Muku Akasaka’s research while affiliated with University of Shizuoka and other places

... Extensive evaluation on the Unconstrained Ear Recognition Challenge 2023 (UERC2023) benchmark [9] demonstrates that EdgeEar achieves performance comparable to state-ofthe-art deep learning architectures across key metrics, such as Equal Error Rate, as shown in Fig. 1, while maintaining a significantly smaller parameter count. These findings highlight the potential of adapting edge-optimized face recognition architectures for ear recognition in resource-constrained scenarios. ...

Reference:

EdgeEar: Efficient and Accurate Ear Recognition for Edge Devices

... Dataset approach is an efficient method for improving fairness among races, yet constructing large-scale datasets with fairness considerations is not straightforward. Additionally, it has been pointed out that racial boundaries are ambiguous [6] and that the impact of fairness can also arise from the interrelationship between racial and environmental factors [12,17]. Many researches has also been conducted to address racial bias by innovating the structure of models to allow fair learning even with existing datasets that are biased towards certain races. ...

Reference:

LabellessFace: Fair Metric Learning for Face Recognition without Attribute Labels

... In addition to the method basis and training data, previous works in the literature can be categorized based on the output resolution, i.e., low-resolution (e.g., 112 × 112) or high-resolution (e.g., 1024 × 1024) reconstructed face images. However, most works in the literature generate lowresolution face images [15], [16], [17], [18], [19], [20], [21]. In [15], an optimization-based method for whitebox TI attacks was proposed, where starting from a random noise or a guiding image an iterative gradient-ascend approach is used to generate an image that has a similar facial template. ...

Reference:

Template Inversion Attack Using Synthetic Face Images Against Real Face Recognition Systems