This repository compare a selfie with images from identity documents and response if the selfie match.

Last update: Jan 27, 2022

Overview

aws-rekognition-facecompare

This repository compare a selfie with images from identity documents and response if the selfie match.

This code was made in a Python Notebook under SageMaker.

Set up:

Create a Notebook Instance in SageMaker

Notebook instance type : ml.t2.medium

Volume Size : 5GB EBS

Create a role for SageMaker with the following policies:

AmazonS3FullAccess

AmazonRekognitionFullAccess

AmazonSageMakerFullAccess

Create a S3 Bucket
Inside bucket create folder to insert the dataset images

Code Explanation

boto3 is needed to use the aws client of S3 and Rekognition. Just like what we do with variables, data can be kept as bytes in an in-memory buffer when we use the io module’s Byte IO operations, so we can load images froms S3. At least Pillow is needed for image plotting.

import boto3
import io
from PIL import Image, ImageDraw, ExifTags, ImageColor

rekognition_client=boto3.client('rekognition')
s3_resource = boto3.resource('s3')

In this notebook I use two functions of AWS Rekognition

detect_faces : Detect faces in the image. It also evaluate different metrics and create different landmarks for all elements of the face like eyes positions.
compare_faces : Evaluate the similarity of two faces.

Case of use

Here I explain how to compare two images

The compare function

IMG_SOURCE ="dataset-CI/imgsource.jpg"
IMG_TARGET ="dataset-CI/img20.jpg"
response = rekognition_client.compare_faces(
                SourceImage={
                    'S3Object': {
                        'Bucket': BUCKET,
                        'Name': IMG_SOURCE
                    }
                },
                TargetImage={
                    'S3Object': {
                        'Bucket': BUCKET,
                        'Name': IMG_TARGET                    
                    }
                }
)

response

{'SourceImageFace': {'BoundingBox': {'Width': 0.3676206171512604,
   'Height': 0.5122320055961609,
   'Left': 0.33957839012145996,
   'Top': 0.18869829177856445},
  'Confidence': 99.99957275390625},
 'FaceMatches': [{'Similarity': 99.99634552001953,
   'Face': {'BoundingBox': {'Width': 0.14619407057762146,
     'Height': 0.26241832971572876,
     'Left': 0.13103649020195007,
     'Top': 0.40437373518943787},
    'Confidence': 99.99955749511719,
    'Landmarks': [{'Type': 'eyeLeft',
      'X': 0.17260463535785675,
      'Y': 0.5030772089958191},
     {'Type': 'eyeRight', 'X': 0.23902645707130432, 'Y': 0.5023221969604492},
     {'Type': 'mouthLeft', 'X': 0.17937719821929932, 'Y': 0.5977044105529785},
     {'Type': 'mouthRight', 'X': 0.23477530479431152, 'Y': 0.5970458984375},
     {'Type': 'nose', 'X': 0.20820103585720062, 'Y': 0.5500822067260742}],
    'Pose': {'Roll': 0.4675966203212738,
     'Yaw': 1.592366099357605,
     'Pitch': 8.6331205368042},
    'Quality': {'Brightness': 85.35185241699219,
     'Sharpness': 89.85481262207031}}}],
 'UnmatchedFaces': [],
 'ResponseMetadata': {'RequestId': '3ae9032d-de8a-41ef-b22f-f95c70eed783',
  'HTTPStatusCode': 200,
  'HTTPHeaders': {'x-amzn-requestid': '3ae9032d-de8a-41ef-b22f-f95c70eed783',
   'content-type': 'application/x-amz-json-1.1',
   'content-length': '911',
   'date': 'Wed, 26 Jan 2022 17:21:53 GMT'},
  'RetryAttempts': 0}}

If the source image match with the target image, the json return a key "FaceMatches" with a non-empty, otherwise it returns a key "UnmatchedFaces" with a non-empty array.

# Analisis imagen source
s3_object = s3_resource.Object(BUCKET,IMG_SOURCE)
s3_response = s3_object.get()
stream = io.BytesIO(s3_response['Body'].read())
image=Image.open(stream)
imgWidth, imgHeight = image.size  
draw = ImageDraw.Draw(image)  

box = response['SourceImageFace']['BoundingBox']
left = imgWidth * box['Left']
top = imgHeight * box['Top']
width = imgWidth * box['Width']
height = imgHeight * box['Height']

print('Left: ' + '{0:.0f}'.format(left))
print('Top: ' + '{0:.0f}'.format(top))
print('Face Width: ' + "{0:.0f}".format(width))
print('Face Height: ' + "{0:.0f}".format(height))

points = (
    (left,top),
    (left + width, top),
    (left + width, top + height),
    (left , top + height),
    (left, top)

