Leiguang Gong

Robust and fast collaborative tracking with two stage sparse optimization

The sparse representation has been widely used in many areas and utilized for visual tracking. Tracking with sparse representation is formulated as searching for samples with minimal reconstruction errors from learned template subspace. However, the computational cost makes it unsuitable to utilize high dimensional advanced features which are often important for robust tracking under dynamic environment. Based on the observations that a target can be reconstructed from several templates, and only some of the features with discriminative power are significant to separate the target from the background, we propose a novel online tracking algorithm with two stage sparse optimization to jointly minimize the target reconstruction error and maximize the discriminative power. As the target template and discriminative features usually have temporal and spatial relationship, dynamic group sparsity (DGS) is utilized in our algorithm. The proposed method is compared with three state-of-art trackers using five public challenging sequences, which exhibit appearance changes, heavy occlusions, and pose variations. Our algorithm is shown to outperform these methods.

Composition of image analysis processes through object-centered hierarchical planning

This paper presents a new approach to the knowledge-based composition of processes for image interpretation and analysis. Its computer implementation in the VISIPLAN (Vision Planner) system provides a way of modeling the composition of image analysis processes within a unified, object-centered hierarchical planning framework. The approach has been tested and shown to be flexible in handling a reasonably broad class of multi-modality biomedical image analysis and interpretation problems. It provides a relatively general design or planning framework, within which problem specific image analysis and recognition processes can be generated more efficiently and effectively. In this way, generality is gained at the design and planning stages, even though the final implementation stage of interpretation processes is almost invariably problem- and domain-specific. >

REAL-TIME MUTUAL-INFORMATION-BASED LINEAR REGISTRATION ON THE CELL BROADBAND ENGINE PROCESSOR

Emerging multi-core processors are able to accelerate medical imaging applications by exploiting the parallelism available in their algorithms. We have implemented a mutual-information-based 3D linear registration algorithm on the Cell Broadband Enginetrade (CBE) processor, which has nine processor cores on a chip and has a 4-way SIMD unit for each core. By exploiting the highly parallel architecture and its high memory bandwidth, our implementation with two CBE processors can compute mutual information for about 33 million pixel pairs in a second. This implementation is significantly faster than a conventional one on a traditional microprocessor or even faster than a previously reported custom-hardware implementation. As a result, it can register a pair of 256times256times30 3D images in one second by using a multi-resolution method. This paper describes our implementation with a focus on localized sampling and speculative packing techniques, which reduce the amount of the memory traffic by 82%

A Parallel GPU algorithm for mutual information based 3D nonrigid image registration

Many applications in biomedical image analysis require alignment or fusion of images acquired with different devices or at different times. Image registration geometrically aligns images allowing their fusion. Nonrigid techniques are usually required when the images contain anatomical structures of soft tissue. Nonrigid registration algorithms are very time consuming and can take hours for aligning a pair of 3D medical images on commodity workstation PCs. In this paper, we present parallel design and implementation of 3D non-rigid image registration for the Graphics Processing Units (GPUs). Existing GPU-based registration implementations are mainly limited to intra-modality registration problems. Our algorithm uses mutual information as the similarity metric and can process images of different modalities. The proposed design takes advantage of highly parallel and multi-threaded architecture of GPU containing large number of processing cores. The paper presents optimization techniques to effectively utilize high memory bandwidth provided by GPU using on-chip shared memory and co-operative memory update by multiple threads. Our results with optimized GPU implementation showed an average performance of 2.46 microseconds per voxel and achieved factor of 28 speedup over a CPU-based serial implementation. This improves the usability of nonrigid registration for some real world clinical applications and enables new ones, especially within intra-operative scenarios, where strict timing constraints apply.

Contextual modeling and applications

There have been increasingly growing interests in computational modeling of contextual knowledge representation and context-based problem solving. Different approaches have been developed to exploring the concept and computational mechanism of context, including various formal and empirical methods and systems based on different interpretations and applications of the notion. This paper discusses two basic interpretations of context and presents an operational definition of context from a computational perspective. Two paradigms for context-based problem solving are formulated. A goal-directed modeling framework for context-based problem solving is reported. In this framework, a context is defined in terms of an object and its relationships with other objects. Every context is centered at such an object, and cannot exist without it. A problem-solving task (e.g. find an object) with this framework can thus be defined as a process of determining a context or a sequence of contexts in which a solution path (e.g. steps for finding the object) can be decided. A set of operations is specified for context manipulation. Some application examples are described.

