Fast and reliable probabilistic face embeddings based on constrained data uncertainty estimation

Chen, Kai; Yi, Taihe; Lv, Qi

doi:10.1016/j.imavis.2022.104429

Cited by 7 publications

(3 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some notable models applied in face recognition include Hidden Markov Model-based system, Eigenfaces using Principal Component Analysis, Fisher faces using Linear Discriminant Analysis, Bayesian method using Probabilistic Distance Metric, etc. The techniques are proven to be efficient in segmenting multiple human faces even with partial occlusion and complex cluttered background (Asha et al, 2022;Huang et al, 2022;Chen et al, 2022;Iqbal et al, 2011;Samaria & Young, 1994). In this strategy the color information is extracted from the video frames to identify the possible location of the human face in the image.…”

Section: Background Studymentioning

confidence: 99%

Intelligent Head-bot, towards the Development of an AI Based Cognitive Platform

Baki,

Akhtaruzzaman,

Refat

et al. 2023

MIJST

View full text Add to dashboard Cite

A cognitive humanoid head is an AI enabled head-bot platform that resembles human's cognitive abilities, such as perception, thinking, learning, and decision-making. The platform is able to interact with human through natural language processing and recognize individuals, thus allowing seamless communication between two parties. No such cognitive platform has been introduced in Bangladesh, thus creating an open field to contribute to the field of Machine Intelligence. This study aims to develop an AI based humanoid head (head-bot) capable of imitating a range of expressions, recognizing individuals, and interacts with visitors through general conversation. The head-bot skeleton is developed using a number of hexagonal blocks of PVC sheet to mimic a human-head-like structure where LCD, camera, microphones, and speaker are mounted. Two separate Machine Learning models are designed for face detection and recognitions, and voice enabled chat-bot implementation. The head-bot platform incorporates 2-DoF neck movements for various head gestures and face tracking. The Artificial Neural Network models are tested with accuracy of 95.05%, and 99.0%, for face detection and recognitions, and speech recognitions and response generation, respectively. According to the overall results and system performances, it seems that the proposed system has a number of good potentials for real life applications such as entertainment, guidance, conversations, interactive receptionists, personal companion, medical assistance, and so on.

show abstract

Section: Background Studymentioning

confidence: 99%

Intelligent Head-bot, towards the Development of an AI Based Cognitive Platform

Baki,

Akhtaruzzaman,

Refat

et al. 2023

MIJST

View full text Add to dashboard Cite

show abstract

“…Recently, probabilistic embeddings gained popularity for face recognition in unconstrained conditions where the recognition systems may heavily suffer from low-quality inputs [11,12,13,25]. For instance, the authors of [11] proposed to represent the input images by Gaussian distributions.…”

Section: Probabilistic Embeddingsmentioning

confidence: 99%

“…There also exists a concept of probabilistic embeddings, which naturally provides the quality estimates as variancerelated probability distribution parameters. Such embeddings are well studied in both face recognition [11,12,13] and speaker recognition research [14,15].…”

Section: Introductionmentioning

confidence: 99%

Magnitude-aware Probabilistic Speaker Embeddings

Kuzmin,

Fedorov,

Sholokhov

2022

Preprint

View full text Add to dashboard Cite

Recently, hyperspherical embeddings have established themselves as a dominant technique for face and voice recognition. Specifically, Euclidean space vector embeddings are learned to encode person-specific information in their direction while ignoring the magnitude. However, recent studies have shown that the magnitudes of the embeddings extracted by deep neural networks may indicate the quality of the corresponding inputs. This paper explores the properties of the magnitudes of the embeddings related to quality assessment and out-of-distribution detection. We propose a new probabilistic speaker embedding extractor using the information encoded in the embedding magnitude and leverage it in the speaker verification pipeline. We also propose several quality-aware diarization methods and incorporate the magnitudes in those. Our results indicate significant improvements over magnitude-agnostic baselines both in speaker verification and diarization tasks.

show abstract