target tensor in the form of a three-dimensional tensor and according to the tensor to be completed Initialize the target tensor step S2, input the target tensor into the tensor completion model, and use the gradient descent method to solve the tensor completion model, and output the solution of the target tensor; the tensor completion model is based on thin t-SVD ( Tensor Truncated Kernel Norm Model for Tensor Singular Value Decomposition). Step S3: Convert the solution of the target tensor into a format corresponding to the data source. Tensor data completion is efficient, the algorithm is fast, and the quality of the completion results is high.

Description

technical field [0001] The present invention relates to the technical field of visual tensor data completion, in particular to a visual tensor data completion method based on truncation kernel norm. Background technique [0002] In computer vision research, many visual data such as color images, video sequences, etc. often have very complex high-order data structures. Traditional data representations such as vectors and matrices cannot reflect these multi-dimensional data structure information well, and tensors are used as a generalization of higher-order (orders greater than or equal to 3) represented by vectors (first-order) and matrices (second-order) , which can better reflect the internal structure of multi-dimensional data such as images and videos. [0003] In the actual process of data acquisition, some elements in the tensor data will be lost due to encoding, transmission, and data conversion. A processing method of tensor completion is low-rank tensor completion....

AI Technical Summary

Problems solved by technology

This patent better retains the high correlation between each data element and surrounding elements; but the quality of the decomposition result directly affects the final completion effect, such as overlapping tensors in practical applications cannot correctly estimate the final completion result , such as the completion of color images, will cause problems such as blurring of image edge details; the model of this method is the sum of trace norms of all decomposed small tensors, and the model itself cannot well approximate the rank of the original tensor, resulting in The final completion effect becomes worse

Images