Video Event Detection Using Temporal Pyramids of Visual Semantics with Kernel Optimization and Model Subspace Boosting

In this study, we present a system for video event classification that generates a temporal pyramid of static visual semantics using minimum-value, maximum-value, and average-value aggregation techniques. Kernel optimization and model subspace boosting are then applied to customize the pyramid for each event. SVM models are independently trained for each level in the pyramid using kernel selection according to 3-fold cross-validation. Kernels that both enforce static temporal order and permit temporal alignment are evaluated. Model subspace boosting is used to select the best combination of pyramid levels and aggregation techniques for each event. The NIST TRECVID Multimedia Event Detection (MED) 2011 dataset was used for evaluation. Results demonstrate that kernel optimizations using both temporally static and dynamic kernels together achieves better performance than any one particular method alone. In addition, model sub-space boosting reduces the size of the model by 80%, while maintaining 96% of the performance gain.

Learning by focusing: A new framework for concept recognition and feature selection

In this paper, we develop a new method for feature selection and category learning. We first introduce two observations from our experiments: (1) It is easier to distinguish two concepts than to learn an isolated concept. (2) To distinguish different concept pairs we can find different selections of optimal features. These two observations may partly explain the success of human vision learning, especially why an infant can simultaneously capture distinguished visual features when learning new concepts. Based on these two observations, we developed a new learning-by-focusing method which first constructs focalized concept discriminators for pairs of concepts, and then builds nonlinear classifiers using the discrimination scores. We build datasets for four concept structure: vehicle, human affliction, sports, and animals, and experiments on all the four datasets verify the success of our new approach.

A 3D point matching algorithm for affine registration

PurposeLandmark point–based registration is an important tool in medical image analysis and applications. An efficient and robust new method is sought that does not require optimization and is less susceptible to noise.MethodsA new quaternion-based affine registration algorithm for matching 3D point sets, a generalization of a previously reported method for 2D point sets, was developed. The new algorithm computes the exact affine transformation and the unknown correspondence between two 3D point sets but does not require any optimization. The method performs robustly in the presence of noise for non-degenerate cases. The method performs a reduction of general affine case to an orthogonal case, and then computes the unknown rotation using the quaternion representation of the 3D points. The method assumes no prior knowledge of point-wise correspondence between the two point sets. The algebraic and geometric concepts underlying the method are shown to be both clear and intuitive.ResultsExperimental evaluation of the method was performed using both randomly generated synthetic 3D point sets and Stanford Bunny dataset. The algorithm performed well even for noisy data. A feasibility test was performed with medical MR scans showing promising results.ConclusionThe algorithm for point-based correspondence registration demonstrated robust results, even in noisy cases, and was shown to be feasible for use with medical images.

Accelerating 3D nonrigid registration using the cell broadband engine processor

Registration or alignment of medical images in clinical applications requires cost-effective high-performance computing. In this paper, we present a parallel design and implementation of a mutualinformation-based multiresolution nonrigid registration algorithm that takes advantage of the Cell Broadband Engine® (Cell/B.E.) Architecture by exploiting the different levels of parallelism and optimization strategies. The new method was tested with a dualprocessor Cell/B.E. processor-based system. The experiments show an average performance of 1.09 µs per voxel and excellent scalability, demonstrating real-time or near-real-time performance for the computationally demanding task of nonrigid image registration.

Anatomical object recognition and labeling by atlas-based focused non-rigid registration and region-growing

Computer-assisted recognition of anatomical objects in medical images is at the center of many important clinical applications. Automatic extraction and recognition of human abdominal structures from CT images has been particularly challenging for medical imaging research and applications. Intra-patient and inter-patient spatial, morphological and intensity variability is typically significantly contributing to great difficulties in developing satisfactory automatic solutions to the problem. In this paper we report on a new approach, which treats recognition of anatomical objects from a given medical image as a task of non-rigid registration followed by segmentation. It uses the knowledge inferred from an atlas or model image to specify a sequence of smaller sub-image spaces or spatial contexts to register progressively the atlas image with the given patient image. The labels of the target objects in the atlas image are carried over to the patient image by the registration process representing the recognition result, which is further improved by a region-growing process. Preliminary experiments of artery blood vessel recognition and labeling with real patient data have demonstrated the potential of the method to be a viable alternative solution to the problem.