Optical Flow Back-Projection Motion vector plays one significant feature in moving object segmentation. However, the motion vector in this application is required to represent the actual motion displacement, rather than regions of visually significant similarity. In this project, Region-based Selective Optical Flow Back-projection (RSOFB) which back-projects optical flows in a region to restore the region’s motion vector from gradient-based optical flows, is proposed to obtain genuine motion displacement. The back-projection is performed based on minimizing the projection mean square errors of the motion vector on gradient directions. As optical flows of various magnitudes and directions provide various degrees of reliability in the genuine motion restoration, the optical flows to be used in the RSOFB are optimally selected based on their sensitivity to noises and their tendency in causing motion estimation errors. In this project a deterministic solution is also derived for performing the minimization and obtaining the genuine motion magnitude and motion direction.	Fall Detection using Modular Neural Networks and Back-projected Optical Flow This project presents a video-based algorithm for fall detection. The algorithm is based on the back-projected optical flow and modular neural networks. From a video sequence, the moving object is first extracted and the pixels with high variance of the extracted object are determined as feature points. Then the proposed back-projected optical flow is employed to estimate the genuine motion of these feature points. The normalized accumulated values of four directions of the estimated motion vectors of the feature points form a to-be-recognized feature vector. The sequence of feature vectors is fed into a time-delay neural network modular to detect whether a falling event occurs. The outputs of different modules, which have learned different moving direction of the object, are fed into a committee neural network for fall detection.
Improving the reliability of RFID-based Psychiatric patient tracking This project is focusing on promoting the reliability of psychiatric patients tracking using RFID system. RFID has been regarded one economic approach in monitoring the psychiatric patients. However, in order that the RFID can be commonly accepted as one of the major methods in patient monitor, it requires both the low false alarm rate and the low miss detection rate. In other words, the reliability of patient tracking needs to achieve a significant high level. However, with current RFID techniques, several reasons which could usually cause the interference and miss detection are (1) the interferences of signals from two locators of overlapped cover ranges and (2) the interference of signals from close tags. All these reasons cause the high false alarm and miss detection rates in psychiatric patients tracking using RFID.	Using Curve Fitting and Spectro-temporal Neural Network for Triggering Feature Analysis and Behavior Modeling of FM Specialized Cells In recent years, Biomedical Informatics has become a new trend of science and technology. Biomedical Informatics is bringing together researchers from bioinformatics, medical informatics and computer science. The principle of this subject is using mathematical computation, statistics and computer analysis for life sciences research. Therefore, their application is very extensively, including genes, medical treatment, medicine and so on.
Tangible Photo-Realistic Virtual Museum (2005/07/21) We present a tangible photo-realistic virtual museum system that couples augmented panorama and vision-based tracking with tangible interface to achieve the real-time interaction between the visitor and the exhibitions. This system explores how to use a physical object named physical control cube (PCC) as a tangible embodiment of the visitor's hand when s/he appreciates the exhibitions. This approach provides a direct mode of browsing the museum in the cyberspace as we walk in the museum. This research was supported in part by the National Digital Archives Program, NSC 93-2422-H-001-0004 and 93-0201-29-戊- 3-6.2.2 from the National Science Council, Taiwan and developed in Institute of Information Science, Academia Sinica.	Reconstructing 3D Model of Real Scenes from Photographs (2005/08/03) This system automatically extracts the 3D information and reconstructs a textured 3D model from a sequence of images of a real scene. No prior knowledge about the scene is needed to build the 3D models. All information such as camera pose and orientation will be estimated through the processes. Therefore, this system offers a high degree of flexibility when taking photographs. The only constraint is the intrinsic camera parameters need to be obtained first. The 3D modeling task is decomposed into 4 successive steps. The camera intrinsic parameters are calibrated using a calibration board first. Second, the camera pose and the epipolar geometry between a stereoscopic image pair are estimated by the corresponding points of this pair. Next, consecutive images of the sequence are treated as stereo pair and the disparity maps are computed by area matching. Finally, the dense 3D points are estimated by the linking matches through consecutive image pairs. Then, these 3D points are visualized as a 3D model which is also texture mapped for photo-realistic appearance. This system has been tested on several real scenes, and some of the reconstructed models are shown in this paper
Vision-Based Gait Analysis This research provides an approach to analyze gait. By this approach, some abnormal types of gait can be distinguished. This research is aimed to help home-care systems and surveillance applications.	Modified MMSE DMC and Error Concealment for Improving H.264 Error Resilience (2005/08/08) This study proposed schemes to improve H.264 error resilience. Besides improving the original spatial and temporal error concealment schemes adopted in H.264 test model, we use double-motion-vector mechanism to decrease error propagation.
