A Dynamic Approach to Face Recognition


With the recent emphasis on homeland security, there is an increased interest in accurate and non-invasive techniques for face recognition. Most of the current techniques perform a structural analysis of facial features from still images. Recently, video-based techniques have also been developed but they suffer from low image-quality. In this paper, we propose a new method for face recognition, which is based on dynamic instead of static facial features. We track the motion of certain features on the face during a facial expression and obtain a vector field that characterizes the deformation of the face. Since it is almost impossible to imitate another personÕs facial expressions these deformation fields are bound to be unique to an individual. To test the performance of our method in face recognition scenarios, we have conducted experiments where we presented individuals wearing heavy make-up as disguise to our matching framework. The results show superior face recognition performance when compared to the popular PCA+ LDA method, which is based on still images.


Our Method

In the training process, we obtain two high-resolution images of an individual, one with a neutral expression and the other with a facial expression, like a subtle smile. We then compute the deformation field from these two images. We separate each deformation field into two images corresponding to the projections of the vector field on X- and Y- axes. We call these images Projection Images. During the testing phase, we compare the Projection Images of the subject with those stored in our database and find the closest match.

Figure 1. Some of the subjects in our database. The projection images have been color-mapped so as to emphasize the variations


The person in fig 1.1 is the person in fig 1.2 with make-up. The person in fig 1.4 is the person in fig 1.5 with make-up. Though these two subjects are wearing make-up, we can see that their Projection Images are quite similar and easily distinguishable from those of other subjects.


Figure 2. Two projection images (right) of identical twins (left)


The subjects in fig 2 are identical twins, but we can see a marked difference between their Projection Images. We can notice that the first twin tends to exercise the left side of his face more when smiling while this is not the case with the second twin.

Results of Comparison with PCA+LDA

In order to validate the effectiveness of our approach, we compared the ability of our approach to recognize people with make-up it with that of PCA+LDA, which is a popular still image-based face recognition method. Table 1 shows the results we obtained with PCA+LDA. We can see that PCA+LDA failed to correctly recognize the query subject when the subject was wearing make-up. Table 2 shows the results we obtained with our approach. As we can see, our approach correctly recognized the query subject in both the cases.


Table 1. Distances obtained with the PCA+LDA algorithm

Query Image Target Image Best Match 2nd Best Match 3rd Best Match
Distance 0.83 0.32 0.45 0.73
Query Image Target Image Best Match 2nd Best Match 3rd Best Match
Distance 0.56 0.47 0.56 0.925


Table 2. Distances obtained with our DISC algorithm. Note, comparisons were done on the projection images and not the images shown here

Query person Target person Best Match 2nd Best Match 3rd best Match
Distances 0.077 0.077 0.709 0.797
Query person Target person Best Match 2nd Best Match 3rd Best Match
Distances 0.088 0.088 0.709 0.789


An important feature of our method is that it is completely based on affordable mainstream technologies. The only hardware we require is a high-resolution digital camera. In our experiments, we have observed that a digital camera with resolution greater than 4 mega pixels is sufficient to capture images that enable us to accurately track the skin pores on an individualÕs face. Nowadays, cameras with such resolutions are commonplace. Some newer cell phone cameras have resolutions as high as 5 mega pixels. The rapid explosion in digital camera technology means that images can now be acquired at hitherto unheard of detail.

Currently we are working on making our method pose-invariant. We are also exploring the use of higher-level analysis methods for characterizing the deformation fields.