---
tags: 生物辨識
---

# face image quality assessment

## Motvation
* 人工標記人臉資料的品質是很麻煩又非常容易出錯的
* 不是要分辨出人看起來清晰的圖像，還是要找出模型可以正確判斷的

## Method

#### SER-FIQ: Unsupervised Estimation of Face Image Quality Based on Stochastic Embedding Robustness
[github](https://github.com/pterhoer/FaceImageQuality)

透過dropout層產出隨機不同embedding，兩兩計算產生質量分數


分數計算公式如下


#### SDD-FIQA: Unsupervised Face Image Quality Assessment with Similarity Distribution Distance
[github](https://github.com/Tencent/TFace/tree/quality)

對每一張圖片隨機選m個類內及m個類外，利用Wasserstein Distance算出分數

[Wasserstein Distance](https://zhuanlan.zhihu.com/p/353418080)

Quality Model:將Recognition model拿掉embedding layer和classification layer加上一個fully-connected layer並利用Huber loss訓練