IRIS Research Assistant

Hong Chang
Ph.D. Student

Multispectral imaging is a technique that provides images of a scene at multiple wavelengths and can generate precise optical spectra at every pixel.  Multispectral images are typically acquired by specialized systems. We proposed using multispectral imaging to improve face recognition performance. Multispectral face imaging system was first built which consists of the VariSpec Liquid Crystal Tunable Filter (LCTF), a black/white video camera and video grabber. With this imaging system, we can customize number of channels and  wavelengths of channels for different applications by programming LCTF filter in a precise and robust manner. After image acquisition, we proposed several image fusion algorithms with the illumination information. These fusion algorithms are tested by a well-known recognition engine- FaceIt. The results show that multispectral images outperform conventional face images under varying illumination. Furthermore, we are studying the spectral band images and have proposed a band range selection algorithm to reduce the number of acquired bands and still obtain better performance than conventional images.

To-be-blended left side image

To-be-blend right side image

Image blending and image quality measure

 Image blending is the second procedure of Image Mosaicking and there are several methods (e.g. average blending,  weighted blending, gradient stitching Laplacian  pyramid and wavelet blending). Some of the methods show better smoothness and sharpness than others. Which method should be chosen depends on the purpose of mosaicking, such as inspection or visualization. In addition, several image quality measurements are involved in this research.

.         

stacking                                             weighted blending

Color Constancy for Image Mosiacking

Color Constancy has been used in many applications. In our research, we focus on the algorithms of color constancy which can reduce the color mismatch for Image Mosaicking.