Improve the efficiency of handcrafted features in image retrieval by adding selected feature generating layers of deep convolutional neural networks

Shamsipour, Ghazal; Fekri-Ershad, Shervan; Sharifi, Mahdi; Alaei, Alireza

doi:10.1007/s11760-023-02934-z

Cited by 28 publications

(3 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…During model training, considering that latent features typically diminish after repeated convolutions, ResNet-50 is employed to counteract this effect. Unlike conventional setups, ResNet-50 continuously incorporates previous latent features during the backward training process [40,42] thereby enhancing the global representation of features.…”

Section: Appearance Embeddingmentioning

confidence: 99%

Res-NeuS: Deep Residuals and Neural Implicit Surface Learning for Multi-View Reconstruction

Wang,

Gao,

Shen

2024

Sensors

View full text Add to dashboard Cite

Surface reconstruction using neural networks has proven effective in reconstructing dense 3D surfaces through image-based neural rendering. Nevertheless, current methods are challenging when dealing with the intricate details of large-scale scenes. The high-fidelity reconstruction performance of neural rendering is constrained by the view sparsity and structural complexity of such scenes. In this paper, we present Res-NeuS, a method combining ResNet-50 and neural surface rendering for dense 3D reconstruction. Specifically, we present appearance embeddings: ResNet-50 is used to extract the appearance depth features of an image to further capture more scene details. We interpolate points near the surface and optimize their weights for the accurate localization of 3D surfaces. We introduce photometric consistency and geometric constraints to optimize 3D surfaces and eliminate geometric ambiguity existing in current methods. Finally, we design a 3D geometry automatic sampling to filter out uninteresting areas and reconstruct complex surface details in a coarse-to-fine manner. Comprehensive experiments demonstrate Res-NeuS’s superior capability in the reconstruction of 3D surfaces in complex, large-scale scenes, and the harmful distance of the reconstructed 3D model is 0.4 times that of general neural rendering 3D reconstruction methods and 0.6 times that of traditional 3D reconstruction methods.

show abstract

Section: Appearance Embeddingmentioning

confidence: 99%

Res-NeuS: Deep Residuals and Neural Implicit Surface Learning for Multi-View Reconstruction

Wang,

Gao,

Shen

2024

Sensors

View full text Add to dashboard Cite

show abstract

“…This approach was applied recently in [22] but only to 1D spectral domain representations (i.e., application of the FFT) and not to time-frequency transforms. In other fields, the use of a hybrid approach, which combines handcrafted features and DL to improve computing efficiency and classification performance, has recently been proposed for generic image processing in [23], though with different features than in this study.…”

Section: Related Workmentioning

confidence: 99%

LoRa Radio Frequency Fingerprinting with Residual of Variational Mode Decomposition and Hybrid Machine-Learning/Deep-Learning Optimization

Baldini,

Bonavitacola

2024

Electronics

View full text Add to dashboard Cite

RFF refers to the technique for identifying and classifying wireless devices on the basis of their physical characteristics, which appear in the digital signal transmitted in space. Small differences in the radio frequency front-end of the wireless devices are generated across the same wireless device model during the implementation and manufacturing process. These differences create small variations in the transmitted signal, even if the wireless device is still compliant with the wireless standard. By using data analysis and machine-learning algorithms, it is possible to classify different electronic devices on the basis of these variations. This technique has been well proven in the literature, but research is continuing to improve the classification performance, robustness to noise, and computing efficiency. Recently, DL has been applied to RFF with considerable success. In particular, the combination of time-frequency representations and CNN has been particularly effective, but this comes at a great computational cost because of the size of the time-frequency representation and the computing time of CNN. This problem is particularly challenging for wireless standards, where the data to be analyzed is extensive (e.g., long preambles) as in the case of the LoRa (Long Range) wireless standard. This paper proposes a novel approach where two pre-processing steps are adopted to (1) improve the classification performance and (2) to decrease the computing time. The steps are based on the application of VMD where (in opposition to the known literature) the residual of the VMD application is used instead of the extracted modes. The concept is to remove the modes, which are common among the LoRa devices, and keep with the residuals the unique intrinsic features, which are related to the fingerprints. Then, the spectrogram is applied to the residual component. Even after this step, the computing complexity of applying CNN to the spectrogram is high. This paper proposes a novel step where only segments of the spectrogram are used as input to CNN. The segments are selected using a machine-learning approach applied to the features extracted from the spectrogram using the LBP. The approach is applied to a recent LoRa radio frequency fingerprinting public data set, where it is shown to significantly outperform the baseline approach based on the full use of the spectrogram of the original signal in terms of both classification performance and computing complexity.

show abstract

“…In these last years, Convolutional Neural Network (CNN) [ 7 , 8 ] models have shown significant performance improvement in several fields as Natural Language Processing (NLP) [ 9 ] and computer vision [ 10 ]. Given their success in such fields, it seems to be efficient for image retrieval.…”

Section: Introductionmentioning

confidence: 99%

Enhancing Medical Image Retrieval with UMLS-Integrated CNN-Based Text Indexing

Gasmi,

Ayadi,

Torjmen

2024

Diagnostics

View full text Add to dashboard Cite

In recent years, Convolutional Neural Network (CNN) models have demonstrated notable advancements in various domains such as image classification and Natural Language Processing (NLP). Despite their success in image classification tasks, their potential impact on medical image retrieval, particularly in text-based medical image retrieval (TBMIR) tasks, has not yet been fully realized. This could be attributed to the complexity of the ranking process, as there is ambiguity in treating TBMIR as an image retrieval task rather than a traditional information retrieval or NLP task. To address this gap, our paper proposes a novel approach to re-ranking medical images using a Deep Matching Model (DMM) and Medical-Dependent Features (MDF). These features incorporate categorical attributes such as medical terminologies and imaging modalities. Specifically, our DMM aims to generate effective representations for query and image metadata using a personalized CNN, facilitating matching between these representations. By using MDF, a semantic similarity matrix based on Unified Medical Language System (UMLS) meta-thesaurus, and a set of personalized filters taking into account some ranking features, our deep matching model can effectively consider the TBMIR task as an image retrieval task, as previously mentioned. To evaluate our approach, we performed experiments on the medical ImageCLEF datasets from 2009 to 2012. The experimental results show that the proposed model significantly enhances image retrieval performance compared to the baseline and state-of-the-art approaches.

show abstract

Improve the efficiency of handcrafted features in image retrieval by adding selected feature generating layers of deep convolutional neural networks

Cited by 28 publications

References 37 publications

Res-NeuS: Deep Residuals and Neural Implicit Surface Learning for Multi-View Reconstruction

Res-NeuS: Deep Residuals and Neural Implicit Surface Learning for Multi-View Reconstruction

LoRa Radio Frequency Fingerprinting with Residual of Variational Mode Decomposition and Hybrid Machine-Learning/Deep-Learning Optimization

Enhancing Medical Image Retrieval with UMLS-Integrated CNN-Based Text Indexing

Contact Info

Product

Resources

About