Search

In this paper, we will introduce a fast and robust method of constituting a probability map by extracting foreground object from background image which is basic technology to be applied to vision-based real-time game interaction. We implement a neural network with one hidden layer to recognize a motion of the probability map. In this process, we extract the foreground probability map which is robust against change of illumination using local binary pattern and improve the speed using integral histogram. Also, we use a novel Gaussian window frame mask which can be applied fast in the integral histogram process and reduce noise by adaptive Gaussian sigma value depending on the strength of edge. In the following, experiments are carried out and satisfactory results are obtained which indicate the fast and robust conditions.

Background Subtraction is an important preprocessing step for extracting the features of tracking objects in the vision-based HCI system. In this paper, the orientation histogram between the foreground image and the background image is compared to extract the foreground probability in the local area. The orientation histogram-based method is partially robust against illumination change and small moving objects in background. There are two major drawbacks of using histograms which are quantization errors in histogram binning and slow computation speed. With Gaussian binning and integral histogram, we present the recursive partitioning method that gives false detection suppression and fast computation speed.

In conversion of Web pages for mobile devices, the original intention of a Web designer is often spoiled since the existing Web conversion systems eliminate images indiscriminately or only convert objects into Hyperlink. Considering current Web environment with improved performance of hardware/software and the transition of information representation from text to visual, we need a new method that not only minimizes the deterioration of the image-layout but also respects the intention of the Web author. In this paper, we propose an adaptation method that minimizes the loss of information and preserves the original intention of a Web designer from the visual aspect. In image conversion, the proposed method extracts meaningful information from the back-ground instead of resizing in the conventional method. It also performs layout-based conversion for the arrangement of image objects. The experimental results show that the proposed method achieved both the accuracy and the usability in comparison with the existing methods.

We present a fast and robust algorithm for segmenting foreground object from background image by comparing local histograms. Background subtraction is a important preprocessing step for extracting the features that can be used for object tracking in surveillance system or HCI system in virtual environment. In this paper, the local histograms of the same area are used to compute a foreground probability. The histogram-based method is partially robust against illumination change and small moving objects in background. However without data quantization to reduce bin size, histograms are generally not suitable for realtime applications. Moreover quantization errors are a major drawback of using histograms. We propose a new method to keep the advantages of histograms without suffering computational load and quantization error using local kernel histogram with the multi-scaled integral histograms. We implement the video game interface with a trained neural network to prove the proposed method is highly applicable to coarse pose recognition.

In recent, development of wireless internet technology gradually has increased multimedia application in mobile device. Specially, application of grand capacity multimedia contents has increased in mobile environment gradually. MPEG-7 can search multimedia data using various type. But in restricted grand capacity of multimedia, MPEG-7 is used to be limited in desktop computer system mostly. Therefore, management of grand capacity multimedia data has become very important researching topic in mobile environment, and so embedded DBMS based on MPEG-7 for multimedia data management in client-server environment is needed. The system supports restricted hardware resource for multimedia application in mobile device. In this thesis, we propose a method for MPEG-7 data clustering that included MPEG-7 data property. On the base of the clustering result, we also develop the embedded multimedia database management system based on MPEG-7 scheme(EMDS:MPEG-7 Scheme Based Embedded Multimedia Database Management System) in mobile device such as PDA.

This paper proposes a new game-control interface using hand poses captured from a digital camera. Extracting skin color from the background-subtracted image is generally used for segmenting hand region. But it suffers from change in illumination, shadows and skin- colored objects so that unwanted subsets reduce accuracy in hand pose recognition. To deal with these problems, first, subtracting skin colors from the background image must be done and second, it mustn 't get affected by changes in background. The paper describes an approach to using variance of homogeneity for skin color extraction. Homogeneity which represents contrast in a window is robust against change in illumination and can be used to effectively extract skin color rather than color or brightness based subtraction from background. The recognition of hand poses proceeds by computing fourier descriptor of extracted pixels. This approach to recognition can robustly identify hand poses comparing to the others. Experiments on real-time game demonstrate the robustness of the proposed method.