An adaptive face image clustering method based on spectral clustering and reinforcement learning

Abstract

An adaptive face image clustering method based on spectral clustering and reinforcement learning, comprising the following steps: 1) Applying deep convolutional neural network DCNN to realize the feature representation of all face images in the database; 2) Applying spectral clustering The algorithm realizes the clustering of image representation; 3) Based on the reinforcement learning algorithm, the optimal cluster number and spectral clustering dimensionality reduction parameters are adaptively searched. First, given the desired cluster search range, set the search step size and the search starting point The starting point; then execute the Q-Learning algorithm, try all possible parameter tuning behaviors within the search range, and choose the behavior that makes the best clustering performance to give the largest positive return; until all behaviors cannot make performance optimization, or meet the maximum iteration number of times to end the search; the end point of the search after multiple trainings is the set value of the optimal parameter. Through dynamic parameter adjustment and self-adaptive search, the present invention can help to find the global optimal clustering number and dimensionality reduction, and improve the face image clustering performance.

Description

technical field [0001] The invention relates to an adaptive face image clustering method, in particular to an adaptive face image clustering method based on spectral clustering and reinforcement learning. Background technique [0002] With the rapid development of computer vision and pattern recognition technology, image, as the most common visual information presentation mode, has broad application prospects. In the era of "big data", a large number of pictures are generated every day. For example, on social media, according to Facebook, an average of 350 million images are generated every day, most of which are images of faces. In judicial investigations, there are still a huge number of pictures that urgently need to be identified and classified. In terms of social security maintenance and monitoring management, a large number of face images captured by cameras need to be authenticated and stored for comparison. However, these face images usually do not have identity t...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the disadvantages of the poor clustering performance of existing face image processing methods, and to significantly improve the face image clustering performance, the present invention provides an adaptive face image clustering method based on spectral clustering and reinforcement learning, Combined with spectral clustering model and reinforcement learning algorithm, through dynamic parameter adjustment and adaptive search, it can help to find the globally optimal number of clusters and dimensionality reduction

Images