Abstract:Identifying similar images within databases is a significant challenge in the field of image retrieval. This challenge is further compounded by escalating demands for heightened precision and speed, propelled by advancements in information technology. This study introduces a distinctive network design to address common problems associated with high-probability image retrieval, such as sluggish retrieval speeds and inadequate discrimination. The network architecture primarily employs CNNs for feature extraction. To maximize the advantages of deep learning in feature extraction, the design includes a second-order attention module and a second-order similarity loss. The incorporation of second-order information enhances the correlation among local spatial locations in the image, adjusting each local contribution to the global context. This facilitates a more comprehensive understanding of the image's structure and content, thereby improving feature extraction. Furthermore, dimension reduction techniques are applied to effectively capture the profound information within images. This aids in the elimination of superfluous features while preserving essential information necessary for the retrieval task. This method improves the model's retrieval accuracy by concentrating on the image's deep features, allowing it to disregard unnecessary information. Experimental results demonstrate that our image retrieval technique not only improves accuracy but also significantly accelerates retrieval speed across various datasets, exceeding previous benchmarks in the field.

High Dimensionality Reduction Using Cur Matrix Decomposition And Auto-Encoder For Web Image Classification

Computational and Theoretical Analysis of Supervised Dimensionality Reduction

Image retrieval using dimensionality reduction

Auto-Encoder Based Dimensionality Reduction

Research on Image Classification Method of Features of Combinatorial Convolution

Representation Learning Via a Semi-Supervised Stacked Distance Autoencoder for Image Classification

Image Classification by Feature Dimension Reduction and Graph based Ranking

Dimensionality Reduction Strategy Based on Auto-Encoder

Supervised Dimensionality Reduction and Image Classification Utilizing Convolutional Autoencoders

Dimensionality Reduction With Category Information Fusion And Non-Negative Matrix Factorization For Text Categorization

A Hybrid Autoencoder Framework of Dimensionality Reduction for Brain-Computer Interface Decoding

A Dimensional Reduction Guiding Deep Learning Architecture for 3D Shape Retrieval

Image retrieval method based on CNN and dimension reduction

Deep Autoencoders for Dimensionality Reduction of High-Content Screening Data

Image retrieval based on dimensionality reduction of second-order information

Fusion of effective dimension reduction and discriminative dictionary learning for high-dimensional classification

Divergence Regulated Encoder Network for Joint Dimensionality Reduction and Classification

Sparse Clustering Algorithm Based on Multi-Domain Dimensionality Reduction Autoencoder

Transferred Dimensionality Reduction

A restorable autoencoder as a method for dimensionality reduction

Guided autoencoder for dimensionality reduction of pedestrian features