Image Processing: Algorithms and Systems XXIV Program

TUESDAY 3 MARCH 2026

Image Processing: Algorithms and Systems

Session Chair: Atanas Gotchev, Tampere University
08:30 - 10:30
Harbour A

08:30IPAS-240
Evaluation of interactions between two-component self-assembled structures for SEM using multifractal analysis, Yoshihiro Sato, Faculty of Design and Data Science, Tokyo City University (Japan); Taito Ogiya, Advanced Ceramics Research Center, Nagoya Institute of Technology (Japan); Fumio Munakata, Faculty of Science and Engineering, Tokyo City University (Japan) [view abstract]

Multifunctional materials are formed by incorporating two or more components that self-organize through intra- or inter-component interactions. The resulting structure is crucial for enhancing the material's functionality, yet methods for its quantitative evaluation remain underdeveloped. Here, we propose a method based on multifractal analysis to quantitatively evaluate the structure of these materials and elucidate the relationship between their components. To validate this approach, we generated simulated images using Baker's transformation and examined how the resulting metrics change with variations in component concentration and the number of layers. Mutual information, derived from the generalized fractal dimension, was employed as a quantitative indicator of the correlation between the components.

08:50IPAS-241
Multi-camera automatic calibration using human body meshes recovered from multiple persons, Chih-Hsien Chou, Futurewei Technologies, Inc. (US) [view abstract]

Human pose and shape estimation (HPSE) is a crucial function for human-centric applications, while deep learning-based monocular 3D HPSE may subject to depth ambiguity and occlusion problems. Multi-camera systems with wide baselines can solve the problem but accurate and robust multi-camera calibration is a prerequisite. The main objective for the project is to develop algorithms for automatic calibration of multi-camera systems which fully utilize human semantic information from multiple persons simultaneously seen by multiple perspective, wide-angle, and fisheye cameras, without using predetermined calibration patterns or objects. The proposed method solves the multi-view matching problem by combining geometric consistency (represented by pose and shape) and appearance similarity (represented by feature) to calculate the affinity scores between human body meshes detected from different views and then calculate the optimal permutation matrix, which is cycle-consistent across all views for all persons. Humans seen by pairs of cameras and identified as the same person are further processed for pairwise camera calibration using Structure-from-Motion (SfM) and RANSAC algorithms to estimate the camera pose. The proposed method supports multiple persons in the common regions and achieves higher accuracy and faster convergence rate than existing methods using deep learning-based 2D human joint estimators with iterative refinement.

09:10IPAS-242
DSR-QBD: A multi-frame approach for disparity-robust reconstruction in 2 2 OCL quad-bayer sensors, Sun-Young Yoo, Samsung Electronics (Republic of Korea); Yong-Joon Song, Samsung Electronics (Republic of Korea); Byung-Wook Jung, Samsung Electronics (Republic of Korea); Junghoon Jung, Samsung Electronics (Republic of Korea); Sang-Gwon Lee, ; Min-Woong Seo, Samsung Electronics (Republic of Korea); Jae-kyu Lee, Samsung Electronics (Republic of Korea); Jonghyun Go, Samsung Electronics (Republic of Korea); Jaihyuk Song, Samsung Electronics (Republic of Korea) [view abstract]

In this paper, we propose the DSR-QBD framework, which integrates Deep Burst Super-Resolution (DBSR) with U-Net-based 2x2 OCL Quad-Bayer Demosaic. Traditional single-frame methods often struggle with the inherent disparity issues present in 2 2 On-Chip Lens (OCL) Quad-Bayer sensors. Our proposed framework addresses these challenges by treating a single 2 2 OCL image as multiple phase-separated frames, enabling the application of advanced multi-frame super-resolution techniques. Unlike conventional single-frame approaches, our method addresses the disparity issue in 2 2 OCL Quad-Bayer sensors by treating a single 2 2 OCL image as multiple phase-separated frames and applying multi-frame techniques. This strategy enables the effective utilization of phase images to enhance reconstruction quality. Furthermore, the integration of U-QBD within DSR-QBD mitigates the limitations of DBSR, particularly in correcting false pattern artifacts that may arise during reconstruction, thereby yielding more stable and natural results.

