Unsupervised Palmprint Image Quality Assessment via Pseudo-Label Generation and Ranking Guidance

Yuchen Zou，Chengcheng Liu，Huikai Shao， Dexing Zhong

Research on palmprint recognition has been progressing toward a contactless and unconstrained direction, where the imaging quality is easily disturbed by various factors. Determining the trustworthiness of palmprint image input to the system is crucial for stable and reliable recognition. However, there are few studies on Palmprint Image Quality Assessment (PIQA) for practical scenarios due to the lack of labeled image quality data. In this paper, we propose a Pseudo-label Generation and Ranking Guidance (PGRG) framework for label-free PIQA. Two-stage preparation is performed before the training of quality model. 1) Estimate the recognizability of palmprint images to generate quality pseudo-labels. 2) Generate pseudo-images and guide their quality ranking to pre-train the model. In the final training phase, we propose a label-based ranking loss. It guides the model to determine quality relative relationships by emphasizing the ranking between pseudo-labels. This allows our framework to be applied to a variety of palmprint recognition models. Adequate experiments are conducted on three benchmark unconstrained palmprint databases. The results show that our PGRG framework can effectively improve the performance of existing advanced palmprint recognition models. The Equal Error Rate (EER) can be reduced by up to 2.41% by discarding a small amount of low-quality images compared to the baseline.