)
draw.line(points, fill='#00d400', width=2)

image.show()

Left: 217
Top: 121
Face Width: 235
Face Height: 328

0: for face in response['FaceMatches']: face_match = face['Face'] box = face_match['BoundingBox'] left = imgWidth * box['Left'] top = imgHeight * box['Top'] width = imgWidth * box['Width'] height = imgHeight * box['Height'] print('FaceMatches') print('Left: ' + '{0:.0f}'.format(left)) print('Top: ' + '{0:.0f}'.format(top)) print('Face Width: ' + "{0:.0f}".format(width)) print('Face Height: ' + "{0:.0f}".format(height)) points = ( (left,top), (left + width, top), (left + width, top + height), (left , top + height), (left, top) ) draw.line(points, fill='#00d400', width=2) image.show()">

# Analisis imagen target
s3_object = s3_resource.Object(BUCKET,IMG_TARGET)
s3_response = s3_object.get()
stream = io.BytesIO(s3_response['Body'].read())
image=Image.open(stream)
imgWidth, imgHeight = image.size  
draw = ImageDraw.Draw(image)
if len(response['UnmatchedFaces']) > 0:
    for face in response['UnmatchedFaces']:
        box = face['BoundingBox']
        left = imgWidth * box['Left']
        top = imgHeight * box['Top']
        width = imgWidth * box['Width']
        height = imgHeight * box['Height']
        print('UnmatchedFaces')
        print('Left: ' + '{0:.0f}'.format(left))
        print('Top: ' + '{0:.0f}'.format(top))
        print('Face Width: ' + "{0:.0f}".format(width))
        print('Face Height: ' + "{0:.0f}".format(height))

        points = (
            (left,top),
            (left + width, top),
            (left + width, top + height),
            (left , top + height),
            (left, top)

        )
        draw.line(points, fill='#ff0000', width=2)
        
if len(response['FaceMatches']) > 0:
    for face in response['FaceMatches']:
        face_match = face['Face']
        box = face_match['BoundingBox']
        left = imgWidth * box['Left']
        top = imgHeight * box['Top']
        width = imgWidth * box['Width']
        height = imgHeight * box['Height']
        print('FaceMatches')
        print('Left: ' + '{0:.0f}'.format(left))
        print('Top: ' + '{0:.0f}'.format(top))
        print('Face Width: ' + "{0:.0f}".format(width))
        print('Face Height: ' + "{0:.0f}".format(height))

        points = (
            (left,top),
            (left + width, top),
            (left + width, top + height),
            (left , top + height),
            (left, top)

        )
        draw.line(points, fill='#00d400', width=2)        
image.show()

FaceMatches
Left: 671
Top: 1553
Face Width: 749
Face Height: 1008

This repository compare a selfie with images from identity documents and response if the selfie match.

Related tags

Overview

aws-rekognition-facecompare

Set up:

Code Explanation

Case of use

Owner

Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation

performing moving objects segmentation using image processing techniques with opencv and numpy

This is a Deep Leaning API for classifying emotions from human face and human audios.

PyTorch implementation of Super SloMo by Jiang et al.

The devkit of the nuScenes dataset.

A High-Performance Distributed Library for Large-Scale Bundle Adjustment

This is the pytorch implementation of the paper - Axiomatic Attribution for Deep Networks.

[BMVC2021] The official implementation of "DomainMix: Learning Generalizable Person Re-Identification Without Human Annotations"

Rot-Pro: Modeling Transitivity by Projection in Knowledge Graph Embedding

3D HourGlass Networks for Human Pose Estimation Through Videos

source code of Adversarial Feedback Loop Paper

In this project we investigate the performance of the SetCon model on realistic video footage. Therefore, we implemented the model in PyTorch and tested the model on two example videos.

Dealing With Misspecification In Fixed-Confidence Linear Top-m Identification

[CVPR2022] Representation Compensation Networks for Continual Semantic Segmentation

Pytorch implementation of One-Shot Affordance Detection

This repository contains the database and code used in the paper Embedding Arithmetic for Text-driven Image Transformation

Keras-1D-ACGAN-Data-Augmentation

Unofficial implementation of Pix2SEQ

Step by Step on how to create an vision recognition model using LOBE.ai, export the model and run the model in an Azure Function

Codes and pretrained weights for winning submission of 2021 Brain Tumor Segmentation (BraTS) Challenge