Top PDF Learning a Scene Background Model via Classification

Learning a Scene Background Model via Classification

Learning a Scene Background Model via Classification

Learning a Scene Background Model via Classification Horng-Horng Lin, Student Member, IEEE, Tyng-Luh Liu, Member, IEEE, and Jen-Hui Chuang, Senior Member, IEEE Abstract—Learning to efficiently construct a scene background model is crucial for tracking techniques relying on background subtraction. Our proposed method is motivated by criteria leading to what a general and reasonable background model should be, and realized by a practical classification technique. Specifically, we consider a two-level approximation scheme that elegantly com- bines the bottom-up and top-down information for deriving a back- ground model in real time. The key idea of our approach is simple but effective: If a classifier can be used to determine which image blocks are part of the background, its outcomes can help to carry out appropriate blockwise updates in learning such a model. The quality of the solution is further improved by global validations of the local updates to maintain the interblock consistency. And a complete background model can then be obtained based on a mea- surement of model completion. To demonstrate the effectiveness of our method, various experimental results and comparisons are in- cluded.
Show more

14 Read more

A Multi-Class SVM Classification System Based on Methods of Self-Learning and Error Filtering

A Multi-Class SVM Classification System Based on Methods of Self-Learning and Error Filtering

{fjh95p, hch94, singling}@cs.ccu.edu.tw Abstract—In this paper, the technique of Sup- port Vector Machine has been used to deal with multi-class Chinese text classification. Several data retrieving techniques including word segmentation, term weighting and feature extraction are adopted to implement our system. To improve classification accuracy, two revised methods, self-learning and error filtering, for straight forward SVM results are pro- posed. The method of self-learning uses misclassified documents to retrain classification system, and the method of error filtering filters out possibly misclas- sified documents by analyzing the decision values from SVM. The experiment result on real-world data set shows the accuracy of basic SVM classification system is about 79% and the accuracy of improved SVM classification system can reach 83%.
Show more

9 Read more

MICK: A Meta-Learning Framework for Few-shot Relation Classification with Small Training Data

MICK: A Meta-Learning Framework for Few-shot Relation Classification with Small Training Data

Few-shot relation classification seeks to classify incoming query instances after meeting only few support instances. This ability is gained by training with large amount of in-domain annotated data. In this paper, we tackle an even harder problem by further limiting the amount of data available at training time. We propose a few-shot learning framework for relation classification, which is particularly powerful when the training data is very small. In this framework, models not only strive to classify query instances, but also seek underlying knowledge about the support instances to obtain better instance representations. The framework also includes a method for aggregating cross-domain knowledge into models by open-source task enrichment. Additionally, we construct a brand new dataset: the TinyRel-CM dataset, a few-shot relation classifica- tion dataset in health domain with purposely small training data and challenging relation classes. Experimental results demonstrate that our framework brings performance gains for most underly- ing classification models, outperforms the state-of-the-art results given small training data, and achieves competitive results with sufficiently large training data.
Show more

10 Read more

Supervised Fuzzy ART: Training of a Neural Network for Pattern Classification via Combining Supervised and Unsupervised Learning

Supervised Fuzzy ART: Training of a Neural Network for Pattern Classification via Combining Supervised and Unsupervised Learning

According to the matching criteria, if the input vector I is mapped to different desired output from that of the hyper- rectangle containing I, a new category is created.. Initial[r]

6 Read more

A machine learning approach for acquiring descriptive classification rules of shape contours

A machine learning approach for acquiring descriptive classification rules of shape contours

Ueda and Suzuki (4) retained perceptually relevant features in shapes to learn shape models. These results show that machine learning techniques really help the tas[r]

8 Read more

A novel learning algorithm for data classification with radial basis function networks

A novel learning algorithm for data classification with radial basis function networks

One important advantage of the proposed learning algorithm, in comparison with the support vector machines, is that the proposed learning algorithm normally takes far less time[r]

