Ipapfa-Gan: Identity-Preserved And Attention-Based Progressive Face Aging Using Generative Adversarial Network

Abstract

Many Studies On Face Aging Have Been Conducted, Ranging From Approaches That Use Pure Image Processing Algorithms To Those That Use Generative Adversarial Networks. It Is Preferable When Computationally Aging A Face That The Age Output Is Close To The Expected Age And That The Individual's Characteristics Are Preserved. Conventionally, Two Types Of Modelling Techniques Have Been Used In This Task: Prototype-Based And Model-Based Methods. When Transforming A Face From A Younger Domain To An Aged Domain, Both Approaches Fail To Retain Individual Characteristics. With Advancements In Computer Vision, Generative Models, Particularly Generative Adversarial Networks, Have Been Used To Perform This Task (Gans). It Is Now Possible To Generate Realistically Aged Faces Of Specific Individuals Using Them. However, These Methods Cannot Meet The Three Essential Requirements Of Face Aging Simultaneously And Usually Generate Aged Faces With Strong Ghost Artefacts When The Age Gap Becomes Large. So, Identity-Preserved And Attention-Based Progressive Face Aging Using Generative Adversarial Network (IPAPFA GAN) Is Proposed To Mitigate These Issues. The Proposed Architecture Uses Self-Attention GAN To Maintain Identity-Related Information And Improve Quality Of The Generated Images. And Also, Pixel-Wise Loss Replaced With Attention Loss To Alleviate Accumulative Blurriness. Further, Least Square GAN Employed For The Discriminator To Improve The Quality Of The Synthesized Images And Stabilize Training Process. Furthermore, Quantitative Experiments Validate The Effectiveness Of Our Approach. The Inception Score Of The Proposed Solution Has Shown Improvements In Generating The Desired Images With Better Quality With The Score Of 34.23. The Pearson Correlation Coefficient Of The Proposed Solution Also Has Shown Improvement In Generating Images With Better Aging Accuracy And Smoothness With The Score Of 0.993. Finally, The Verification Confidence Score Of The Proposed Solution Has Shown Improvements In Identity Preservation Between Input Faces And Generated Face Images With The Score Of 97.03, 95.12, And 90.42 Respectively From Three Age Groups. Comprehensive Experiments Demonstrate Superior Performance Of The Proposed Solution Over The Existing State-Of-The-Art Methods On CACD Benchmarked Dataset With Attention Loss, Least Square GAN, And Self-Attention Mechanism.