Deep artifact-free residual network for single-image super-resolution

Nasrollahi, Hamdollah; Farajzadeh, Kamran; Hosseini, Vahid; Zarezadeh, Esmaeil; Abdollahzadeh, Milad

doi:10.1007/s11760-019-01569-3

Cited by 10 publications

(5 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…2.8. To reduce artifacts, additional steps are required when generating SR images, such as using residual layers 80 …”

Section: Resultsmentioning

confidence: 99%

“…To reduce artifacts, additional steps are required when generating SR images, such as using residual layers. 80 Finally, the DenseED block in ML model architectures helps to generate SR images when the ML model is trained with a small dataset. The performance improvement depends on optimizing other hyper-parameters and parameters of the network, including learning rate, non-linear activation, loss function, and weight decay, on regularizing the over-fitting.…”

Section: Srdenseed With Computational Sr Techniquesmentioning

confidence: 99%

See 1 more Smart Citation

Small training dataset convolutional neural networks for application-specific super-resolution microscopy

Mannam

Howard

2023

J. Biomed. Opt.

View full text Add to dashboard Cite

Significance: Machine learning (ML) models based on deep convolutional neural networks have been used to significantly increase microscopy resolution, speed [signal-to-noise ratio (SNR)], and data interpretation. The bottleneck in developing effective ML systems is often the need to acquire large datasets to train the neural network. We demonstrate how adding a "dense encoderdecoder" (DenseED) block can be used to effectively train a neural network that produces superresolution (SR) images from conventional microscopy diffraction-limited (DL) images trained using a small dataset [15 fields of view (FOVs)].Aim: The ML helps to retrieve SR information from a DL image when trained with a massive training dataset. The aim of this work is to demonstrate a neural network that estimates SR images from DL images using modifications that enable training with a small dataset.Approach: We employ "DenseED" blocks in existing SR ML network architectures. DenseED blocks use a dense layer that concatenates features from the previous convolutional layer to the next convolutional layer. DenseED blocks in fully convolutional networks (FCNs) estimate the SR images when trained with a small training dataset (15 FOVs) of human cells from the Widefield2SIM dataset and in fluorescent-labeled fixed bovine pulmonary artery endothelial cells samples.Results: Conventional ML models without DenseED blocks trained on small datasets fail to accurately estimate SR images while models including the DenseED blocks can. The average peak SNR (PSNR) and resolution improvements achieved by networks containing DenseED blocks are ≈3.2 dB and 2×, respectively. We evaluated various configurations of target image generation methods (e.g., experimentally captured a target and computationally generated target) that are used to train FCNs with and without DenseED blocks and showed that including DenseED blocks in simple FCNs outperforms compared to simple FCNs without DenseED blocks.Conclusions: DenseED blocks in neural networks show accurate extraction of SR images even if the ML model is trained with a small training dataset of 15 FOVs. This approach shows that microscopy applications can use DenseED blocks to train on smaller datasets that are application-specific imaging platforms and there is promise for applying this to other imaging modalities, such as MRI/x-ray, etc.

show abstract

“…2.8. To reduce artifacts, additional steps are required when generating SR images, such as using residual layers 80 …”

Section: Resultsmentioning

confidence: 99%

Section: Srdenseed With Computational Sr Techniquesmentioning

confidence: 99%

Small training dataset convolutional neural networks for application-specific super-resolution microscopy

Mannam

Howard

2023

J. Biomed. Opt.

View full text Add to dashboard Cite

show abstract

“…Metode super-resolution berbasis deep learning telah menunjukkan potensi tinggi dalam bidang interpolasi dan restorasi gambar dibandingkan dengan algoritma interpolasi berbasis piksel konvensional (Kim et al, 2021). Beberapa metode super-resolution terkini yang berbasis CNN (Convolutional Neural Network) (Nasrollahi et al, 2020;J. Wang et al, 2022) dan GAN (Generative Adversarial Network) (Ren et al, 2020;Tian et al, 2022) telah membuktikan kemampuannya dalam meningkatkan resolusi gambar yang rendah.…”

Section: Pendahuluanunclassified

Peningkatan Performa Pengenalan Wajah pada Gambar <i>Low-Resolution</i> Menggunakan Metode<i> Super-Resolution</i>

