Associate Professor, Electrical & Computer Engineering
Donadeo ICE 11-380
Imaging is critical to intelligent activity in various sectors of the economy. To enable and simplify the impossible and difficult visual tasks, progress is necessary from image sensors to image understanding. By advancing the former via electronic engineering and the latter via computer science, Dr. Joseph's lab pursues excellence in imaging science and technology (IS&T), a multi-disciplinary research area, through the lens of computer engineering.
In September 2016, the IS&T lab published a paper in the Journal of Computational Science on a new tensor computing framework, based on Einstein notation and accompanied by C++ and MATLAB libraries . The paper unifies and advances a growing body of work on high-dimensional algebra and software for technical computing. Although the IS&T lab was motivated by specific problems in computer vision, the paper illustrates benefits of the framework to canonical polyadic decomposition, a high-dimensional variant of singular value decomposition.
In March 2016, the IS&T lab published a paper in the Journal of Imaging Science and Technology on a new tone mapping operator. Tone mapping is essential for viewing images from high/wide dynamic range cameras on standard display devices. The novel operator, based on histogram adjustment, uses a model of the camera noise to ensure that the tone mapping does not amplify the noise above a display threshold. The novel operator and its fixed-point design are validated through offline and real-time experiments with a logarithmic CMOS image sensor.
In October 2015, the IS&T lab published a paper in Sensors, an open-access journal ranked ahead of the IEEE Sensors Journal, on fixed pattern noise and photometric correction of CMOS image sensors . Using low-degree polynomials, the proposed methods may be implemented efficiently for nonlinear, as well as linear, pixels. Nonlinear pixels are interesting because they can capture a high/wide dynamic range of light at very high frame rates. The methods were demonstrated experimentally with a logarithmic CMOS image sensor.