6 Read more

Robot motion classification from the standpoint of learning control

Robot motion classification from the standpoint of learning control

In robot learning control, the learning space for executing the general motions of multi-joint robot ma- nipulators is very complicated. Thus, when the learning controllers are employed as major roles in motion governing, the motion variety requires them to consume excessive amount of memory. Therefore, in spite of their ability to generalize, the learning controllers are usually used as subordinates to conventional controllers or the learning process needs to be repeated each time a new trajectory is encountered. To simplify learning space complexity, we propose, from the standpoint of learning control, that robot motions be classi$ed ac- cording to their similarities. The learning controller can then be designed to govern groups of robot motions with high degrees of similarity without consuming excessive memory resources. Motion classi$cation based on using the PUMA 560 robot manipulator demonstrates the e9ectiveness of the proposed scheme.
Show more

12 Read more

An EM based multiple instance learning method for image classification

An EM based multiple instance learning method for image classification

Abstract In this paper, we propose an EM based learning algorithm to provide a comprehensive procedure for maximizing the measurement of diverse density on given multiple Instances. Furthermore, the new EM based learning framework converts an MI problem into a single- instance treatment by using EM to maximize the instance responsibility for the corresponding label of each bag. To learn a desired image class, a user may select a set of exemplar images and label them to be conceptual related (positive) or conceptual unrelated (negative) images. A positive image consists of at least one object that the user may be interested, and a negative image should not contain any object that the user may be interested. By using the proposed EM based learning algorithm, an image retrieval prototype system is imple- mented. Experimental results show that for only a few times of relearning cycles, the prototype system can retrieve user’s favor images from WWW over Internet.
Show more

5 Read more

Using Back Propagation Model to Design a MIDI Music Classification System

Using Back Propagation Model to Design a MIDI Music Classification System

Table 6 compares the final training errors from different learning rates under the same initial condition and 10,000 training iterations. Based on the final training errors from the experiments, we found the larger the learning rate, the smaller the training errors. As a result, it seems that it is better to select a larger learning rate for the back propagation model. However, a smaller training error for the training samples may not correspond to a better test result. Based on our experience, an acceptable set of network parameters should be good for both the training and test patterns. In our model, we can have the best classification result when the learning rate is equal to 2.5. Fig. 2 plots the correct training result for this case. When the learning rate is set to 3.0, we may obtain an incorrect result as shown in Fig. 3.
Show more

6 Read more

Multiclass support vector classification via coding and regression

Multiclass support vector classification via coding and regression

2. Classification by multiresponse regression Consider the problem of multiclass classification with J classes based on d measurements of input attributes x A R d1 . Denote the membership set by J ¼ f1; 2; . . . ; Jg and each indivi- dual membership by g A J . Suppose we have training data fðx i ; g i Þ AR d1  J g n i ¼ 1 . Our goal is to construct a classification rule which, given a new input, can correctly predict the associated class label of this new input. Aside from various support vector approaches mentioned in Section 1 originating from the machine learning community, regression-based methods for classification have some long history in statistical literature [14,30,20,2]. The two articles [20,19] have an in-depth discussion of using multi- response regression for discriminant purpose. In general, the regression approach for classification consists of three major steps: encoding, linear regression and decoding. Nonlinear extension can be done by conventional adaptive nonparametric regression strategies, for instance, MARS [15], or by the kernel trick [34]. Hastie et al. [20] have used a particular scoring scheme, namely the optimal scoring, for encoding class labels. For regression-based classification, it is natural to ask if different coding schemes for transforming class labels into response scores lead to different classification results. To answer this question, we use general output codes to encode the class labels and give an equivalence criterion on coding and scoring schemes. Also, we adopt the kernel trick for nonlinear extension and employ mSVR
Show more

12 Read more

A Street Scene Surveillance System for Moving Object Detection,Tracking and Classification

A Street Scene Surveillance System for Moving Object Detection,Tracking and Classification

Fig. 1. Processing pipeline of the detection and tracking module. In this work, the background image subtraction method is used to segment the dynamic object in the scene. A background model of the scene is first generated based on the statistics of several input image frames. An inten- sity threshold is then used to segment the moving object from the difference between the background model and the current image frame. Since the background region of the scene will also change slightly due to the non- uniform outdoor illumination condition, it should be continuously updated after a period of time. To model the background scene caused by the illumination changes exclusively, the image is first converted to the HSV color space and only the brightness component is updated [13]. This process ensures that the pixel intensity values do not change significantly and the background scene is more robust for object segmentation. By thresholding the derived difference image, segmented and detected objects are represented by a binary image. After the morpho- logical erosion and dilation operations for noise reduction, the bounding boxes of the targets are given by the nonzero horizontal and vertical projections of the blobs. Fig. 2 shows the result of difference image and the detected object location.
Show more