Rohim,

Nisa,

Hindratno

et al. 2024

JTIIK

View full text Add to dashboard Cite

Kartu Tanda Penduduk Elektronik (KTP-el) merupakan identitas wajib bagi penduduk Indonesia. Penyimpanan pada cip KTP-el yang mana selain digunakan untuk menyimpan gambar potret wajah individu, juga harus dapat menyimpan identitas lain seperti biodata, tanda tangan, dan sidik jari kiri dan kanan. Keterbatasan tersebut mengharuskan gambar potret wajah disimpan pada ukuran low-resolution (LR) sehingga sistem pengenalan wajah tidak optimal. Dalam penelitian ini, kami menggunakan Poznan University of Technology (PUT) Face database yang terdiri atas 200 gambar dari 100 individu. Data tersebut dilakukan proses down sampling menggunakan bicubic interpolation untuk menghasilkan data LR. Kami menginvestigasi penggunaan metode super-resolution (SR) berbasis deep learning, termasuk DFDNet, LapSRN, GFPGAN, Real-ESRGAN, Real-ESRGAN+GFPGAN, dan FaceSPARNet. Hal ini bertujuan untuk meningkatkan kualitas gambar LR. Evaluasi performa dilakukan dengan menggunakan matriks False Rejection Rate(FRR) pada beberapa tingkatan False Acceptance Rate (FAR). Hasil penelitian menunjukkan bahwa beberapa metode SR terutama FaceSPARNet menunjukkan peningkatan performa face recognition hingga 2%. Sedangkan, metode SR yang berbasis GAN (GFPGAN, Real-ESRGAN, Real-ESRGAN+GFPGAN) cenderung meningkatkan false reject rate. Penelitian ini menunjukkan bahwa metode SR dari kategori General Basic CNN-based FSR dapat digunakan untuk meningkatkan kinerja face recognition pada gambar LR, seperti pada KTP-el.

show abstract

“…2.8. To reduce artifacts, additional steps are required when generating super-resolution images, such as using residual layers [80].…”

Section: Srdesnseed With Computational Sr Techniquesmentioning

confidence: 99%

Small Training Dataset Convolutional Neural Networks for Application Specific Super-Resolution Microscopy

Mannam

Howard

2022

Preprint

View full text Add to dashboard Cite

Machine learning (ML) models based on convolutional neural networks (CNNs) have been used to significantly increase microscopy resolution, speed (signal-to-noise ratio), and data interpretation. The bottleneck in developing effective ML systems is often the need to acquire large datasets to train the neural network. This paper demonstrates how adding a dense encoder-decoder block can be useful to effectively train a CNN that provides super-resolution images from conventional diffraction-limited microscopy images when trained using a small dataset containing 15 field-of-views (FOVs). DenseED blocks use a dense layer that concatenates features from the previous convolutional layer to the next convolutional layer. Demonstrate using DenseED blocks in fully convolutional networks (FCNs) to estimate the super-resolution images when trained with a small training dataset (15 FOVs) of human cells from the Widefield2SIM dataset and the fluorescent-labeled fixed bovine pulmonary artery endothelial cells (BPAE samples). Conventional ML models without DenseED blocks trained on small datasets fail to accurately estimate super-resolution images while models, including the DenseED blocks can. The average resolution and peak signal-to-noise ratio (PSNR) improvements achieved using DenseED blocks in FCNs when trained with 15 FOVs are 2 times and ~3.2 dB, respectively. In addition, we evaluated various configurations of target image generation methods (experimentally captured target and computationally generated target) that are used to train the FCNs with and without DenseED blocks and showed with DenseED blocks outperform compared to simple FCNs without DenseED blocks. Hence, the proposed approach indicates that microscopy applications can use DenseED blocks to train on smaller datasets that are application specific/experimental modality specific imaging platforms such as MRI/X-ray and other in vivo imaging modalities.

show abstract

Deep artifact-free residual network for single-image super-resolution

Cited by 10 publications

References 42 publications

Small training dataset convolutional neural networks for application-specific super-resolution microscopy

Small training dataset convolutional neural networks for application-specific super-resolution microscopy

Peningkatan Performa Pengenalan Wajah pada Gambar <i>Low-Resolution</i> Menggunakan Metode<i> Super-Resolution</i>

Small Training Dataset Convolutional Neural Networks for Application Specific Super-Resolution Microscopy

Contact Info

Product

Resources

About