Open AccessExtracting Facial Features in Filter Modified Images using Neural Networks
Author(s) -
M. G. Sumithra,
M. Shyamala Devi,
A. Poornima,
Shatadru Sarkar,
Harish Patel,
Preethi Mahankali,
B.J. Navya
Publication year - 2021
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/1074/1/012034
Subject(s) - artificial intelligence , computer science , pattern recognition (psychology) , filter (signal processing) , face (sociological concept) , curse of dimensionality , feature (linguistics) , feature extraction , feature selection , noise (video) , computer vision , image (mathematics) , social science , linguistics , philosophy , sociology
Facial identification plays a vital role in this digital era. Recent advancements in technology provide various filtering effects while capturing the images. Extracting the facial features from the filter modified image becomes challenging. To identify the important facial attribute values from those images, special features extraction techniques needs to be applied. Facial points are the key factors to recognize the face. When these factors are altered, facial detection becomes crucial. Initially apply the image pre-processing techniques to reduce the noise level in the images then perform feature selection. To overcome the curse of dimensionality, Occam’s razor approach is followed to limit the selection of attributes. In different stages the features are going to be filtered by replacing the weaker attributes. Later Pearson Correlation, a filter based correlation technique is applied to evaluate the deviated numerical value from the target. The deviated features can be treated with Lasso effect to suppress the poor weighted features. Finally through recursive feature elimination, highest influencing attribute factor can be identified. In this approach some of the facial feature which is used in training the neural network can be discarded based on the deviation rates. In terms of accuracy the proposed system can recognize the dataset with 70% of accuracy with filter effected images.