A Face Recognition under Varying Illumination - Advances in Image and Graphics Technologies - page 127

the illumination from different angles, moreover, the background may cause error rec-
ognition. In contrast, the processed images, regardless of the retinex or our method, can
solve the problem of illumination and useless background. Compare the 2th and 3th
lines, the images in 2th line are much white than those in 3th line. Though distinct edge
isn't affected by illumination, it can't present enough important basic information which
distinguishes some people with similar edge. Consequently, our method adds some low
frequency component to invariant to add additional information to improve accuracy.
Table 1 presents the recognition accuracies of three conditions. In order to verify the
effectiveness of our method, we employ the recognition rate of original condition as a
standard. Obviously, the accuracy of our method is higher than that of the standard in
table 1. Compare our method and the retinex, the accuracy of our method is 4% higher
than that using retinex. The reason why our method is better is that an appropriate low
frequency component adds useful information to reflectance to make the images dis-
tinguishing easy.

Table 1. Comparison of different methods

Methods Accuracy

Original image 82%

Retinex 90%

Our method 94%

5 Conclusion

Illumination, pose and expression are the three important factors to cause low recog-
nition accuracy in face recognition. This essay proposes a method based on retinex and
the wavelet transformation to the problem of face recognition under the illumination
condition. First, the illumination invariant and variant are generated by retinex theory.
Second, the reflectance and illumination are rotated and segmented to identify the
effective fields according to eye positions. Third, decompose the illumination com-
ponent via the wavelet transformation and set its low-frequency coefficients to zero.
Therefore, the processed illumination component is obtained by the inverse transfor-
mation. Last, restructure the two components to a new image. The experiment de-
monstrates that our method performs well. In the future, we will prepare to employ this
method to the image acquired from camera in real time, this will make our method well
applicable to the public environment.

References

1. Wei, C.C., Wang, X.P., Yan, J.W., et al.: Method for eliminating the Illumination Effects of
Face Recognition ( 一种消除光照影响的人脸识别方法 ). Electronic Test 7, 19-23 (2012)