Current Issue : July - September Volume : 2019 Issue Number : 3 Articles : 5 Articles
Image classification is an important problem in computer vision. The sparse coding spatial\npyramid matching (ScSPM) framework is widely used in this field. However, the sparse coding\ncannot effectively handle very large training sets because of its high computational complexity,\nand ignoring the mutual dependence among local features results in highly variable sparse codes\neven for similar features. To overcome the shortcomings of previous sparse coding algorithm,\nwe present an image classification method, which replaces the sparse dictionary with a stable\ndictionary learned via low computational complexity clustering, more specifically, a k-medoids\ncluster method optimized by k-means++. The proposed method can reduce the learning complexity\nand improve the featureâ??s stability. In the experiments, we compared the effectiveness of our method\nwith the existing ScSPM method and its improved versions. We evaluated our approach on two\ndiverse datasets: Caltech-101 and UIUC-Sports. The results show that our method can increase the\naccuracy of spatial pyramid matching, which suggests that our method is capable of improving\nperformance of sparse coding features....
With explosive growth of malware, Internet users face enormous threats from Cyberspace, known as â??fifth dimensional space.â?\nMeanwhile, the continuous sophisticatedmetamorphismofmalware such as polymorphismand obfuscationmakes itmore difficult\nto detect malicious behavior. In the paper, based on the dynamic feature analysis of malware, a novel feature extraction method\nof hybrid gram (H-gram) with cross entropy of continuous overlapping subsequences is proposed, which implements semantic\nsegmentation of a sequence of API calls or instructions. The experimental results show the H-gram method can distinguish\nmalicious behaviors and is more effective than the fixed-length n-gram in all four performance indexes of the classification\nalgorithms such as ID3, Random Forest, AdboostM1, and Bagging....
State-of-the-art human detection methods focus on deep network architectures to achieve\nhigher recognition performance, at the expense of huge computation. However, computational\nefficiency and real-time performance are also important evaluation indicators. This paper presents\na fast real-time human detection and flow estimation method using depth images captured by a\ntop-view TOF camera. The proposed algorithm mainly consists of head detection based on local\npooling and searching, classification refinement based on human morphological features, and tracking\nassignment filter based on dynamic multi-dimensional feature. A depth image dataset record\nwith more than 10k entries and departure events with detailed human location annotations is\nestablished. Taking full advantage of the distance information implied in the depth image, we achieve\nhigh-accuracy human detection and people counting with accuracy of 97.73% and significantly reduce\nthe running time. Experiments demonstrate that our algorithm can run at 23.10 ms per frame on a\nCPU platform. In addition, the proposed robust approach is effective in complex situations such as\nfast walking, occlusion, crowded scenes, etc....
Machine learning techniques are a standard approach in spam detection. Their quality depends on the quality of the learning set,\nandwhen the set is out of date, the quality of classification falls rapidly.Themost popular publicweb spamdataset that can be used to\ntrain a spamdetectorâ??WEBSPAM-UK2007â??is over ten years old. Therefore, there is a place for a lifelong machine learning system\nthat can replace the detectors based on a static learning set. In this paper, we propose a novel web spam recognition system.The\nsystem automatically rebuilds the learning set to avoid classification based on outdated data. Using a built-in automatic selection\nof the active classifier the system very quickly attains productive accuracy despite a limited learning set. Moreover, the system\nautomatically rebuilds the learning set using external data from spam traps and popular web services. A test on real data from\nQuora, Reddit, and Stack Overflow proved the high recognition quality. Both the obtained average accuracy and the F-measure\nwere 0.98 and 0.96 for semiautomatic and fullâ??automatic mode, respectively....
As we move towards improving the skill of computers to play games like chess against\nhumans, the ability to accurately perceive real-world game boards and game states remains a\nchallenge in many cases, hindering the development of game-playing robots. In this paper, we\npresent a computer vision algorithm developed as part of a chess robot project that detects the\nchess board, squares, and piece positions in relatively unconstrained environments. Dynamically\nresponding to lighting changes in the environment, accounting for perspective distortion, and using\naccurate detection methodologies results in a simple but robust algorithm that succeeds 100% of\nthe time in standard environments, and 80% of the time in extreme environments with external\nlighting. The key contributions of this paper are a dynamic approach to the Hough line transform,\nand a hybrid edge and morphology-based approach for object/occupancy detection, that enable the\ndevelopment of a robot chess player that relies solely on the camera for sensory input....
Loading....