Recently, motion history images are used as a decent spatiotemporal template to represent human action. Each human action has its distinctive profile and a motion sequence that can be suitably represented by a histogram. In this paper a histogram oriented action recognition method is presented. In the proposed method, we use a mixture of histogram information as a feature vector for action representation. Histograms are taken from directional motion history images and their corresponding local binary pattern images which are then added together. We measured the performance of the proposed method along with some variants of it by employing KTH action dataset and found higher accuracies. The presented results also justify the superiority of the proposed method compared to other approaches for action recognition found in literature.

This paper introduces a simple algorithm for pedestrian detection on low resolution images. The main objective is to create a successful means for real-time pedestrian detection. While the framework of the system consists of edge orientations combined with the LBP feature extractor, a novel way of selecting the threshold is introduced. This threshold improves significantly the detection rate as well as the processing time. Furthermore, it makes the system robust to uniformly cluttered backgrounds, noise and light variations. The test data is the INRIA pedestrian dataset and for the classification, a support vector machine with an RBF kernel is used. The system performs at a state-of-the-art detection rates while being intuitive as well as very fast which leaves sufficient processing time for further operations such as tracking and danger estimation.

In this paper, we propose a new surface-based method to solve the rotation transformation of 3D rigid registration. Different from the criterion that minimize the sum of the squares of the errors between the corresponding parts of objects, here we estimate the spatial modalities of objects by their orientations. In details, Gauss map of surface is introduced to measure the distribution of surface curvature. To depict the characters of surface, we add sign weights to Gauss sphere according to the local total curvature. The stereographic projection is utilized to transform the spherical statistics to bifacial plane. As well, an optimization solution is done by iterated algorithm. We applied our proposed method on lung triangular surface registration. The experiments were completed by mutual information of block statistic.

Temporal Subtraction Technique is one of effective tools for diagnosis of lung cancer from thoracic CT images. By comparing two images of the same subject but shot at different time, the detection of temporal changing becomes facilitated. To acquire a more accurate subtraction image, the registration of these two images is critical. However it is not easy as the influence of the slight structures such as lung blood vessel. In this paper, a novel nonrigid image registration method based on vessel structure information is proposed. The similarity of the vessels structure is defined by means of the likelihood function of vessels structure and their direction. We combine this similarity of vessels with the intensity information of images. And the metrics are used as similarity measure in registration procedure. The proposed method has been applied to thoracic MDCT images, and the improvement of registration accuracy was investigated. The efficiency of our proposed method was indicated.

In this paper, we present some spatio-temporal (XYT) approaches for global-based action representation at the top of the motion history image method. Action representations can be considered as image templates, which can be useful for understanding various actions or gestures as well as for recognition and analysis. These spatio-temporal representations can be useful for various games environments. The motion history image represents how motion moves in image sequence. We present our SURF-based history image. In this method, we extract visual features to select candidate points based on the SURF detector. Then we extract motion features to calculate motion vectors at each point by employing optical flow. In this paper, we also propose a method called intensity-accumulated image method that can manage partial occlusion or some missing moving information. Based on our analyses, we strongly feel that our proposed methods - SURF-based motion history image method and the intensity-accumulated image method can be useful for various applications related to games, gesture and action understanding for analyses.

The development of an automatic human motion recognition system leads to the solution to the problems concerning the video-based applications in recognizing human activities. Such a system is to be investigated in the context of human motion analysis. Although there were a large number of researches in this area for a long time, there was little attention given to the development of a structured database for successful retrieval of motion data incorporating the time-space trade-off. We have proposed a system which is capable of dealing with large set of motion data employing an efficient database structure with improved performance. We have analyzed two motion representation techniques to realize the effectiveness of the system. Performance evaluation is performed by synthesized 3D human motions observed from eight camera directions. Finally, our results show that the proposed recognition scheme performs well for the captured motions.

This paper discusses an approach to human action recognition via local fea-ture tracking and robust estimation of background motion. The main contri-bution is a robust feature extraction algorithm based on KLT tracker and SIFT as well as a method for estimating dominant planes in the scene. Multiple in-terest point detectors are used to provide large number of features for every frame. The motion vectors for the features are estimated using optical flow and SIFT based matching. The features are combined with image segmenta-tion to estimate dominant homographies, and then separated into static and moving ones regardless the camera motion. The action recognition approach can handle camera motion, zoom, human appearance variations, background clutter and occlusion. The motion compensation shows very good accuracy on a number of test sequences. The recognition system is extensively com-pared to state-of-the art action recognition methods and the results are im-proved. 1

Motion overwriting due to motion self-occlusion is a big concern in motion and activity recognition. This paper presents a directional motion recognition approach that can solve the motion overwriting for complex actions or activities. Optical flow is split into four directions to compute motion templates. These templates are used to create feature vectors by Hu moments. Very satisfactory recognition results are achieved for various complex actions, which encompass motion overwriting. This method is compared with the basic motion history image method and multi-level motion history image method. The latter method professed that it can overcome motion selfocclusion problem and hence we compare these methods for several complex datasets with complex dimensions.

