The paper titled “Rank Pooling for Image set based Face Recognition”  authored by Harinarayanan KK , Principal Research Scientist, postgraduate from IISc Bangalore and Pramod YN ,Research & Development Engineer, postgraduate from IIIT Hyderabad has been accepted at the International Conference on Intelligent Systems and Green Technology (ICISGT-2019) to be held at Andhra University in Visakhapatnam.

ICISGT-2019, organized by IEEE Hyderabad Section and IEEE Vizag Bay sub Section, aims to bring together leading academicians, scientists, researchers and research scholars to exchange and share their experiences and research results on all aspects of related to Intelligent Systems and Green Technologies. It provides an interdisciplinary platform for researchers, practitioners, and educators to present and discuss the recent innovations, trends, and concerns, as well as practical and theoretical challenges encountered and solutions adopted in these fields. It also aims to showcase the benefits to humanity with these Intelligent Systems and Green Technologies.

The following is the abstract of the paper.

Rank Pooling for Image set based Face Recognition:

Single image face recognition is limited by many factors like the pose of the person, illumination, occlusions induced by wearables, hair growth, etc. In many situations, these constraints are imposed because the recognition is performed independently for each image. As an illustrative example, in multi-view surveillance systems the captured faces have low resolution and bad illumination more often than not. However considering all the frames captured across time and cameras, we have a favorable number of face images with a variety of poses and illumination. By considering these images as a set for recognition, we will be able to infer information specific to the identity of the person in cases where pose and illumination variance is common.

It is important to note that there is a difference in how Face Verification and Face Recognition work. Face Verification is basically an Image Similarity problem wherein, two candidate face images of a person is compared to verify if it identifies to the same person in question. Essentially it is a one-one matching problem. In contrast, Face recognition is one-many matching problem wherein from a dictionary of people, the process of face verification has to be performed with each of them to assign the identity of the person. Hence the dictionary size has a lot of impact on the final processing time for face identification in case of Face Recognition problem. Essentially Face Verification is a subset of the bigger Face Recognition problem.

The paper proposes a simple Rank pooling algorithm for Image Set based Face Recognition (RPFR) which builds on metric learning approach to map the face images/frames to an embedding space were based on nearest neighbor pooling in the feature space, the person can be identified.

The conference is being held on 29th -30th June 2019 at YVS Murthy Auditorium, AU North Campus, Visakhapatnam.

Harinarayanan KK (aka Hari ) majorly works on AI for Cancer Detection, one of the most important and impactful problems among AI solutions to Medical image processing. Pramod YN works on developing Computer vision solutions to automate critical processes affecting the Industrial and Defence sector at large.