A Motion Emphasized 3-D SPIHT for Visual Improved Video Coding This research proposes to rearrange bit rates from eye-sensitive regions which is the motion of the video to eye-insensitive regions which is the static regions of the video, so that eye-sensitive regions would have higher fidelity so as to obtain a better visual condition on the whole video in low bit rate situations.	The Integration of Physiological Signals from Multiple Devices Plan And Home Health Care Box (2005/08/11) In this plan we will implement a intelligent home health care box with extensibility. The care box has functions of collecting physiological signal and information exchange and the ability of intelligently automatic alarm notification. Through the system configuration, the family health care box can connect various care devices and accept the related physiological signal of the patient. The built-in function of physiological signal detection and alarm function can analyze the physiological situation to judge the unusual change and other symptom reaction. It will broadcast alarm when the unusual situation happened to increase the immediacy of care and convenience of home care to the patient. And through physiological signal monitoring function , the care center can understand well the health condition and related physiology signal of the patient.
3-D Localization of Clustered Microcalcifications Using Cranio-Caudal and Medio-Lateral Oblique Views This research presents a 3-D localization method to register clustered microcalcifications on mammograms from cranio-caudal (CC) and medio-lateral oblique (MLO) views. The method consists of three major components: registration of clustered microcalcifications in CC and MLO views, 3-D localization of clustered microcalcifications and 3-D visualization of clustered microcalcifications. The registration is performed based on three features, gradient, energy and local entropy codes that are independent of spatial locations of microcalcifications in two different views and are prioritized by discriminability in a binary decision tree. The 3-D localization is determined by a sequence of coordinate corrections of calcified pixels using the breast nipple as a controlling point. Finally, the 3-D visualization implements a virtual reality modeling language viewer (VRMLV) to view the exact location of the lesion as a guide for needle biopsy. In order to validate our proposed 3-D localization system, a set of breast lesions, which appear both in mammograms and in MR Images is used for experiments where the depth of clustered microcalcifications can be verified by the MR images.	Comparative Evaluation of classifiers and Feature Selection Methods for Mass Screening in Whole Digitized Mammograms The objective of this research is to extract the features from Digitized Mammograms block by block based on a variety of texture features, and improve Correct Classification Rate for mass screening by comparing classifiers with difference Feature Selection Methods. Unlike traditional methods that are employed to perform detection based on gray level, this paper has adopted three groups of characteristics related to mass texture, namely, SGLD(Spatial Gray Level Dependence), TS(Texture Spectrum) and TFCM(Texture Feature Coding Method). Totally 19 texture features are offered to describe the characteristics of masses and normal textures. Next, under the testing by classifiers, three Feature Selection Methods--SBS (Sequential Backward Selection), SFS (Sequential Forward Selection) and SFSM (Sequential Floating Search Method) are used to find out suboptimal subset from 19 features in order to improve the performance of mass detection. Finally, the performances are compared when two classifiers—PNN (Probabilistic Neural Network) and SVM (Support Vector Machine) are applied--to find out the optimum correct classification rate. The experimental images in this research are obtained from MIAS MiniMammographic Database offered by Mammographic Image Analysis Society (MIAS). The experimental data show that, the testing samples of this system can offer a 98% Detection Rate, with only 1.4% False Alarm Rate.
Tissues Classification for Breast MRI Contrast Enhancement Using Kalman Filter-based Linear Mixing Method Among the most recent techniques of breast examination, a great attention is being paid to breast MRIs. Since the contrast-enhanced breast MRIs acquired by traditional contrast-injection has shown to be very sensitive in the detection of breast cancer, this system adopts a spectral signature detection technology, Kalman Filter-based Linear Mixing Method (KFLM), which could successfully classify breast MRIs into four major tissues and present the classified results in high contrast tissue-separated images. A series of experiments using real MRIs and phantoms are conducted and compared to the commonly used c-means (CM) method for performance evaluation. After compare with CM algorithm and contrast-injected breast MRIs, the results showed that the high contrast images generated by spectral signature detection technologies had a superior quality.	Using Feature Selection with Support Vector Machine in Gastric Histology Classification (2004/08/01) This study presented a computer-aided diagnosis system using sequential forward floating selection (SFFS) with support vector machine (SVM) to assist physicians to obtain gastric histology from endoscopic images without invasive biopsy during endoscopy. At first, lots of candidate images features of endoscopic images are extracted via discrete wavelet transform, color and texture criterion in the feature extraction stage. However, lots of candidate image features cannot effectively describe histological results. Then, in the feature selection stage, SFFS is applied to select a subset of features, which performs the best classification result under SVM. In the feature classification stage, SVM can do the classification task well based on the selected image features. In order to enhance the performance of the classifier, finding proper threshold in SVM for different histological results is necessary. Based on this methodology, a new diagnosis system is implemented to provide physicians the instant gastric histology results during the endoscopy without invasive biopsy.