Romeil Sandhu dot com

Abstract: In this project, we present a non-rigid approach to jointly solve the tasks of 2D-3D pose estimation and 2D image segmentation.  In general, most frameworks which couple both pose estimation and segmentation assume that one has the exact knowledge of the 3D object (see project below).  However, in non-ideal conditions, this assumption may be violated if only a general class to which a given shape belongs to is given (e.g., cars, boats, or planes).  Thus, we lift this restriction by proposing to solve the pose estimation and  image segmentation tasks via (non)linear manifold learning of 3D embedded for a general class of objects. Read the rest of this entry »

Abstract: In this project, we seek to minimize the gap-to-capacity (given by Shannon’s theoretical limit) of a rate 1/3 code. This is done via a convolutional encoder/decoder for varying memory elements as well for both soft and hard decoding scheme. We show that the gap-to-capacity can be minimized with respect to the suboptimal un-coded code word or a (3,1) repetition code. Although better schemes are available such as LDPC and turbo codes, we have chosen the convolutional code for its simplicity and generality.  In this paper, we present the basic concepts associated with convolution codes and results comparing the gap-to-capacity of the algorithm implemented with respect to Shannon’s optimal code. Read the rest of this entry »

Abstract: In many computer vision applications, it is valuable to know the 3D pose of an object or its real-world location when only images of a 2D scene are given.  This is known as 2D-3D pose estimation. On the other hand, it may also be necessary to seperate or segment an object from its background within an image.  Interestingly, these fundamental tasks of 2D image segmentation and 2D-3D pose estimation are usually decoupled and are rarely studied in a unified framework.  By leveraging the advantages of certain techniques from each problem, we couple both tasks in a variational manner through a single unique energy functional. Read the rest of this entry »

Abstract: Whether you are dealing with imagery that pertains to medical analysis, visual tracking, or 3D reconstruction, one fundamental assumption is that the data has already been properly aligned with respect to the same coordinate system.  For example, a 3D laser scanner creates a 3D model by taking a several partial scans of the object from different angles combing to form a 3D model.   However, how do you properly align such scans?  This is known as registration, specifically point set registration.  Thus, this project concerns itself with the problem of registering two (or more) point sets using particle filtering.  Read the rest of this entry »

Abstract: What is segmentation?  What is medical imaging?  While this work does not completely answer the latter question, it sheds light on one major area of medical imaging — segmentation.  This well studied problem in computer vision is the fundamental task of partitioning an image into disjoint regions.  With respect to biological vision, segmentation is consider a low-level task of being able to group a scene into certain classes (e.g., a room may be decomposed into chairs, tables, carpets, people, etc.).   In many medical applications, one may need to capture a certain structure such as the corpus collusm (seen right). We discuss one such framework and algorithm here for several challenging medical image segmentation tasks. Read the rest of this entry »