Kambiz Rahbar
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Kambiz Rahbar has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
Fatemeh Taheri
Department of Computer Engineering, ST.C., Islamic Azad University, Tehran, Iran
9 shared publicationsZiaeddin Beheshtifard
Department of Computer Engineering, ST.C., Islamic Azad University, Tehran, Iran
5 shared publicationsMaryam Khademi
Department of Applied Mathematics, ST.C., Islamic Azad University, Tehran, Iran
3 shared publicationsMehrnaz Niazi
Pishtazan Higher Education Institute, Department of Computer Engineering, Shiraz, Iran
3 shared publicationsAlireza Nazari
Department of Computer Engineering, ST.C., Islamic Azad University, Tehran, Iran
2 shared publicationsMansour Sheikhan
Department of Electrical Engineering, South Tehran Branch, Islamic Azad University, Tehran, Iran
2 shared publications
Latest Publications
- 22 July 2026
- Published by Springer Nature in Signal, Image and Video Processing
- Vol. 20 (10) , 520
Object image retrieval based on self-supervised learning and zero-shot learning
- 22 July 2026
- Published by Springer Nature in Signal, Image and Video Processing
- Vol. 20 (10) , 520
Object image retrieval based on self-supervised learning and zero-shot learning
- Open Access
- 20 May 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Enhancing image retrieval via Siamese network-based hashing with gated residual connections
This paper presents a supervised hashing framework built on a Siamese architecture, where gated residual connections substantially enhance the quality of compressed representations. The model employs ConvNeXt-Base as the backbone, combining the inductive biases of convolutional networks with modern architectural principles inspired by transformers. In the hash-generation pathway, the learned representation must simultaneously preserve essential identity information and apply nonlinear transformations for effective compression and separation. To achieve this, a learnable gate is introduced between two complementary branches: (1) a shallow identity/residual branch that preserves the core feature structure extracted by the backbone, and (2) a deeper transformed branch that performs nonlinear projection and disentanglement. The adaptive gating mechanism dynamically balances these two paths, enabling the network to retain discriminative local and global cues while suppressing irrelevant variations. As a result, the proposed design produces stable, semantically consistent, and highly discriminative binary hash codes. The Siamese network is trained using Triplet Loss to enforce similarity preservation in Hamming space. Extensive experiments on fine-grained benchmarks (CUB-200-2011 and NABirds) as well as a coarse-grained dataset (CIFAR-10) demonstrate that the proposed framework consistently outperforms or matches state-of-the-art hashing methods, validating its robustness and generalization across different levels of semantic granularity.
- 20 May 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessEnhancing image retrieval via Siamese network-based hashing with gated residual connections
This paper presents a supervised hashing framework built on a Siamese architecture, where gated residual connections substantially enhance the quality of compressed representations. The model employs ConvNeXt-Base as the backbone, combining the inductive biases of convolutional networks with modern architectural principles inspired by transformers. In the hash-generation pathway, the learned representation must simultaneously preserve essential identity information and apply nonlinear transformations for effective compression and separation. To achieve this, a learnable gate is introduced between two complementary branches: (1) a shallow identity/residual branch that preserves the core feature structure extracted by the backbone, and (2) a deeper transformed branch that performs nonlinear projection and disentanglement. The adaptive gating mechanism dynamically balances these two paths, enabling the network to retain discriminative local and global cues while suppressing irrelevant variations. As a result, the proposed design produces stable, semantically consistent, and highly discriminative binary hash codes. The Siamese network is trained using Triplet Loss to enforce similarity preservation in Hamming space. Extensive experiments on fine-grained benchmarks (CUB-200-2011 and NABirds) as well as a coarse-grained dataset (CIFAR-10) demonstrate that the proposed framework consistently outperforms or matches state-of-the-art hashing methods, validating its robustness and generalization across different levels of semantic granularity.
