Adobe Systems Inc.
From Images to Insights
With the success of digital photography during the past few years we have witnessed a revolution in the way photographs and videos are captured and processed. Today, our ability to acquire images and video far outstrips our ability to make sense of that data. This is true not only in personal and commercial applications but also in the sciences, where huge amounts of image data are acquired from scanners, microscopes, telescopes, and various other instruments.
We desperately need better abstractions that can improve our ability to gain insight from large collections of image data. I argue that proper data analysis can transform the data into a meaningful and perceptually intuitive representation. However, conceiving the right representation is not straightforward and it can benefit greatly from appropriate data visualization and human involvement. Fortunately, once the right abstraction is found it leads to a better and simpler acquisition method. To summarize, the whole process often involves a complex interplay among data acquisition, data visualization, and data representation.
In this talk I will discuss a number of data-driven hierarchical representations that tame the complexity of high-dimensional visual data. First, I will address the representation of spatially-varying appearance using a tree-structured factorization method and a new matrix decomposition algorithm. Then I will show how to generalize these ideas to decompose time-lapse video into simple and intuitive components that can be edited. Finally, I will discuss the MERL face-scanning project, where we collected a database of over 400 subjects with thousands of images each in order to build high-quality statistical models of human faces.
Friday, September 21, 2007
JEC 3117 - 4:00 p.m. to 5:00 p.m.
Refreshments at 3:30 p.m.