http://www.bmva.org/bmvc/2007/papers/paper-70.pdf

Automatic player detection, labeling and tracking in broadcast soccer video

We propose the n -dimensional scale invariant feature transform ( n-SIFT) method for extracting and matching salient features from scalar images of arbitrary dimensionality, and compare this method's performance to other related features. The proposed features extend the concepts used for 2-D scalar images in the computer vision SIFT technique for extracting and matching distinctive scale invariant features. We apply the features to images of arbitrary dimensionality through the use of hyperspherical coordinates for gradients and multidimensional histograms to create the feature vectors. We analyze the performance of a fully automated multimodal medical image matching technique based on these features, and successfully apply the technique to determine accurate feature point correspondence between pairs of 3-D MRI images and dynamic 3D + time CT data.

/pdf/n-sift-n-dimensional-scale-invariant-feature-transform-447i9jdeny.pdf

$n$ -SIFT: $n$ -Dimensional Scale Invariant Feature Transform

Multi-core processors are proliferated across different domains in recent years. In this paper, we study the performance of frequent pattern mining on a modern multi-core machine. A detailed study shows that, even with the best implementation, current FP-tree based algorithms still under-utilize a multi-core system due to poor data locality and insufficient parallelism expression. We propose two techniques: a cache-conscious FP-array (frequent pattern array) and a lock-free dataset tiling parallelization mechanism to address this problem. The FP-array efficiently improves the data locality performance, and makes use of the benefits from hardware and software prefetching. The result yields an overall 4.0 speedup compared with the state-of-the-art implementation. Furthermore, to unlock the power of multi-core processor, a lock-free parallelization approach is proposed to restructure the FP-tree building algorithm. It not only eliminates the locks in building a single FP-tree with fine-grained threads, but also improves the temporal data locality performance. To summarize, with the proposed cache-conscious FP-array and lock-free parallelization enhancements, the overall FP-tree algorithm achieves a 24 fold speedup on an 8-core machine. Finally, we believe the presented techniques can be applied to other data mining tasks as well with the prevalence of multi-core processor.

Optimization of frequent itemset mining on multiple-core processor

Visual feature extraction with scale invariant feature transform (SIFT) is widely used for object recognition. However, its real-time implementation suffers from long latency, heavy computation, and high memory storage because of its frame level computation with iterated Gaussian blur operations. Thus, this paper proposes a layer parallel SIFT (LPSIFT) with integral image, and its parallel hardware design with an on-the-fly feature extraction flow for real-time application needs. Compared with the original SIFT algorithm, the proposed approach reduces the computational amount by 90% and memory usage by 95%. The final implementation uses 580-K gate count with 90-nm CMOS technology, and offers 6000 feature points/frame for VGA images at 30 frames/s and ~ 2000 feature points/frame for 1920 × 1080 images at 30 frames/s at the clock rate of 100 MHz.

https://ir.nctu.edu.tw:443/bitstream/11536/22155/1/000321926600020.pdf

Fast SIFT Design for Real-Time Visual Feature Extraction

Existing parallel mining algorithms for frequent itemsets lack a mechanism that enables automatic parallelization, load balancing, data distribution, and fault tolerance on large clusters. As a solution to this problem, we design a parallel frequent itemsets mining algorithm called FiDoop using the MapReduce programming model. To achieve compressed storage and avoid building conditional pattern bases, FiDoop incorporates the frequent items ultrametric tree, rather than conventional FP trees. In FiDoop, three MapReduce jobs are implemented to complete the mining task. In the crucial third MapReduce job, the mappers independently decompose itemsets, the reducers perform combination operations by constructing small ultrametric trees, and the actual mining of these trees separately. We implement FiDoop on our in-house Hadoop cluster. We show that FiDoop on the cluster is sensitive to data distribution and dimensions, because itemsets with different lengths have different decomposition and construction costs. To improve FiDoop’s performance, we develop a workload balance metric to measure load balance across the cluster’s computing nodes. We develop FiDoop-HD, an extension of FiDoop, to speed up the mining performance for high-dimensional data analysis. Extensive experiments using real-world celestial spectral data demonstrate that our proposed solution is efficient and scalable.

