Research And Implementation Of Facial Image Age Estimation

Human facial age estimation has attracted much attention in the field of pattern recognition and computer vision in recent decades,due to its potential applications in human computer interaction,surveillance monitoring and entertainment.The study of automatic age estimation means that the computer can automatically estimate the exact age or the age range according to the input facial image.Although researchers have made great efforts to solve the problem,they still face great difficulties and challenges.On the one hand,human facial age estimation is affected by gender,race,lifestyle and other external factors.On the other hand,the existing age estimation methods can not achieve high accuracy with the lack of labeled training data.Therefore,we need further to make continuous efforts and attention for age estimation.Exiting approaches usually solve this problem from two sides,which are feature and estimation method.Focusing on the insufficiency of labeled facial images,we deeply analyze the characteristics of the age estimation problem which is different from the general classification or regression problem.And we present a novel age estimation algorithm by taking ordinal information among all age labels in consideration.(1)Age estimation based on multi-label ranking.Each facial image is labeled by multiple age tags by replacing the single age label.At the same time,the proposed method aggregates the prediction models for different age labels into a matrix,and casts age label ranking into a matrix recovery problem.The proposed method considers the ordinal relationship among all age labels.It also introduces the matrix trace norm to explicitly control the model complexity so that a reliable prediction model can be learned for age label ranking.(2)Age estimation based on structured sparsity.To exploit the intrinsic property of ordinal relationships,the learning problem is formulated as a structured sparse model regularized by the structured sparsity.The structured sparse regularization encodes the ordinal relationship among different age labels,and ensures that the samples with similar age labels are close to each other in the feature space.In addition,we introduce the grouped sparsity which constrains all age labels to share a common subspace.In order to put the algorithm into application,we build an real-time automatic age estimation system using C++ language and OpenCV library.Taking the advantage of the ordinal relationship among all age labels,both the algorithms have achieved good experimental performance on public datasets,such as the FG-NET and Refined-MORPH datasets.And the age estimation algorithm based on structured sparsity performs better than the algorithm based on multi-label ranking.