Fusion kernel-based linear nucleation feature space grouping modeling method

Abstract

The invention provides a fusion kernel-based linear nucleation feature space grouping modeling method, which comprises the following steps of: giving a training sample, and sampling a plurality of sample signals from the training sample according to a predetermined probability distribution function to form a reduction matrix; determining a fusion kernel function of Euclidean and cosine distance measurement, combining the fusion kernel function with the method, performing kernel matrix approximate construction on the training sample and the reduced matrix, and decomposing the kernel matrix by using rank k characteristics to obtain a virtual sample; decomposing the virtual sample into a virtual training sample and a virtual test sample, constructing a prediction model based on elastic network regularization, and training and testing the prediction model on the virtual training sample and the virtual test sample by using an elastic network regularization learning method to obtain a finalprediction model. By implementing the method, the defects in the prior art can be overcome, the extracted features can better reconcile the contradiction between the sample scale and the feature dimension, and the problem of small-scale samples is avoided, so that the performance is improved.

Description

technical field [0001] The invention relates to the technical fields of face recognition and computer technology, in particular to a modeling method for grouping of linear kernelized feature spaces based on fusion kernels. Background technique [0002] Face recognition refers to the problem of matching human images from different camera perspectives in a multi-camera system. It provides critical assistance in the analysis of different aspects such as person identity and behavior, and has developed into a key component in the field of intelligent video surveillance. [0003] At present, the main methods of face recognition can be divided into the following two categories: 1) face recognition methods based on feature representation; 2) face recognition methods based on metric learning. [0004] Among the face recognition methods based on feature representation, there are methods based on the most commonly used underlying visual features and methods based on semantic attribute...

AI Technical Summary

Problems solved by technology

However, this method is based on the theoretical assumption that the sample distribution obeys the Gaussian distribution, but in reality, the samples not only do not obey the Gaussian distribution perfectly, but may even deviate seriously, resulting in performance degradation

In addition, in actual situations, the sample size is often much smaller than the feature dimension, which makes the calculation of the Mahalanobis distance in metric learning difficult or even unsolvable.

Images