From dabc97ddd0579d1ec5e0bcf461b5299eac870e65 Mon Sep 17 00:00:00 2001 From: Darien Pardinas Diaz Date: Mon, 1 Jan 2024 20:52:45 +0000 Subject: [PATCH] Fixed spelling error --- README.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index 2d2e96d..02776e8 100644 --- a/README.md +++ b/README.md @@ -89,11 +89,11 @@ wsl -d Ubuntu-22.04 docker run hello-world If you want to do Typescript/Javascript development directly on Windows, install Node.js and NPM on Windows. With Powershell this can be achieved via `winget install OpenJS.NodeJS.LTS --accept-source-agreements`. -## Cropping to photo standard dimentions +## Cropping to photo standard sizes -In order to crop and scale the face of the person to a particular passport requirement, the following approach was investigated. Given the set of detected facial landmarks *A*, *B*, *C* and *D*, we would like to estimate *P* and *Q* with accuracy that is sufficient to ensure that the face in the output photo fall within the limits of the size requirements. In other words, the estimated location of the crown (*P’*) and chin point (*Q’*) should be such that the distance *P’Q’* scaled by the distance between the ideal location of the crown (*P*) and chin point (*Q*) falls within the tolerance range allowed in photo ID specifications. For instance, for the case of Australian passport, the allowed scale drift range is **±5.88%** given that the face height (chin to crown) must be between 32mm and 36mm: $\dfrac{1}{2} \times \dfrac{36 - 32}{0.5 \times (32+36)} = \pm 5.88$%. +In order to crop and scale the face of the person to a particular photo ID requirement, the following approach was investigated. Given the set of detected facial landmarks *A*, *B*, *C* and *D*, we would like to estimate *P* and *Q* with accuracy that is sufficient to ensure that the face in the output photo fall within the limits of the size requirements. In other words, the estimated location of the crown (*P’*) and chin point (*Q’*) should be such that the distance *P’Q’* scaled by the distance between the ideal location of the crown (*P*) and chin point (*Q*) falls within the tolerance range allowed in photo ID specifications. For instance, for the case of Australian passport, the allowed scale drift range is **±5.88%** given that the face height (chin to crown) must be between 32mm and 36mm: $\dfrac{1}{2} \times \dfrac{36 - 32}{0.5 \times (32+36)} = \pm 5.88$%.
To develop and validate the proposed approach, facial landmarks from the [SCFace database](http://www.scface.org/) were used. The SCFace database contains images for 130 different subjects and frontal images of each individual were carefully annotated by the [Biometric Recognition Group - ATVS at Escuela Politecnica Superior of the Universidad Autonoma de Madrid [ATVS]](https://atvs.ii.uam.es/scfacedb_landmarks.html). -The procedure to estimate *P’* and *Q’* from *A*, *B*, *C* and *D* is as follow: Firstly, points *M* and *N* are found as the center of segments *AB* and *CD* respectively. *P’* and *Q’* are expected to fall in the line that passes through *M* and *N*. Then using a normalization distance *KK = |AB| + |MN|* and scale constants *α* and *β*, we estimate *P’Q’* = *αKK* and *M’Q’* = *βKK*. From the dataset *α* and *β* were adjusted to minimize the estimation error of *P'* and *Q'*. +The procedure to estimate *P’* and *Q’* from *A*, *B*, *C* and *D* is as follow: Firstly, points *M* and *N* are found as the center of segments *AB* and *CD* respectively. *P’* and *Q’* are expected to fall in the line that passes through *M* and *N*. Then using a normalization distance *KK = |AB| + |MN|* and scale constants *α* and *β*, we estimate *P’Q’* = *αKK* and *M’Q’* = *βKK*. From the dataset *α* and *β* were adjusted to minimize the estimation error of *P'* and *Q'*. Having estimated positions *P'* and *Q'* allows us to crop to most photo ID standards defined throughout the world.