FiDoop: Parallel Mining of Frequent Itemsets Using MapReduce

Scale invariant feature transform (SIFT) is an approach for extracting distinctive invariant features from images, and it has been successfully applied to many computer vision problems (e.g. face recognition and object detection). However, the SIFT feature extraction is compute-intensive, and a real-time or even super-real-time processing capability is required in many emerging scenarios. Nowadays, with the multi- core processor becoming mainstream, SIFT can be accelerated by fully utilizing the computing power of available multi-core processors. In this paper, we propose two parallel SIFT algorithms and present some optimization techniques to improve the implementation 's performance on multi-core systems. The result shows our improved parallel SIFT implementation can process general video images in super-real-time on a dual-socket, quad-core system, and the speed is much faster than the implementation on GPUs. We also conduct a detailed scalability and memory performance analysison the 8-core system and on a 32-core chip multiprocessor (CMP) simulator. The analysis helps us identify possible causes of bottlenecks, and we suggest avenues for scalability improvement to make this application more powerful on future large-scale multi- core systems.

SIFT implementation and optimization for multi-core systems

In this paper, we present a tree-partition algorithm for parallel mining of frequent patterns. Our work is based on FP-Growth algorithm, which is constituted of tree-building stage and mining stage. The main idea is to build only one FP-Tree in the memory, partition it into several independent parts and distribute them to different threads. A heuristic algorithm is devised to balance the workload. Our algorithm can not only alleviate the impact of locks during the tree-building stage, but also avoid the overhead that do great harm to the mining stage. We present the experiments on different kinds of datasets and compare the results with other parallel approaches. The results suggest that our approach has great advantage in efficiency, especially on certain kinds of datasets. As the number of processors increases, our parallel algorithm shows good scalability.

/pdf/tree-partition-based-parallel-frequent-pattern-mining-on-2c3829k83n.pdf

Tree partition based parallel frequent pattern mining on shared memory systems

Soccer highlight detection is an active research topic in recent years. One of the difficult problems is how to effectively fuse multi-modality cues, i.e. audio, visual and textual information, to improve the detection performance. This paper proposes a novel two-dependence Bayesian network (2d-BN) based fusion approach to soccer highlight detection. 2d-BN is a particular Bayesian network which assumes that each variable depends on two other variables at most. Through this assumption, 2d-BN can not only characterize the relationships among features but also be trained efficiently. Extensive experiments demonstrate the effectiveness of the proposed method.

/pdf/soccer-highlight-detection-using-two-dependence-bayesian-5g28lz0u6p.pdf

Soccer Highlight Detection using Two-Dependence Bayesian Network

Generally this disclosure describes interactive media methods and systems. A method may include capturing an image, detecting at least one face in the image, determining an identity and expression corresponding to the at least one face, generating an icon for the at least one face based on the corresponding expression, and displaying the icon on a video monitor.

Interactive media systems

Replays are key cues for events detection in sport videos since they are the immediate consequence of highlights or important events happened in sports. In many sports videos, replays are usually sandwiched with two identical logo transitions, prompt the beginning and end of a replay. A logo transition is a kind of special digital video effects, usually contains 12-35 consecutive frames, describe a flying or variable object. In this paper, a novel automatic logo detection approach is proposed. It contains two main stages: a logo transition template is automatically learned by dynamic programming and unsupervised clustering, a key frame is also extracted; then the extracted key frame and the learned logo template are used jointly to detect logos in sports videos. The optical flow features are used to depict the motion characteristics of the logo transitions. Experiments on different types of sports videos show that the proposed approach can reliably detect logos in sports videos efficiently.

Yimin Zhang

Papers

SIFT implementation and optimization for multi-core systems

Tree partition based parallel frequent pattern mining on shared memory systems

Soccer Highlight Detection using Two-Dependence Bayesian Network

Interactive media systems

A novel sports video logo detector based on motion analysis