IRIS Research Staff

Mingzhong Yi, Masters Student

This block diagram courtesy of [Lehigh University] http://www.eecs.lehigh.edu/SPCRL/IF/image_fusion.htm

Image fusion refers to a process that extracts redundant and complementary information from a set of input images and fuses it into a single and more complete image. The fused image should be more informative and thus will be more useful for human or machine perception. The fusion of the two images can take place at the signal, pixel (Image), feature or symbol level (courtesy of [Lehigh University] http://www.eecs.lehigh.edu/SPCRL/IF/image_fusion.htm). Our research is to combine Multi-spectral image fusion technology with Face Recognition technology. Our objective is to find some satisfactory methods for Multi-Spectral Face Recognition. Now, I have already got some satisfactory Image Fusion Rules based on Wavelet transformation to fuse visible and infrared images for Face Recognition. The following is a Block diagram of a generic wavelet-based image fusion approach. In future, this approach will also be explored for Multi-Spectral Face Recognition.

Experiments are administrated to test (using FaceIt) the effectiveness of different combination of Wavelet coefficients of visible and infrared images, and thus find optimum fusion rules to fuse visible and infrared images using MATLAB Wavelet toolbox.

Using popular and authoritative graphs, such as Cumulative Match Characteristics (CMC), to show the effectiveness of the fusion rules of IRIS.

Multi-spectral data acquisition is performed using unique illumination situation and imaging process of IRIS lab such as sequential image acquisition from band 480 to 720 and to gray, and thus it's no need to do registration.

Design unique analysis tools to test (using FaceIt) the effectiveness of different bands on Face Recognition.

Future work including designing unique experiments and strategies to try to find optimum fusion rules for fusing multi-spectral images for Face Recognition.

Applications

High accurate Face Recognition: Using Multi-Spectral Image Fusion to get higher face recognition performance, such as higher Identification Rate.

Robust Face Recognition: Using Multi-Spectral Image Fusion to compensate the decrease in Face Recognition performance due to illumination change.

High confident Face Recognition: Using Multi-Spectral Image Fusion to increase the confidence of face recognition for those faked face images such as images with makeup.