Facial Paralysis Dataset

Facial paralysis is the loss of facial muscle movement

either in one side or both sides of the face due to the facial

nerve damage. Currently, the subjective assessments are widely

used techniques to determine the measure of degree with which

the patient is affected. However, the subjective assessments are

highly dependant on the expert’s view and a few sets of grading

rules. In this paper, the quantitative assessment to measure the

degree of facial paralysis is proposed. The video database of

facially paralyzed patients, which consists of seven different views

and multiple subjects with ten different expressions are collected

under three experts supervision. Inorder to capture the variations

present in multiple views and subjects across all the expressions, a

large Gaussian mixture model (GMM) is trained. A feature vector

is obtained from each expression using a maximum a posteriori

adaptation (MAP). The dimension of the adapted feature vector

is very high and contains redundant attributes. So, we reduce

the dimension of the feature vector using factor analysis, which

contains pertinent attributes. The proposed work is evaluated on

the video database of 39 facially paralyzed patients of different

age groups and gender. Based on the facial paralysis effect,

experts assign subjective scores to patients using Yanagihara

grading rules, which are further used as ground truth. We also

show the efficacy of the proposed approach by measuring the

different degree of facial paralysis for all 10 types of expressions

better than existing approaches for quantitative assessment.