2014-04-25 EECS 556 abstracts / schedule, 4-6pm 1005 EECS

All students need to be present for all presentations, unless prior
arrangements were made with Prof. Fessler, to be able to earn full
credit for the oral presentation score and to be eligible for any
project prizes.
Be sure to test your laptop in the room before the presentation day
to iron out any technical issues, to be able to earn full credit
for oral presentations.


4:00-4:20
Image classification by non-negative sparse coding, low-rank and sparse decomposition
Aniket Anand Deshmukh, Naveen Murthy

4:20-4:40
Gender recognition based on SIFT features
Han Lin, Tianrui Luo, Chengcan Ye

4:40-5:00
MRI bias field correction based on tissue labeling
Jiyang Chu, Jie Li, Lianli Liu, Zhen Zeng
:
Intensity inhomogeneity, also known as "bias field" in MR images, mainly
arises from imperfection in RF profiles, and it harms the performance of many
image analysis algorithms.  Traditional Segmentation based bias field
correction methods is constrained by its low intensity based segmentation
accuracy.  To improve segmentation accuracy, we propose to integrate rich-
feature segmentation in bias field estimation.  A Laplacian regularization
scheme is also designed to encourage the smoothness of estimated bias field.
We use synthetic data from BrainWeb and augmented them with artificial bias
field.  Experiment on the dataset shows that our method provides more robust
and accurate segmentation result, which results in better bias field estimation.


5:00-5:20
Improving the accuracy of volumetric image registration using ranked order statistics and importance sampling
Kenny Cha, Kathleen Ropella, Victor Wu
:
Image registration is important in the field of medical imaging for multi-
modality fusion and investigating structures during longitudinal studies.
Image registration is the process of spatially aligning two images - a fixed
image and a moving image - based on some similarity measure.  The
transformation used to align the images can be classified as either rigid or
non-rigid.  Rigid registration uses a single set of transform parameters
applied to every pixel whereas non-rigid registration uses transform
parameters that are pixel-specific.  We focus on non-rigid registration of
intra-modality images.  Determining the transform parameters can be formulated
as an optimization problem where the cost function is based on the intensity
differences between the fixed image and the transformed moving image.  This
problem can be solved by the gradient descent algorithm, which often requires
a large amount of computation.  To reduce this computation, we calculate a
stochastic approximation of the gradient by sampling the image based on the
modified selective statistical estimator (MSSE) and importance sampling.  This
combination encourages sampling from a subset of the population that is likely
to determine an improving direction in the optimization algorithm.  We test
this method on 2D test images and coronal slices of lung CT images during
inhalation and exhalation.


5:20-5:40
Unified blind method for multi-image super-resolution and single/multi-image blur deconvolution
Abhishek Bafna, Tatyana Dobreva, Yaohui Li, Rebecca Malinas

5:40-6:00
Single image local defocus blur kernel estimation
Cody Hyman, Paul Schroeder