Faces Similarity

Workflow Faces Similarity in HGrid247-3.5

Purpose
This workflow is used to:

Calculating the level of facial similarity
Identify whether two images contain the same person
Sort results by highest facial similarity

In contrast to ordinary image similarity , here the focus is only on the facial area (face region) .

Concept Face Similarity

The main process consists of 3 stages:

Face Detection → detects faces in images
Feature Extraction → extracting facial features
Similarity Matching → comparing faces

Step 1 – Copy Workflow

From the previous workflow:

    WFImageSimilarity

Copy to:

    WFFaceSimilarity

Purpose:

reuse pipeline
just replace the similarity method

Step 2 – Transformator_2 Modification

On Transformator_2 :
Delete function:

    ImageSimilarity

Because it will be replaced with a special one for the face

Step 3 – Adding FaceSimilarity

Add function:

    FaceSimilarity

Input-output relationship:

    image1 ─┐
            ├── FaceSimilarity ──> similarity
    image2 ─┘

Important Parameters

1. Face Detector

    CLMFaceDetector

Function:

detecting face position in an image
ensure that only the facial area is processed

2. Facial Feature Extractor

    DoGSIFTFeature

Function:

extract important facial features
such as eyes, nose, mouth

3. Key Point Descriptor

    SIFT

Function:

compare facial features between images

Step 4 – Compile Workflow

Click the button:

    Generate ()

Output:

    workshop_mr.jar

Step 5 – Running on Hadoop

Run on an Apache Hadoop cluster :

    hadoop jar workshop_mr.jar workshop.WFFaceSimilarity photo output_fsim photo/photo1.jpg

Parameter

    Parameter       Description
    ------------------------------------
    photo           facial image dataset
    output_fsim     results folder
    photo1.jpg      reference face

Sample Output

    photo/photo1.jpg        photo/photo1.jpg        1.0
    photo/photo1.jpg        photo/photo5.jpg        0.7046571656156865
    photo/photo1.jpg        photo/photo7.JPG        0.6650472587046466
    photo/photo1.jpg        photo/photo2.jpg        0.3304030217553085
    photo/photo1.jpg        photo/photo4.jpg        0.3246716833058123
    photo/photo1.jpg        photo/photo3.jpg        0.3175215036337799
    photo/photo1.jpg        photo/photo6.jpg        0.032529146286689375

Interpretation of Results

    Similarity Value        Meaning
    -----------------------------------------
    1.0                     identical faces
    > 0.7                   very similar
    0.3 – 0.7               quite similar
    < 0.1                   not similar

Note : The file of workflow WFFaceSimilarity can be found in the following link : WFFaceSimilarity.csd