Image Processing & Recognition
[1] Colorization
We develop a colorization algorithm for monochrome still image and video with easy interactive processing. Colorization is performed by sowing color seeds and propagating colors spatially to the remaining pixels.
[2] Advanced Retinex
The MSR (Multi-Scale Retinex) is effective for suppressing the unwanted banding artifacts around high contrast edges, but takes high computation costs. This study develops a simple algorithm for realizing the same banding-free effect based on contrast adjustment in the SSR (Single-Scale Retinex).
[3] Versatile Gamut Mapping
Gamut mapping was accomplished that was mapped the image into the printer gamut. But the observation method of the image has various devices now. This study aims to develop in both compression for wide gamut and expansion for narrow gamut adapted to various devices.
[4] Computational Vision for Color constancy
We are interested in vision systems and algorithms for realizing color constancy. A vision system may predict any information about surface reflectance. The surface-spectral reflectance of an object is an inherent physical property of its surface． Given the reflectance, an extremely wide range of problems is solved in not only realization of color constancy but also digital archive, rendering realistic color images, and accurate color reproduction. We have developed algorithms for estimating illuminant-spectral distribution of a light source and surface-spectral reflectance of an object.
[5] Iris Detection for Eyegaze Interface
The iris detection is very important for eyegaze interface by tracking human iris. This study aims at realizing an accurate iris detection from a monocular camera.
[6] Natural Scene Illuminant Estimation
Natural Scene Illuminant Estimation tries to estimate the spectrum distribution of unknown illuminant by using a spectral camera system. This study aims to estimate the complicated distribution that composed of fluorescent lamp and ambient light (incandescent lamp or daylight) ,which was difficult to estimate so far.