7 Read more

Layered Scene Decomposition via the Occlusion-CRF

Layered Scene Decomposition via the Occlusion-CRF

We propose a new optimization approach, named Fu- sion Space (FS), to infer an occlusion-CRF model from an RGBD image. FS repeatedly proposes a restricted solution space for each variable, and solves a multi-labeling prob- lem to update the solution by TRW-S [10]. The restricted space must contain the current solution to guarantee mono- tonic convergence. We have seven types of proposals and try them one by one after a random permutation. We repeat this process three times. The exception is the first two pro- posals in the first iteration. The first proposal must be the surface adding proposal to generate surface labels, as ini- tially no surface labels exist. The second proposal must be a background hull proposal, which effective recovers a back- ground architectural structure. Now, we explain the details
Show more

9 Read more

Movie scene segmentation using background information

Movie scene segmentation using background information

c Institute of Information Science, Academia Sinica, Taipei, Taiwan Received 15 June 2006; received in revised form 24 July 2007; accepted 31 July 2007 Abstract Scene extraction is the first step toward semantic understanding of a video. It also provides improved browsing and retrieval facilities to users of video database. This paper presents an effective approach to movie scene extraction based on the analysis of background images. Our approach exploits the fact that shots belonging to one particular scene often have similar backgrounds. Although part of the video frame is covered by foreground objects, the background scene can still be reconstructed by a mosaic technique. The proposed scene extraction algorithm consists of two main components: determination of the shot similarity measure and a shot grouping process. In our approach, several low-level visual features are integrated to compute the similarity measure between two shots. On the other hand, the rules of film-making are used to guide the shot grouping process. Experimental results show that our approach is promising and outperforms some existing techniques.
Show more

10 Read more

N-Gram Model with a Background Distribution

N-Gram Model with a Background Distribution

{yanzehua,fli}@sjtu.edu.cn http://lt-lab.sjtu.edu.cn Abstract. Automatic thread extraction for news events can help peo- ple know different aspects of a news event. In this paper, we present a method of extraction using a topical N-gram model with a background distribution (TNB). Unlike most topic models, such as Latent Dirich- let Allocation (LDA), which relies on the bag-of-words assumption, our model treats words in their textual order. Each news report is repre- sented as a combination of a background distribution over the corpus and a mixture distribution over hidden news threads. Thus our model can model “presidential election” of different years as a background phrase and “Obama wins” as a thread for event “2008 USA presidential elec- tion”. We apply our method on two different corpora. Evaluation based on human judgment shows that the model can generate meaningful and interpretable threads from a news corpus.
Show more

9 Read more

Situated Learning in Class Using Pocket PCs via a Mobile Learning System

Situated Learning in Class Using Pocket PCs via a Mobile Learning System

3. Activities of learning and assessment The assessment of situated learning should focus on learning processes and outcomes (McLellan, 1996). McLellan (1996) contend that learning and assessment are simultaneously. Students begin conducting their exploration activities following the explanations of teachers. Learning activity guidance and testing was presented via the personal pocket PCs. Students sought the questions answers from authentic situations. Students then responded to the questions on pocket PCs. For example, the screen of the pocket PC presented “Find a plum blossom on elementary school campus, and then look to see how many petals each blossom has”. Students thus must go to find the plant and examine it before they can give an answer. Additionally, using a pocket PC with its embedded camera, students can identify specific plants and take a picture as requested.
Show more

4 Read more

A decision support system for constructing an alert classification model

A decision support system for constructing an alert classification model

a b s t r a c t As the rapid growth of network attacking tools, patterns of network intrusion events change gradually. Although many researches had been proposed to analyze network intrusion behaviors in accordance with low-level network data, they still suffer a large mount of false alerts and result in difficulties for network administrators to discover useful information from these alerts. To reduce the load of administrators, by collecting and analyzing unknown attack sequences systematically, administrators can do the duty of fix- ing the root causes. Due to the different characteristics of each intrusion, none of analysis methods can correlate IDS alerts precisely and discover all kinds of real intrusion patterns. Therefore, an alert-based decision support system is proposed in this paper to construct an alert classification model for on-line network behavior monitoring. The architecture of decision support system consists of three phases: Alert Preprocessing Phase, Model Constructing Phase and Rule Refining Phase. The Alert Processing Phase is used to transform IDS alerts into alert transactions with specific data format as alert subsequences, where an alert sequence is a kind of well-aggregated alert transaction format to discover intrusion behaviors.
Show more

