Graph-Based Semi-Supervised Deep Image Clustering With Adaptive Adjacency Matrix.

Area of Science:

• Machine Learning
• Computer Vision
• Deep Learning

Background:

• Current graph-based semi-supervised deep clustering methods overlook shallow-level features.
• Feature extraction networks often use odd convolutional kernels, leading to uneven receptive field intensity.
• Fixed, precomputed adjacency matrices limit adaptability to evolving sample relationships.

Purpose of the Study:

• To propose a novel graph-based semi-supervised deep clustering method for image clustering.
• To address limitations in feature extraction and adaptability of existing methods.
• To improve clustering accuracy and efficiency.

Main Methods:

• Utilized a parity cross-convolutional feature extraction and fusion module for high-quality feature extraction.
• Incorporated a clustering constraint layer to enhance clustering efficiency.
• Customized an output layer for unsupervised regularization training and inferred an adaptive adjacency matrix via network prediction.

Main Results:

• The proposed method significantly outperforms state-of-the-art approaches on USPS, MNIST, SVHN, and FMNIST datasets.
• Achieved superior performance in terms of Accuracy (ACC), Normalized Mutual Information (NMI), and Adjusted Rand Index (ARI).

Conclusions:

• The novel graph-based semi-supervised deep clustering method effectively addresses limitations of existing techniques.
• The approach demonstrates superior performance and adaptability in image clustering tasks.
• The method offers a promising advancement for machine learning and computer vision applications.