Takanori Fujisawa
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Takanori Fujisawa has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Masaaki Ikehara
Department of Electronics and Electrical Engineering, Keio University, Yokohama, Japan
4 shared publications - Hiroki Tomosada
Department of Electronics and Electrical Engineering, Keio University, Yokohama 223-8522, Japan
1 shared publication Masaaki Ikehara
Department of Electronics and Electrical Engineering, Keio University, Yokohama, Kanagawa, 223-8522 Japan
1 shared publicationRyo Fujimoto
Department of Electronics and Electrical Engineering, Keio University
1 shared publicationSora Harada
Keio Univ.,EEE Dept.,Yokohama, Kanagawa,Japan,223-8522
1 shared publicationTakahiro Kudo
1 shared publication
Latest Publications
- Open Access
- 29 September 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 135224-135233
GAN-Based Image Deblurring Using DCT Loss With Customized Datasets
In this paper, we propose a high quality image deblurring method that uses discrete cosine transform (DCT) and requires less computational complexity. We train our model on a new dataset which is customized to include images with large motion blurs. Recently, Convolutional Neural Network (CNN) and Generative Adversarial Network (GAN) based algorithms have been proposed for image deblurring. Moreover, multi-scale and multi-patch architectures of CNN restore blurred images clearly and suppress more ringing or blocking artifacts, but they take a longer time to process. To improve the quality of deblured images and reduce the computational time, we propose a method called "DeblurDCTGAN" that preserves texture and suppresses ringing artifacts in the restored image without multi-scale or multi-patch architecture using DCT based loss.This loss compares the restored image and the ground truth image in the frequency domain. With this loss, DeblurDCTGAN can reduce block noise and ringing artifacts while maintaining deblurring performance. Our experimental results show that DeblurDCTGAN gets the highest performances in terms of PSNR, SSIM, and running time compared with conventional methods. In terms of real image datasets, DeblurDCTGAN shows a better performance by using a customized training dataset made from GoPro, DVD, NFS and HIDE training datasets. Experimented code with pre-trained weights, datasets and results are available at https://github.com/Hiroki-Tomosada/DCTGAN-master.
- 29 September 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 135224-135233
Open AccessGAN-Based Image Deblurring Using DCT Loss With Customized Datasets
In this paper, we propose a high quality image deblurring method that uses discrete cosine transform (DCT) and requires less computational complexity. We train our model on a new dataset which is customized to include images with large motion blurs. Recently, Convolutional Neural Network (CNN) and Generative Adversarial Network (GAN) based algorithms have been proposed for image deblurring. Moreover, multi-scale and multi-patch architectures of CNN restore blurred images clearly and suppress more ringing or blocking artifacts, but they take a longer time to process. To improve the quality of deblured images and reduce the computational time, we propose a method called "DeblurDCTGAN" that preserves texture and suppresses ringing artifacts in the restored image without multi-scale or multi-patch architecture using DCT based loss.This loss compares the restored image and the ground truth image in the frequency domain. With this loss, DeblurDCTGAN can reduce block noise and ringing artifacts while maintaining deblurring performance. Our experimental results show that DeblurDCTGAN gets the highest performances in terms of PSNR, SSIM, and running time compared with conventional methods. In terms of real image datasets, DeblurDCTGAN shows a better performance by using a customized training dataset made from GoPro, DVD, NFS and HIDE training datasets. Experimented code with pre-trained weights, datasets and results are available at https://github.com/Hiroki-Tomosada/DCTGAN-master.
- Open Access
- 26 January 2020
- Published by Society for Imaging Science & Technology in Electronic Imaging
- Vol. 32 (10) , 181-1
Non-Blind Image Deconvolution Based on “Ringing” Removal Using Convolutional Neural Network
Fast track article for IS&T International Symposium on Electronic Imaging 2020: Image Processing: Algorithms and Systems proceedings.
- 26 January 2020
- Published by Society for Imaging Science & Technology in Electronic Imaging
- Vol. 32 (10) , 181-1
Open AccessNon-Blind Image Deconvolution Based on “Ringing” Removal Using Convolutional Neural Network
Fast track article for IS&T International Symposium on Electronic Imaging 2020: Image Processing: Algorithms and Systems proceedings.
