Face Age Recognition

Experiments were performed on the publicly available FGNET. The lack of a large face aging database until recently, limited research on age invariant face recognition. There are two desired attributes of a face aging database: (i) large number of subjects, and (ii) large number of face images per subject captured at many different ages. In addition, it is desired that these images should not have large variations in pose, expression, and illumination. The MORPH dataset has a large number of subjects while FGNET database has a large number of images. MORPH dataset contains about 55,000 face images from 13,000 different people. The FGNET database contains 1002 color and gray face images of 82 persons across a range of different ethnicities. There is a large variation in lighting, expression and pose across the different images. The image size is 300 x 400 in pixel units, on the average. The ages vary from 0 to 69 years. There are on the average, 12 images per person across different ages.


For this experiment we distribute the FG-NET Database into three age subspaces as 0-15, 16-30 and 31 - 69. We take a sample size that has a time length of 10 divisions. We then choose 12 samples from each subspace. Thus we take 120 images from subspace 1 and assign them as class 1. Similarly, we choose 120 images from subspace 2 and assign them as class 2 and do the same for subspace 3. For each of the images we take 7 Anthropometric features. Each sample provides a 10x7 matrix i.e. 10 images, with each having 7 features. In all, there are 36 samples. The results of this experiment is given below.

MeasureNeuCubeMLPSVM Naïve Baye