Age Estimation Based on Complexity-Aware Features
The research related to age estimation using face images has become increasingly important. We propose an age estimator using two kinds of local features, the gradient features which well describe the local characteristic, and the Gabor wavelets which ref
- PDF / 2,533,151 Bytes
- 14 Pages / 439.37 x 666.142 pts Page_size
- 64 Downloads / 200 Views
Abstract. The research related to age estimation using face images has become increasingly important. We propose an age estimator using two kinds of local features, the gradient features which well describe the local characteristic, and the Gabor wavelets which reflect the multi-scale directional information. The RealAdaBoost algorithm with a complexity penalty term in the feature selection module is applied to choose meaningful regions from human face for feature extraction, while balancing the discriminative capability and the computation cost at the same time. Furthermore, the hierarchical classifier, which is composed of an age group classification (e.g., 15–39 years old, 40–59 years old etc.) and a detailed age estimation (e.g. 19, 53 years old, etc.) are utilized to get the final age. Experimental results show that the proposed approach outperforms the methods using single feature on PAL and FG-NET database. It also achieves competitive accuracy with the state-of-the-art algorithms.
1
Introduction
Systems based pattern recognition have been proven to be very useful in many areas such as security and access control, human detection, human computer interaction, and brain computer interface. Recently, the research related to age estimation using face images is more important than ever. The potential applications include automatic guest enrollment, parent TV program control, video surveillance, etc. In general, age estimation systems consist of two steps: feature extraction and classification/regression [25]. The features used in age estimation can be categorized into the local features and the global features. The local features are extracted on some regions which might contain specific facial characteristic, such as wrinkles and freckles. They have been used to classify people into age groups. Conversely, the global features are extracted based on whole face shape or all facial feature points. They are generally used to estimate the exact age. Some researchers also use hybrid features to improve the estimation accuracy, which is the combination of local features and global features. After feature extraction, the classification/regression module is utilized to train the age estimator. The commonly-used algorithms include the age group c Springer International Publishing Switzerland 2015 D. Cremers et al. (Eds.): ACCV 2014, Part I, LNCS 9003, pp. 115–128, 2015. DOI: 10.1007/978-3-319-16865-4 8
116
H. Ren and Z.-N. Li
classification, single-level estimation and the hierarchical age estimation. Age group classification is an approach that roughly predicts an age group, whereas single-level method focuses on detailed age prediction. The hierarchical method is a coarse-to-fine method which integrates the single-level and age group methods together. Regarding the efficiency issue, local features based methods perform better compared to global features based methods utilizing ASM [1] or AAM [2]. Unfortunately, the use of the local features for age estimation has not been well investigated. The methods extracting a dense f
Data Loading...
