Image classification and recognition method based on kernel learning and dictionary learning of data dependence

Abstract

The invention discloses an image classification and recognition method based on kernel learning and dictionary learning of data dependence. Kernel learning is applied to a Riemannian manifold dictionary learning and sparse coding method based on a kernel method; three variables of kernel parameter, dictionary learning and sparse coding are optimized at the same time, the optimal kernel parameter and dictionary are learned, a better coding result can be obtained by using the learned kernel parameter and dictionary, the classification and recognition effects of the image are improved, and the method has good robustness for the image quality problem. The invention further provides an image classification and recognition device, a terminal and a storage medium.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to an image classification and recognition method based on data-dependent kernel learning and dictionary learning. Background technique [0002] The so-called dictionary learning and sparse coding are to learn several codebooks, and use these codebooks to perform sparse coding on feature data. In the past, most feature data were data in Euclidean space. However, in recent years, with the increasingly widespread and in-depth application of machine learning, many feature data are non-Euclidean data. After these non-Euclidean data are endowed with a certain topology and Riemannian measure, they can form a Riemannian manifold. Although the Riemannian manifold is a metric space, it is not a linear space, and dictionary learning and sparse coding cannot be performed directly on the Riemannian manifold. [0003] The linear combination of SPD manifold data limits the combination ...

AI Technical Summary

Problems solved by technology

Kernel functions commonly used on SPD manifolds include Stein kernel function, Jeff kernel function, LE kernel function, and kernel function based on geodesic distance. Since the existing SPD manifold has a single form of kernel function, it is used in machine learning algorithms. The kernel function is basically fixed, so the RKHS mapped by the kernel function is also fixed. This kind of processing cannot get a suitable RKHS according to the training samples, and the algorithm model obtained by training is not optimal.

Images