09:30IPAS-243
JIST-first-2025-011: Analyzing the impact of color spaces and color augmentation on material segmentation and feature representation, Soroush Shahbaznejad, Rochester Institute of Technology (US); Mekides Assefa Abebe, Rochester Institute of Technology (US); Alireza Rabbanifar, Rochester Institute of Technology (US); Michael Murdoch, Rochester Institute of Technology (US) [view abstract]

Material segmentation hinges on pixel-level cues in hue, chroma, and lightness, yet most pipelines remain in device-referred sRGB and use ad hoc photometric tweaks. We disentangle the effects of color encoding versus principled, opponent-axis augmentations by training the same FCN ResNet-101 across a 3 4 grid: three encodings (sRGB, CIELAB, YCbCr) and three colorimetric perturbations luminance translation (brightness), luminance scaling (contrast), and chroma-magnitude scaling (saturation) plus a no-augmentation baseline. Using overlap-aware metrics (intersection-over-union, IoU; F1) and confirmatory mixed-effects analyses, we find that augmentation not encoding is the primary driver of accuracy. Contrast (luminance scaling) delivers the most reliable global gains; brightness is especially effective for specular materials (metal, glass), while saturation benefits chroma-dominant diffuse classes (paper, fruits). Color encoding acts as a secondary modulator: YCbCr s luminance chrominance separation sharpens neutral or composite boundaries (e.g., plastic), whereas CIELAB s approximate perceptual uniformity aids materials with broader chroma spread. Qualitative overlays and backbone-feature embeddings via t-distributed stochastic neighbor embedding (t-SNE) corroborate these trends, showing cleaner boundaries and more separable latent clusters under luminance-scaled training. Practically, we recommend contrast as the safe default, favoring YCbCr when specular cues dominate and CIELAB when chroma carries the signal.

09:50IPAS-353
ACAM-KD: Adaptive and cooperative attention masking for knowledge distillation, Qizhen Lan, UThealth (US); Qing Tian, University of Alabama at Birmingham (US) [view abstract]

Dense visual prediction tasks, such as detection and segmentation, are crucial for time-critical applications (e.g., autonomous driving and video surveillance). While deep models achieve strong performance, their efficiency remains a challenge. Knowledge distillation (KD) is an effective model compression technique, but existing feature-based KD methods rely on static, teacher-driven feature selection, failing to adapt to the student's evolving learning state or leverage dynamic student-teacher interactions. To address these limitations, we propose Adaptive student-teacher Cooperative Attention Masking for Knowledge Distillation (ACAM-KD), which introduces two key components: (1) Student-Teacher Cross-Attention Feature Fusion (STCA-FF), which adaptively integrates features from both models for a more interactive distillation process, and (2) Adaptive Spatial-Channel Masking (ASCM), which dynamically generates importance masks to enhance both spatial and channel-wise feature selection. Unlike conventional KD methods, ACAM-KD adapts to the student's evolving needs throughout the entire distillation process. Extensive experiments on multiple benchmarks validate its effectiveness. For instance, on COCO2017, ACAM-KD improves object detection performance by up to 1.4 mAP over the state-of-the-art when distilling a ResNet-50 student from a ResNet-101 teacher. For semantic segmentation on Cityscapes, it boosts mIoU by 3.09 over the baseline with DeepLabV3-MobileNetV2 as the student model.

10:10IPAS-355
End-to-end optical–computational co-design for Fourier light field microscopy using Fourier- domain diffractive optical elements, Atanas Gotchev, Tampere University (Finland); Lisi Huang, Tampere University (Finland); Ugur Akpinar, Tampere University (Finland); Jani Makinen, Tampere University (Finland); Erdem Sahin, Tampere University (Finland) [view abstract]

Fourier light field microscopy (FLFM) enables single-shot volumetric imaging but remains limited by the reduced spatial resolution caused by the insertion of a microlens array (MLA), which fundamentally constrains the achievable reconstruction fidelity. In this work, we investigate an end-to-end optical computational co-design approach for FLFM by jointly optimizing a Fourier-domain diffractive optical element (DOE) together with a post-processing reconstruction network. Using a unified forward model and reconstruction pipeline, we compare a conventional FLFM system without DOE and a Fourier DOE-enhanced FLFM system, where both configurations employ identical reconstruction networks and training protocols. Quantitative evaluations demonstrate that the jointly trained Fourier DOE consistently achieves higher PSNR and SSIM than the conventional FLFM baseline. Qualitative results further show sharper structural details in the reconstructed volumes. These results highlight the effectiveness of end-to-end optical–computational co-design and demonstrate that Fourier-domain DOEs provide a powerful mechanism for enhancing FLFM reconstruction quality.