- 1 May 2026
- Published by SPIE-Intl Soc Optical Eng in Journal of Electronic Imaging
- Vol. 35 (03) , 1-22
Optimizing the number of neurons in the final hash layer of an image retrieval architecture using the Boruta–Shap algorithm to achieve feature space compression and enhanced feature discriminability
Hashing methods in image classification and retrieval convert high-dimensional feature vectors into compact codes, improving storage efficiency and search speed. Learnable hashing improves class separation by minimizing intraclass and maximizing interclass distances but faces challenges. Most approaches maintain similarity between feature vectors and hash codes through extensive training and complex loss functions. In transfer learning, generating distinctive hash codes can lead to model overfitting due to the limited number of target domain samples. To address this, a common approach is to use features from the fully connected layer as the hash unit. In hashing-based methods, the presence of irrelevant or unimportant neurons in the output layer can negatively affect similarity computation and the retrieval of relevant samples. Conversely, limiting the number of neurons in this layer reduces the complexity of the hash unit, resulting in a lightweight architecture that enables model training with fewer samples. However, excessive compression of hash codes due to aggressive neuron reduction may degrade model performance. We investigate the optimal selection of the number of neurons in the output layer of the hash unit by exploring the trade-off between hash code compression and retrieval performance using the Boruta–Shap algorithm. The results demonstrate that the Euclidean distance becomes more discriminative as the number of neurons in the hash unit is reduced, thereby improving the retrieval of images similar to the query. Experimental results show that reducing the number of neurons from 1920 to 698 (63% reduction) improves retrieval accuracy on CIFAR-10. Furthermore, quantitative analysis indicates that the average Euclidean distance between query images and retrieved samples is reduced by up to 44% (from 9.16 to 5.21), whereas the average cosine similarity increases by up to 8% (from 0.72 to 0.78) compared with using a fully sized hashing layer.
- 1 May 2026
- Published by SPIE-Intl Soc Optical Eng in Journal of Electronic Imaging
- Vol. 35 (03) , 1-22
Optimizing the number of neurons in the final hash layer of an image retrieval architecture using the Boruta–Shap algorithm to achieve feature space compression and enhanced feature discriminability
Hashing methods in image classification and retrieval convert high-dimensional feature vectors into compact codes, improving storage efficiency and search speed. Learnable hashing improves class separation by minimizing intraclass and maximizing interclass distances but faces challenges. Most approaches maintain similarity between feature vectors and hash codes through extensive training and complex loss functions. In transfer learning, generating distinctive hash codes can lead to model overfitting due to the limited number of target domain samples. To address this, a common approach is to use features from the fully connected layer as the hash unit. In hashing-based methods, the presence of irrelevant or unimportant neurons in the output layer can negatively affect similarity computation and the retrieval of relevant samples. Conversely, limiting the number of neurons in this layer reduces the complexity of the hash unit, resulting in a lightweight architecture that enables model training with fewer samples. However, excessive compression of hash codes due to aggressive neuron reduction may degrade model performance. We investigate the optimal selection of the number of neurons in the output layer of the hash unit by exploring the trade-off between hash code compression and retrieval performance using the Boruta–Shap algorithm. The results demonstrate that the Euclidean distance becomes more discriminative as the number of neurons in the hash unit is reduced, thereby improving the retrieval of images similar to the query. Experimental results show that reducing the number of neurons from 1920 to 698 (63% reduction) improves retrieval accuracy on CIFAR-10. Furthermore, quantitative analysis indicates that the average Euclidean distance between query images and retrieved samples is reduced by up to 44% (from 9.16 to 5.21), whereas the average cosine similarity increases by up to 8% (from 0.72 to 0.78) compared with using a fully sized hashing layer.
- 23 September 2025
- Published by Springer Nature in Health Information Science and Systems
- Vol. 13 (1) , 1-15
Introduction of hierarchical deep features based on the VGG-16 model for enhancing diabetic retinopathy classification
Diabetic Retinopathy (DR) is a common ocular disease that presents a significant risk of vision loss in individuals with diabetes. Accurate DR classification is critical for preventing disease progression and preserving patients’ vision. However, DR classification is often complex due to the high degree of similarity and overlap among features across its different stages. To address these challenges, this study introduces a computer-aided diagnosis framework that leverages deep neural networks to extract hierarchical features. In this framework, a hierarchical semantic resolution pyramid of retinal images is generated by integrating feature maps from the pooling layers of a deep neural network model. Handcrafted features are subsequently extracted from the HSRP and then fused to create a comprehensive feature vector. This approach fuses pre-trained neural networks with handcrafted features to effectively identify distinctive image characteristics. Notably, due to the independence of feature maps from input data, the proposed architecture does not require retraining or fine-tuning, enhancing its generalizability across different image domains. To identify the most effective discriminative features for classifying DR stages, the Boruta-Shap algorithm is applied to the feature vector. With the high discriminative power of the selected features and the limited dataset size, a random forest classifier is used to categorize DR into five stages. Comparative performance analysis with existing methods demonstrates the effectiveness of the proposed approach in overcoming the challenges associated with DR classification.