- 1 November 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 401, 1836-1841
Multi-Pitch Estimation using NHF with Multi-Dictionary Distinguishing Attack and Reverberation of Sounds
This paper proposes a multiple pitch estimation algorithm for the piano music which improves both precision and processing time. The conventional method applies non-negative matrix factorization (NMF) and singular value decomposition to ensure the continuity of sound and pattern of musics. However, audio signal spectrogram has large elements and processing singular value decomposition is inefficient in the computational time. Our method separates the input audio spectrogram into the block of sound. Then we apply a NMF with group sparsity constraint to enforce the continuity of sound. In addition, we use the different dictionaries for the attack part and the reverberation part of the sound. It improves the precision of the estimation for each part.
- 1 November 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 401, 1836-1841
Multi-Pitch Estimation using NHF with Multi-Dictionary Distinguishing Attack and Reverberation of Sounds
This paper proposes a multiple pitch estimation algorithm for the piano music which improves both precision and processing time. The conventional method applies non-negative matrix factorization (NMF) and singular value decomposition to ensure the continuity of sound and pattern of musics. However, audio signal spectrogram has large elements and processing singular value decomposition is inefficient in the computational time. Our method separates the input audio spectrogram into the block of sound. Then we apply a NMF with group sparsity constraint to enforce the continuity of sound. In addition, we use the different dictionaries for the attack part and the reverberation part of the sound. It improves the precision of the estimation for each part.
- Open Access
- 18 February 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 7, 22719-22728
High-Accuracy Image Rotation and Scale Estimation Using Radon Transform and Sub-Pixel Shift Estimation
Rotation and scale estimation of images are fundamental tasks in image registration. The conventional estimation method uses log-polar transform and 1D shift estimation to estimate rotation and scale regardless of the shift of images. However, this transform requires interpolation of the frequency components, which causes estimation error. We propose a rotation and scale estimation algorithm based on Radon transform and sub-pixel shift estimation. Radon transform can estimate the rotation independent of the shift and can reduce the influence of interpolation error. In addition, sub-pixel shift estimation using linear approximation of the phase component improves the precision of 1D shift estimation and achieves accurate rotation estimation. The proposed method was evaluated on test images, and the results demonstrate that the proposed method has higher accuracy compared to log-polar transform.
- 18 February 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 7, 22719-22728
Open AccessHigh-Accuracy Image Rotation and Scale Estimation Using Radon Transform and Sub-Pixel Shift Estimation
Rotation and scale estimation of images are fundamental tasks in image registration. The conventional estimation method uses log-polar transform and 1D shift estimation to estimate rotation and scale regardless of the shift of images. However, this transform requires interpolation of the frequency components, which causes estimation error. We propose a rotation and scale estimation algorithm based on Radon transform and sub-pixel shift estimation. Radon transform can estimate the rotation independent of the shift and can reduce the influence of interpolation error. In addition, sub-pixel shift estimation using linear approximation of the phase component improves the precision of 1D shift estimation and achieves accurate rotation estimation. The proposed method was evaluated on test images, and the results demonstrate that the proposed method has higher accuracy compared to log-polar transform.
- 1 January 2018
- Published by Institute of Electronics, Information and Communications Engineers (IEICE) in IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences
- Vol. E101.A (1) , 267-272
Sub-Pixel Shift Estimation of Image Based on the Least Squares Approximation in Phase Region
This paper proposes a novel method to estimate non-integer shift of images based on least squares approximation in the phase region. Conventional methods based on Phase Only Correlation (POC) take correlation between an image and its shifted image, and then estimate the non-integer shift by fitting the model equation. The problem when estimating using POC is that the estimated peak of the fitted model equation may not match the true peak of the POC function. This causes error in non-integer shift estimation. By calculating the phase difference directly in the phase region, the proposed method allows the estimation of sub-pixel shift through least squares approximation. Also by utilizing the characteristics of natural images, the proposed method limits adoption range for least squares approximation. By these improvements, the proposed method achieves high accuracy, and we validate through some examples.
- 1 January 2018
- Published by Institute of Electronics, Information and Communications Engineers (IEICE) in IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences
- Vol. E101.A (1) , 267-272
Sub-Pixel Shift Estimation of Image Based on the Least Squares Approximation in Phase Region
This paper proposes a novel method to estimate non-integer shift of images based on least squares approximation in the phase region. Conventional methods based on Phase Only Correlation (POC) take correlation between an image and its shifted image, and then estimate the non-integer shift by fitting the model equation. The problem when estimating using POC is that the estimated peak of the fitted model equation may not match the true peak of the POC function. This causes error in non-integer shift estimation. By calculating the phase difference directly in the phase region, the proposed method allows the estimation of sub-pixel shift through least squares approximation. Also by utilizing the characteristics of natural images, the proposed method limits adoption range for least squares approximation. By these improvements, the proposed method achieves high accuracy, and we validate through some examples.