Enhanced Face Identification Performance Using Online Mining Strategy in Multi-Task Cascaded Mask Convolutional Networks

作     者：Mony, Krishnaraj Raj, Jeberson Retna 

作者机构：Sathyabama Inst Sci & Technol Sch Comp Dept Comp Sci & Engn Chennai 600119 Tamil Nadu India 

出 版 物：《TRAITEMENT DU SIGNAL》 (Trait. Signal)

年 卷 期：2025年第42卷第1期

页      面：143-152页

核心收录：

学科分类：0808[工学-电气工程] 0809[工学-电子科学与技术（可授工学、理学学位）] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：We would like to express our sincere gratitude to the Department of Computer Science and Engineering  School of Computing  Sathyabama Institute of Science and Technology  Chennai  Tamil Nadu  for their invaluable support and resources that greatly contributed to the success of this research 

主　　题：Age-Invariant Face Recognition (AIFR) cross-age image collection Deep learning face detection hard sample mining task Cascaded Mask Convolutional Network (MTCMCN) real-time performance symmetry and occlusion handling 

摘      要：Due to the variety of lighting, postures, and occlusions, symmetry of faces and identification in an unrestricted area are difficult. The latest study demonstrates that deep learning techniques can do remarkably well on these two challenges. The complex transmitted multitask structure the developers provide in this research takes advantage of the natural relationship between them to improve efficiency. The suggested Multi-task Cascaded Mask Convolutional Network (MTCMCN) has three layers of carefully planned deep convolution networks that work together to figure out where faces and landmarks are from a wide range of angles. Additionally, they provide a novel, continuous, difficult sample mining approach for learning procedures, which may automatically boost efficiency without the manual choice of samples. The use of a sizable cross-age image collection containing gender and age descriptors advances the creation of Age-Invariant Face Recognition (AIFR) and FAS. MTCMCN outperforms existing methods by achieving state-of-the-art accuracy on benchmarks like FDDB and WIDER FACE, exceeding 95% accuracy in some cases. It has a Central Processing Unit (CPU) speed of 16 frames per second and a GPU speed of 99 frames per second, ensuring real-time performance. The proposed system achieves this by using a special identification conditional block and live hard sample mining, thereby improving face recognition regardless of age.