Abstract:
Research related to age estimation using face images has become increasingly important due to its potential use in various applications such as advertising and access control. Although there are several methods proposed in the state of the art of age estimation, it is still challenging task due to the several factors and variation that affect on the face age estimation process.
The primary aim of this thesis is to develop a robust age estimation method based on wrinkle and other local features. First, a robust wrinkle detection method was proposed, which localize wrinkles regions using simple and powerful matched filter. Then adaptive thresholding technique was introduced for the first time to segment wrinkles regions. Experimental results revealed that the adaptive thresholding technique significantly increases the accuracy of the proposed algorithm when evaluated on FERET, Social Habits and our proposed in-house Sudanese dataset.
Second, features representation method called Wrinkle Local-based Descriptor (WLD) was proposed to represent face aging features based on Scattering Transform and wrinkle pattern. Three benchmarks datasets namely FG-NET, FERET and PAL were used to evaluate the proposed method, in addition to our collected Sudanese dataset. Results showed that WLD have a great potential in age estimation, when evaluated on high and mid resolution images. Where the combination of Scattering Transform(ST) and wrinkle pattern yielded a superior result on FERET and PAL with MAE 2.09 and 2.15 years respectively, and 4.72 years for Sudanese dataset. Finally, The effect of smoking on wrinkle was investigated using computer vision algorithms. The result showed that the density of wrinkles for smokers in two regions around the mouth was significantly higher than the non-smokers, at p-value of 0.05.
