基于多任务卷积神经网络的虹膜图像质量评估方法

为了自动、准确、高效地评估采集图像的质量,设计了一个名为MTIQA的卷积神经网络。该网络能够输出与主观评价指标保持较高一致性的客观评估结果。MTIQA采用多任务学习策略,包含网络训练质量评估和失真类型分类两个任务,将两个任务的损失融合并构成新的损失函数。为了评估算法所得到的客观指标的可靠性,建立了名为SIR2019的单眼虹膜质量评估数据集,并召集志愿者进行主观实验以得到主观评价指标。在SIR2019和CASIA-Iris-Distance-Lamp数据集上的实验结果表明,该网络在虹膜图像质量评估上具有较好的可行性、准确性和鲁棒性。

An Iris Image Quality Evaluation Method Based on Multi-task Convolutional Neural Network

In order to automatically, accurately and efficiently evaluate the quality of collected images, a convolutional neural network named MTIQA is designed, which can output objective evaluation results with high consistency with subjective evaluation indexes.MTIQA adopts the multi-task learning strategy, classifies the two tasks through a network training quality assessment and distortion type, and finally fuses the losses of the two tasks into a new loss function.In order to evaluate the reliability of the objective indexes obtained by the algorithm, a monocular iris quality evaluation dataset named SIR2019 is established, and volunteers are recruited for subjective experiments to obtain the subjective evaluation indexes.The proposed network is tested on SIR2019 and CASIA-Iris-Distance-Lamp datasets.The experimental results show that the proposed network has good accuracy, robustness and feasibility in iris image quality evaluation.