Elkarazle, K.; Raman, V.; Then, P. Facial Age Estimation Using
Download 0.59 Mb. Pdf ko'rish
|
BDCC-06-00128
Age manifold
: Age manifold focuses on treating ageing patterns as a trend for several subjects at various ages instead of finding a specific ageing pattern for each person. The flex- ibility of the ageing manifold method allows for subject representation to be in the form of Big Data Cogn. Comput. 2022, 6, 128 8 of 22 one image or several images at different ages. Comparing the age manifold to a close equiv- alent (AGES), we find that models based on this method can learn low-dimensional ageing patterns that AGES could ignore. The low subspace is defined using conformal embedding analysis (CEA) [ 39 ], which obtains features and reduces dimensionality through discrimi- nating analysis and conformal mapping. Both these techniques project high-dimensional data onto a unit hypersphere. Unsupervised dimensionality reduction techniques are not effective for handling discriminators. Instead, [ 40 ] suggested an alternative to reduce dimensionality using a supervised algorithm, denoted as orthogonal locality preserving projections (OLPP), whereby age is first predicted using a regression function and is then locally adjusted to match the correct values within a boundary. 5.2. Deep Learning Models Deep learning models work differently than their manual counterparts. After acquir- ing and pre-processing the images, we feed them to a deep neural network. The network may consist of several convolutional neural network layers (CNN) [ 41 ], pooling layers, dropout layers, batch normalisation layers, or residual connections. We then define the number of filters and the size of each kernel. The layers in the network will automatically extract ageing features as the network continue processing the input images. Deep learning models are usually more accurate than their handcrafted counterparts as we allow the model to decide on the essential features to learn. One of the significant drawbacks to this is the high computational cost, as these models can become enormous and consume more computational power and time. Deep learning models can be either trained from scratch or based on pre-trained models. 1. Download 0.59 Mb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling