Dulal Chakraborty
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Dulal Chakraborty has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Akash Ghosh1 shared publication
Dipankar Das
1 shared publication- Foujia Islam
Department of ICT, Comilla University, Bangladesh
1 shared publication - Moinur Rahman
Comilla University
1 shared publication - Nargis Parvin
Bangladesh Army International University of Science and Technology
1 shared publication Pronab Kumar Paul
1 shared publication
Latest Publications
- Open Access
- 24 July 2026
- Published by Elsevier in SSRN Electronic Journal
Stop-Gradient Gating for Deformable Image Registration: Monotonicity and Capacity-Constrained Collapse
Most deformable image registration models enforce uniform smoothness, spreading deformation capacity evenly across the image. This assumption becomes a bottleneck for medical images with heterogeneous displacement demands. It leads to over-deformation in structurally stable regions and under-deformation in high-motion areas. To address this limitation, we introduce a gated image registration framework that spatially modulates deformation capacity using a learned multiplicative gate. In practice, the gate allocates deformation capacity across regions to achieve optimal alignment. A stop-gradient operator decouples the gating branch from the registration objective. This defines three training regimes, including none, partial, and full. In the partial regime, the stop-gradient blocks gradients of the registration objective, and only sparsity regularization influences the gating parameters. As a result, the mean gate activation drops from ≈ 0.29 to 0.0033, and registration collapses (NCC falls from 0.985 to 0.809) once deformation demand exceeds the model’s effective representational capacity. We characterize this collapse behavior, showing monotonic decay and sharp transition determined by a critical gate threshold 𝑔†. Controlled synthetic experiments and three medical imaging datasets confirm the collapse, which remains absent in low-deformation settings. We further show that this instability can be mitigated by increasing the learning-rate ratio of the gate to the deformation network or by introducing controlled gradient leakage with a target-aware sparsity objective. This insight provides a simple design principle for stable gated registration. Never leave sparsity alone without balancing it against a registration gradient or a high gate learning rate.
- 24 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessStop-Gradient Gating for Deformable Image Registration: Monotonicity and Capacity-Constrained Collapse
Most deformable image registration models enforce uniform smoothness, spreading deformation capacity evenly across the image. This assumption becomes a bottleneck for medical images with heterogeneous displacement demands. It leads to over-deformation in structurally stable regions and under-deformation in high-motion areas. To address this limitation, we introduce a gated image registration framework that spatially modulates deformation capacity using a learned multiplicative gate. In practice, the gate allocates deformation capacity across regions to achieve optimal alignment. A stop-gradient operator decouples the gating branch from the registration objective. This defines three training regimes, including none, partial, and full. In the partial regime, the stop-gradient blocks gradients of the registration objective, and only sparsity regularization influences the gating parameters. As a result, the mean gate activation drops from ≈ 0.29 to 0.0033, and registration collapses (NCC falls from 0.985 to 0.809) once deformation demand exceeds the model’s effective representational capacity. We characterize this collapse behavior, showing monotonic decay and sharp transition determined by a critical gate threshold 𝑔†. Controlled synthetic experiments and three medical imaging datasets confirm the collapse, which remains absent in low-deformation settings. We further show that this instability can be mitigated by increasing the learning-rate ratio of the gate to the deformation network or by introducing controlled gradient leakage with a target-aware sparsity objective. This insight provides a simple design principle for stable gated registration. Never leave sparsity alone without balancing it against a registration gradient or a high gate learning rate.
- Open Access
- 10 December 2025
- Published by FOREX Publication in International Journal of Electrical and Electronics Research
- Vol. 13 (4) , 730-734
Fundamental Frequency Extraction by Utilizing the Combination of Spectrum in Noisy Speech
Speech is the audible acoustic signal generated by the articulatory system (lungs, vocal folds, vocal tract, tongue, lips) to communicate language. The fundamental frequency (F_0) is the lowest frequency component of a speech waveform and corresponds to the vibration rate of the vocal folds during voiced speech. It also determines the pitch of the speaker’s voice. In speech signal processing, this acoustic waveform is captured and analyzed to extract information about what is being said, how it is being said and who is saying it. In various speech processing applications such as voice synthesis, speaker recognition and emotion analysis accurate extraction of the fundamental frequency (F_0) is a vital task. However, the vocal tract’s formants can sometimes significantly alter the glottal waveform’s shape, making it difficult to identify the true pitch. Additionally, in the presence of background noise, traditional pitch detection techniques often experience a considerable decline in performance. This work proposes a robust method for extracting the fundamental frequency by utilizing the complementary advantages of the power spectrum and logarithmic spectrum in noisy speech environments. The power spectrum mitigates noise effects, while the logarithmic operation effectively separates vocal tract characteristics from the source excitation. The proposed approach integrates autocorrelation-based power spectral analysis with cepstral techniques derived from the log spectrum to improve pitch estimation accuracy under adverse conditions. Experimental results on noisy speech datasets show that the proposed hybrid method achieves lower gross pitch error and greater robustness than traditional methods such as BaNa, autocorrelation and cepstral techniques.
- 10 December 2025
- Published by FOREX Publication in International Journal of Electrical and Electronics Research
- Vol. 13 (4) , 730-734
Open AccessFundamental Frequency Extraction by Utilizing the Combination of Spectrum in Noisy Speech
Speech is the audible acoustic signal generated by the articulatory system (lungs, vocal folds, vocal tract, tongue, lips) to communicate language. The fundamental frequency (F_0) is the lowest frequency component of a speech waveform and corresponds to the vibration rate of the vocal folds during voiced speech. It also determines the pitch of the speaker’s voice. In speech signal processing, this acoustic waveform is captured and analyzed to extract information about what is being said, how it is being said and who is saying it. In various speech processing applications such as voice synthesis, speaker recognition and emotion analysis accurate extraction of the fundamental frequency (F_0) is a vital task. However, the vocal tract’s formants can sometimes significantly alter the glottal waveform’s shape, making it difficult to identify the true pitch. Additionally, in the presence of background noise, traditional pitch detection techniques often experience a considerable decline in performance. This work proposes a robust method for extracting the fundamental frequency by utilizing the complementary advantages of the power spectrum and logarithmic spectrum in noisy speech environments. The power spectrum mitigates noise effects, while the logarithmic operation effectively separates vocal tract characteristics from the source excitation. The proposed approach integrates autocorrelation-based power spectral analysis with cepstral techniques derived from the log spectrum to improve pitch estimation accuracy under adverse conditions. Experimental results on noisy speech datasets show that the proposed hybrid method achieves lower gross pitch error and greater robustness than traditional methods such as BaNa, autocorrelation and cepstral techniques.