Searching images with images: characterization, retrieval and ranking for multi-modal imaging
Daniela Ushizima, Ph.D.
Data Scientist, Lawrence Berkeley National Laboratory
Imagine inspecting 1 terabyte of images in one second. Forget binge watching, this means 200 movies in a blink of the eye. This amount of images overwhelms scientists at the national laboratories, who depend on experimental data. But how to visually examine that many pictures? What we need is to construct computer programs that recognize relevant patterns from digital images. That is what I have done for the past 20 years: I use computer vision and machine learning techniques to enable people to use pictures in decision-making. I construct algorithms applied to material science, aiming to control the quality of compounds and, for example, improve durability of aircraft. Another project is to design computational tools for cell microstructure description and interactions, for example, to enable neurologists to detect cells and functions in cellular communities.