Real Time Face Matching With Document Image Using Transfer Learning
In our project, we use an automated system to verify or check our identity and other details by showing our ID documents containing face images such as passports, driving licence and other documents to human operators. Then the user must show their person’s face image and will detect the 21 facial landmarks by using SDM algorithm. Once it has done, we will verify the real time face with document image by matching them using CNN algorithm. Previously, they have used a pair of sibling network to learn domain specific parameters from heterogeneous face pairs. But the cross validation is done only by CNN based general face matcher. The dataset used to train the system also require more space for storage and in case of any errors, manual checking process is the only solution. To get the better solution, we put forward a technique called Dynamic Weight Imprinting (DWI) to renovate the differentiator weights which permit quick merging and more extraditable portrayal. And a pair of sibling network with partly shared parameters are trained to learn a unified face portrayal with domain specific parameters.