- 23 September 2025
- Published by Springer Nature in Health Information Science and Systems
- Vol. 13 (1) , 1-15
Introduction of hierarchical deep features based on the VGG-16 model for enhancing diabetic retinopathy classification
Diabetic Retinopathy (DR) is a common ocular disease that presents a significant risk of vision loss in individuals with diabetes. Accurate DR classification is critical for preventing disease progression and preserving patients’ vision. However, DR classification is often complex due to the high degree of similarity and overlap among features across its different stages. To address these challenges, this study introduces a computer-aided diagnosis framework that leverages deep neural networks to extract hierarchical features. In this framework, a hierarchical semantic resolution pyramid of retinal images is generated by integrating feature maps from the pooling layers of a deep neural network model. Handcrafted features are subsequently extracted from the HSRP and then fused to create a comprehensive feature vector. This approach fuses pre-trained neural networks with handcrafted features to effectively identify distinctive image characteristics. Notably, due to the independence of feature maps from input data, the proposed architecture does not require retraining or fine-tuning, enhancing its generalizability across different image domains. To identify the most effective discriminative features for classifying DR stages, the Boruta-Shap algorithm is applied to the feature vector. With the high discriminative power of the selected features and the limited dataset size, a random forest classifier is used to categorize DR into five stages. Comparative performance analysis with existing methods demonstrates the effectiveness of the proposed approach in overcoming the challenges associated with DR classification.
- 6 May 2025
- Published by Springer Nature in Neural Computing and Applications
- Vol. 37 (19) , 13943-13963
GoogleNet’s semantic hierarchical feature fusion for the classification of lung cancer CT images
- 6 May 2025
- Published by Springer Nature in Neural Computing and Applications
- Vol. 37 (19) , 13943-13963
GoogleNet’s semantic hierarchical feature fusion for the classification of lung cancer CT images
- 6 April 2025
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 63 (9) , 2717-2731
Enhancing breast cancer diagnosis: transfer learning on DenseNet with neural hashing for histopathology fine-grained image classification
Breast cancer is one of the most common types of cancer worldwide. The number of breast cancer cases highlights the importance of disease management at various levels. One complementary method for breast cancer classification is microscopic imaging. Manual histopathological image analysis is time-consuming and prone to human errors. Computer-aided diagnosis (CAD) has emerged as a popular and feasible solution for analyzing medical images due to extensive advancements. Microscopic image analysis can assist physicians in more accurate diagnosis. However, the performance of CAD models needs improvement for practical purposes. In the proposed approach, a baseline model called DenseNet is considered for extracting features from histopathological images. The pre-trained DenseNet model alone is not sufficient for fine-grained feature discrimination between benign and malignant histopathological image samples. Therefore, two hash layers are incorporated at the end of the network to enhance feature separability of the two classes, benign and malignant. The performance of the proposed method is evaluated on the BreakHis histopathological image dataset, with magnifications of 40 × , 100 × , 200 × , and 400 × . The evaluation results confirm the effectiveness of the proposed approach compared to other existing approaches. Furthermore, the interpretability of the proposed approach is demonstrated using the LIME technique. Graphical Abstract
- 6 April 2025
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 63 (9) , 2717-2731
Enhancing breast cancer diagnosis: transfer learning on DenseNet with neural hashing for histopathology fine-grained image classification
