Research On Age Estimation Of Face Images Based On LSTM Fine-Grained Classification

Facial appearance is one of the most important visual features,which will change with age,so automatic age estimation is a challenging research topic in the field of facial feature analysis.Facial feature extraction is one of an essential factor in the task of age estimation,in addition to extracting the global features based on facial facial images,this thesis also takes into account the fine-grained features of age-sensitive area,and proposes a method based on Attention LSTM network for Fine-Grained age estimation in the wild based on the idea of Fine-Grained categories and visual attention mechanism.This method combines residual networks(ResNets)or residual networks of residual networks(RoR)models with LSTM unit to construct AL-ResNets or AL-RoR networks to extract age-sensitive local regions,which effectively improves age estimation accuracy.Firstly,deep neural networks have achieved great success in many different fields,where many models have been developed.The innovation points of several classical structures are analyzed in this thesis,and the principles of the recurrent neural network are also described in detail.Secondly,this thesis analyzes the feasibility of the fine-grained age estimation method and puts forward the overall framework of the method.ResNets and RoR models pre-trained on ImageNet,IMDB-WIKI-101,and target age datasets are chosen as the basic models;Based on pre-trained ResNets or RoR,LSTM is inserted into the last residual block to build AL-ResNets and AL-RoR models;The newly constructed network is used for age-sensitive area training,and the training method is mainly composed of four modules.Finally,the evaluation protocol is introduced;Adience,LAP,MORPH Album 2 and FG-NET are selected as target age datasets,and performed corresponding preprocessing methods on the different data sets.By introducing oversampling methods,the age group classification experiment is performed directly on the Adience dataset.Age value estimation experiments are performed by the Deep EXpectation algorithm(DEX)on MORPH Album 2,FG-NET and LAP datasets.Experiments illustrate the effectiveness of AL-ResNets and AL-RoR for age estimation,where it achieves new state-of-the-art performance than all other CNN methods on the four age datasets.