关键词: 年龄估计, 年龄编码, 有序回归, 排序, 多标签分类

Abstract: The traditional age estimation task based on ranking and regression methods cannot effectively utilize the evolutionary characteristics of human faces and build correlations between different ranking labels. Moreover, multi-class classification methods for age prediction inevitably lead to inconsistencies in the ranking. Based on the above problems, an age estimation method based on integrated ranking matrix encoding and consistency preservation was proposed to fully utilize the correlation between age and ranking values and suppress the problem of inconsistent ranking. A new indicator, the proportion of samples with inconsistent rankings, was proposed to measure the problem of inconsistent rankings in the two-class ranking method. First, age categories were converted into a sort matrix form through a designed ranking coding method. Then, the ResNet34 (Residual Network) feature extraction network is used to extract facial features, which are then learned through the proposed encoding learning module; Finally, the network prediction results are decoded into the predicted age of the image through a ranking decoder based on a metric method. The experimental results on the Morph II dataset achieved a Mean Absolute Error (MAE) of 2.18, and have better results on other publicly available datasets compared to methods such as OR-CNN(Ordinal regression with multiple output CNN for age estimation) and CORAL(Rank consistent ordinal regression for neural networks with application to age estimation), which are also based on ranking and ordinal regression methods. At the same time, the proportion of samples with inconsistent ranking is suppressed, and the performance of ranking inconsistency measurement is improved by about 65% compared to the OR-CNN method.

Key words: age estimation, age encoding, ordinal regression, ranking, multi label classification