We have presented motion history-based human motion recognition technique with various formats of feature vectors. Since the inception of the motion history image (MHI) template for motion recognition, various progresses have been adopted to improve this basic MHI. Stages of development of appearance-based representation and recognition approach are presented here on the basic motion history-based approach to solve self-occlusion problem using our method. Excellent recognition rate for various motions has been found. This is based on gradient-based optical flow calculation. For recognition, Hu moments are considered to calculate feature vectors. Various feature vectors are considered in this paper.

This paper describes a novel display system of a three-dimensional (3-D) image. The proposed system employs a projector, a camera, and a hand-held screen. User can observe a 3-D image embedded in the space by holding up the hand-held screen at an arbitrary position. In the proposed method, the 3-D position of the hand-held screen is measured by the shadow created by projection light. By using this technique, the border between the real world and the digital data becomes seamless, and an intuitive operation is expected. The proposed system can be applied to various visualization systems.

This paper describes a technique for extracting moving objects from a video image sequence taken by a fixed or slowly moving camera by background subtraction. The background subtraction method is effective for extracting moving objects from a video provided by a fixed camera. But the latest background image should be employed for the subtraction in order not to be influenced by the light intensity change. A temporal median filter is proposed in this paper which detects the latest background images sequentially. From a video image stream provided by a slowly moving camera, the camera motion is estimated using a registration algorithm and the temporal median filter is applied to the common image area among a set of successive image frames to extract the background. In order that the camera motion estimation may become more exact effective, local correlation maps are calculated which can exclude outliers in estimated motion vectors. The technique was applied to the video images obtained from a hand-held camera and those taken from a camera set at the front seat of a car, and satisfactory results were obtained.

This paper presents a novel technique for motion capture applicable to temporary events. Every existent optical motion capture system employs stereo vision technique for 3D shape recovery. It needs multiple fixed cameras and their calibration. This restricts the system from applying to temporary events that need immediate image taking without camera calibration such as traffic accidents or unannounced street events. The optical motion capture technique described in this paper does not need camera calibration at the spot of an event interested. A 3-D model of the event is yielded from the captured video images by the application of the factorization technique. An experiment was conducted in which table tennis player's motions were captured at an athletic gym and its 3-D model was yielded at the lab. Two uncalibrated mobile cameras were employed for the experiment and satisfactory results were obtained. Discussion is given on future application of the technique.

Segmentation for lung areas from CT images is an important task on understanding tissue construction, computing and extracting abnormal areas. Many segmentation methods based on contour model are presented. SNAKES (active contour model), on the other hand, are used extensively in computer vision and image processing applications particularly to locate the object boundaries. In lung segmentation, SNAKES is used for extracting the detail of ROI. However, a completely automatic segmentation method is not yet published, since it needs some manual efforts for initial contouring and constructing the contour models. In this paper, we propose a segmentation method for lung areas based on SNAKES without considering any manual operations. Furthermore, abnormal area including ground-glass opacity or lung cancer is classified by voxel density on the CT slice set. Experiment is performed employing nine thorax CT image sets and satisfactory results are obtained. Obtained results are shown along with a discussion.

In general, segmentation is difficult because surrounding soft tissues and organs have similar CT values and sometimes contact with each other. We propose a new technique for automatic segmentation of lung regions and its classification for ground-glass opacity from the segmented lung regions by computer based on a set of the thorax CT images. In this paper, we segment the lung region for extraction of the region of interest employing binarization and labeling process from the inputted each slices images. The region having the largest area is regarded as the tentative lung regions. Furthermore, the ground-glass opacity is classified by correlation distribution on the slice to slice from the extracted lung region with respect to the thorax CT images. Experiment is performed employing twenty six thorax CT image sets and 96% of recognition rates were achieved. Obtained results are shown along with a discussion.

A method is described for obtaining 3-D object models, particularly human models, based on multiple cameras and a computer. Existing optical 3-D modeling systems employ calibrated cameras, which inevitably restricts the cameras at fixed positions, giving constraints to object motions. The proposed technique gives various ways of capturing images of object motions. It contains a mobile cameras system for taking images of an object moving widely by following it. It also provides a simultaneous entire shape recovery strategy by the employment of a surrounding cameras system. The proposed recovery technique puts its theoretical basis on the factorization method. The present paper focuses its attention on the issue of defining a measurement matrix having full entries. The technique is described and its performance is shown by the experiments on 3-D human motion modeling.

This paper employs an appearance-based eigenspace technique for representing and recognizing various human motions in terms of their postures. Various human motions are captured by video camera and they are sampled into defined image frames. These sequential images create a multi-dimensional eigenspace in which the motions are represented based on their postures change and this eigenspace is used for recognizing unknown postures/motions based on a MDL (minimum description length) principle. Since human clothes have an influence on creating an eigenspace, we employ blurred edge images throughput learning and training stages instead of original gray-scale images. The proposed method provides eigenspace updating when a new observation becomes available. Experimental results show satisfactory performance on representing and recognizing various motions and/or postures in the proposed approach.

Development of a mean eigenspace (MES) from some of selected human posture sets is proposed for detecting human standing postures. An average of several posture sets from some of selected subjects is used (instead of using just one set) fro creating MES and different unknown postures are recognized from it. In contrast to the conventional method, the paper proposes to employ image filtering to the input images for better performance of the eigenspace technique instead of using just original images for this particular study. For the selection of posture sets for the MES and for the recognition test, a leave out method is employed. The proposed approach of producing a MES is described and the robustness of the method is effectively proved in the experiment.

This paper describes a technique for recovering 3D shape of non-rigid objects employing factorization with stereo cameras. In the proposed image capture system, two video cameras move under the constraint that they capture several static landmarks commonly visible during observation other than non-rigid objects whose shape recovery is of primary concern. Employing the factorization technique, orientations of the video cameras are calculated first, which are then employed for computing 3D coordinates of the feature points specified on the non-rigid objects. Performance of the proposed technique is shown by an experiment.

This paper presents a technique for automating human scoliosis detection by computer based on moiré topographic images of human backs. Scoliosis is a serious disease often suffered by teenagers. For prevention, screening is performed at schools in Japan employing a moiré method in which doctors inspect moiré images of subjects' backs visually. The inspection of a large number of moiré images collected by the school screening causes exhaustion of doctors and leads to misjudgment. Computer-aided diagnosis of scoliosis has, therefore, been requested eagerly by orthopedists. To automate the inspection process, unlike existent three-dimensional techniques, displacement of local centroids is evaluated two-dimensionally between the left-hand side and the right-hand side of the moiré images in the present technique. The technique was applied to real moiré images to draw a distinction between normal and abnormal cases. According to the leave-out method, the entire 120 image data (60 normal and 60 abnormal) were separated into three data sets. The linear discriminant function based on Mahalanobis distance was defined on the two-dimensional feature space employing one of the data sets containing 40 moiré images and classified 80 images in the remaining two sets. The technique finally achieved the average classification rate of 88.3%.

An algorithm for identifying symmetry of a 3-D object given by its octree is presented, and the symmetry degree (a measure of object symmetry) is proposed. The algorithm is based on traversals of the octree obtained by the principal axis transform of an input octree. An object can be in an arbitrary position and with arbitrary orientation within the octree space, and a wide range of symmetries represented by groups of proper and improper rotations can be identified. It is shown that the octree data structure supports these operations well, especially for objects whose symmetry types are simpler or equal in complexity with a fourfold rotational symmetry. The operation of the algorithm is illustrated using some synthetic test objects. The results, which are composed of identified symmetry types and the corresponding symmetry degrees, were satisfactory.>

A technique for associating an image is described in terms of the network constraint analysis. Pixels and their gray-values are called units and labels, respectively, and a set of n pixels called the unit constraint set T provides the unit-label constraint set R for memorized images. Given an incomplete image X which can have occlusion, noise, distortion, etc., R receives screening by X to yield the reduced set R* (R* contained in R), since the elements of X constrain the elements in R. A depth first search is then applied to the elements of R* to obtain consistent solutions, if any, which are associated images. The proposed technique is free from the interference among memorized images which other association techniques suffer from. An iterative technique is also proposed for speeding up the depth first search. Performance of the proposed association technique is shown by the experiment employing 26 alphabetical letters.>

Describes a technique for symmetry detection and measurement of 3D medical entities based on an algorithm for general 3D symmetry identification. The technique can recognize a number of symmetries including bilateral, central, and rotational; and it expresses them in terms of a symmetry measure called 'the symmetry degree'. It is based on traversals of the octree obtained by the principal axis transform of the input octree, therefore it requires that the medical entity be represented by its octree. This may be the only technique for 3D symmetry measurement that can deal with arbitrary medical entities. The procedure is demonstrated on two skull models reconstructed from axial CT scans but it is applicable to entity models obtained by other techniques as well. It is also shown how to use this procedure for invariant alignment of different images of the same entity for their comparative analysis.>

A necessary and sufficient condition is described for a polyhedron to be reconstructed from its incomplete line drawings pair obtained from a stereo vision system. A practical judgement procedure using this condition is presented and the amount of knowledge that should be prepared for the judgement is also demonstrated experimentally.>