11 Read more

A Conceptual Model of Knowledge Evaluation via Knowledge

A Conceptual Model of Knowledge Evaluation via Knowledge

Despite that knowledge management system has been extensively studied to attain potential solutions for business use, there has been less work in dealing with the knowledge evaluation. The knowledge evaluation is one of the most important tasks to ensure that a specific knowledge object is substantially valuable to an organization. In this paper, based on knowledge flow manipulation we propose a conceptual model for knowledge evaluation (KEM KFM ). The model adopts a mechanism tracing and keeping information that a specific knowledge object is created, acquired, stored, shared, and implemented. Value that it performs is also considered. The conceptual model produces ultimately a result for the importance of each knowledge object to the organization, and core knowledge may be defined accordingly.
Show more

1 Read more

A Learning State Space Model for Image Retrieval

A Learning State Space Model for Image Retrieval

In many cases, P2P tra ffic traverses long distances across core networks and multiple Internet service provider (ISP) networks, even though the content could have been retrieved from a much closer location. Internet video is similar and often delivered from distant servers via multiple, redundant, unicast streams. To address this, there has been a spate of recent e ffort on systems, information exchange, and control to enable efficient P2P and video content distribution. New systems and protocols are needed to enable Internet content to work in concert with the network to be delivered from the best source or over the least congested links. Localized P2P tra ffic may only traverse a few hops instead of ten or twenty; allowing a vast decrease in core network bandwidth.
Show more

12 Read more

Constructing a role playing interactive learning content model

Constructing a role playing interactive learning content model

9 RPG-like Learning Content Figure 7: The OOICM running process After the learning activity starts, the Petri Net runs and becomes State 2 in Figure 8. All scene objects are set according to their frames. The player can press up, down, left, and right key to move John. When the player presses left key, the user event interrupt happens, and if John collides with Mary, CsAT will be triggered as shown in Figure 7. In the inference process 1, the Collision slot value of John frame is added and then the attached procedure is triggered. In 2, Actor slot value of CsAT is added. In 3, the process should check whether PreCondition is satisfied. In 4, if PreCondition is true, the procedure of InActivity is run. In 5, the procedure of InActivity causes John and Mary to converse. In 6, after the conversation, the procedure of InActivity adds result to Result slot. In 7, Result slot is added, so attached procedure is triggered. In 8, the PostAction procedure is run to set Appearance slot of Mary frame. In 9, Appearance slot of Mary frame is updated. In 10, the GoalTest procedure is run to add a token in Petri Net, and therefore Petri Net for SCF becomes State 3 in Figure 8. Final, the Petri Net runs again and becomes State 4 in Figure 8. The learning activity will continue until P E in Petri Net has one token and the story ends.
Show more

8 Read more

A Spatial-Extended Background Model for Moving Blobs Extraction in Indoor Environments

A Spatial-Extended Background Model for Moving Blobs Extraction in Indoor Environments

To model the relation among pixels, we need to use the relations among near pixels to reduce time and storage consumption, and then extend the relations into a more global form. The methods based on the Markov random field (MRF) are well known for extend- ing the neighboring relations among pixels into a more global form. Image segmentation methods based on MRF [17, 19] assume that most pixels belonging to the same object have the same label and these pixels form a group in an image. The MRF combines col- ors among a clique of pixels in a neighboring system and uses an energy function to measure the color consistency. Then, the maximum a posterior estimation method is used to minimize the energy for all the cliques to find the optimal labels. In the MRF-based methods, the final segmentation results are strongly dependent on the energy functions of the labels in different cliques. If a high energy is assigned to the clique with unique la- bels, the extracted foreground regions will become more complete than those of the pixel-wise background models. An additional noise removal process is not required.
Show more

19 Read more

Show all 10000 documents...