Breast cancer is one of the most common types of cancer worldwide. The number of breast cancer cases highlights the importance of disease management at various levels. One complementary method for breast cancer classification is microscopic imaging. Manual histopathological image analysis is time-consuming and prone to human errors. Computer-aided diagnosis (CAD) has emerged as a popular and feasible solution for analyzing medical images due to extensive advancements. Microscopic image analysis can assist physicians in more accurate diagnosis. However, the performance of CAD models needs improvement for practical purposes. In the proposed approach, a baseline model called DenseNet is considered for extracting features from histopathological images. The pre-trained DenseNet model alone is not sufficient for fine-grained feature discrimination between benign and malignant histopathological image samples. Therefore, two hash layers are incorporated at the end of the network to enhance feature separability of the two classes, benign and malignant. The performance of the proposed method is evaluated on the BreakHis histopathological image dataset, with magnifications of 40 × , 100 × , 200 × , and 400 × . The evaluation results confirm the effectiveness of the proposed approach compared to other existing approaches. Furthermore, the interpretability of the proposed approach is demonstrated using the LIME technique. Graphical Abstract
- 24 May 2024
- Published by Oxford University Press (OUP) in The Computer Journal
- Vol. 67 (9) , 2799-2808
Coverless Image Steganography Using Content-Based Image Patch Retrieval
Image steganography is the process of concealing secret information within a cover image. The main challenge of steganography is to ensure that the embedding process does not significantly alter the cover file. In this paper, instead of modifying a cover image to carry information, steganography is performed using a set of images. These images are selected from a dataset of natural images. Each image in the dataset is divided into a number of non-overlapping patches. Then, indexing of the patches is performed based on their features. The secret image is also divided into a set of non-overlapping patches. Similar versions of the patches in the secret image are searched in the dataset to identify candidate patches. The final candidate is selected by calculating the minimum distance between the feature vector of the patches in the secret image and the patches in the dataset. Finally, the receiver retrieves the secret image using the pieces of selected images. Since, instead of embedding information in a cover image, a set of patches from natural images are selected without any changes, this approach can resist change-tracking tools, as demonstrated by experimental results, and also offers the advantage of high embedding capacity.
- 24 May 2024
- Published by Oxford University Press (OUP) in The Computer Journal
- Vol. 67 (9) , 2799-2808
Coverless Image Steganography Using Content-Based Image Patch Retrieval
Image steganography is the process of concealing secret information within a cover image. The main challenge of steganography is to ensure that the embedding process does not significantly alter the cover file. In this paper, instead of modifying a cover image to carry information, steganography is performed using a set of images. These images are selected from a dataset of natural images. Each image in the dataset is divided into a number of non-overlapping patches. Then, indexing of the patches is performed based on their features. The secret image is also divided into a set of non-overlapping patches. Similar versions of the patches in the secret image are searched in the dataset to identify candidate patches. The final candidate is selected by calculating the minimum distance between the feature vector of the patches in the secret image and the patches in the dataset. Finally, the receiver retrieves the secret image using the pieces of selected images. Since, instead of embedding information in a cover image, a set of patches from natural images are selected without any changes, this approach can resist change-tracking tools, as demonstrated by experimental results, and also offers the advantage of high embedding capacity.
- 7 May 2024
- Published by Springer Nature in Multimedia Tools and Applications
- Vol. 84 (11) , 9065-9091
Enhancing image retrieval through entropy-based deep metric learning
- 7 May 2024
- Published by Springer Nature in Multimedia Tools and Applications
- Vol. 84 (11) , 9065-9091
Enhancing image retrieval through entropy-based deep metric learning
- 9 April 2024
- Published by Springer Nature in The Visual Computer
- Vol. 40 (10) , 6867-6881
Content-based image retrieval through fusion of deep features extracted from segmented neutrosophic using depth map
- 9 April 2024
- Published by Springer Nature in The Visual Computer
- Vol. 40 (10) , 6867-6881
Content-based image retrieval through fusion of deep features extracted from segmented neutrosophic using depth map
- 8 April 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 80 (11) , 15917-15939
Fog-Marketing: auction-based multi-tier decentralized markets for fog resource provisioning
- 8 April 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 80 (11) , 15917-15939
Fog-Marketing: auction-based multi-tier decentralized markets for fog resource provisioning