We propose a registration method for cross-sectional images to reconstruct three-dimensional shape of ocular fundus. Scanning lines to obtain cross-sectional images by an Optical Coherence Tomography are displaced caused by eye movement. In the proposed method, we register for cross-sectional images using parameters of Rol-l-Pitch-Yaw rotations to consider eye movement. In addition, we apply objective function considering shape of a disc to accurate reconstruction of ocular fundus. To evaluate the proposed method, registration results demonstrates visualization of the three-dimensional shape of ocular fundus.
Segmentation of lung areas from CT data is important as pre-processing for thoracic CAD that analyzes Pulmonary diseases. This paper presents a segmentation method for lung areas With huge pathological lesions diseases. In case with lung diseases, it is very difficult to identify lung areas precisely by only use of threshold-based method. In our study, we tried to obtain segmented lung areas by use of registration-based method.
Recently, in diagnosis of pulmonary embolism and pulmonary tumor, CT images are used mainly. In the diagnosis of pulmonary embolism, the information of pulmonary arteries is necessary, and in the discrimination of the benign and malignant pulmonary tumors, the information of pulmonary veins is necessary. So, it is important to classify pulmonary blood vessels into pulmonary arteries and pulmonary veins. But, it is impossible to distinguish easily between pulmonary arteries and pulmonary veins, because there are not differences of pulmonary arteries and pulmonary veins on CT values. Moreover, in the improvement for performance of CT devices, the number of slices of CT images increases. So, it has become the burdens of the radiologists. We have developed an CAD system for classifying pulmonary arteries and pulmonary veins automatically for decreasing the burdens of the radiologists. As stated above, pulmonary blood vessels can't be classified easily by a threshold processing. In this study, we made use of the relation that pulmonary arteries are in parallel to the bronchial tubes. We used the distance transform for the pulmonary blood vessels and the bronchial tubes using this relationship, and tried to classify pulmonary blood vessels based on the information of distance between the pulmonary blood arteries and the bronchial tubes. In our method, we performed classification for pulmonary blood vessels and bronchial tubes except for the pulmonary hilum, because the relation between pulmonary blood vessels and the bronchial tubes can't be used in the pulmonary hilum.
To diagnose the diffuse lung diseases, the computed tomography (CT) images are considered as effective. In general, however, the attenuation patterns of diffuse lung diseases in the CT image are complex and various. Hence an accuracy of the diagnosis depends on doctor's experience and skill. To improve the quality of the diagnosis, another doctors' opinions should be taken into consideration in general, that is, called the "second opinion". However, it increases a burden on the doctors. In this study, we propose the method that uses the result of the image classification by computer as a second opinion to reduce their burdens. We used a Counter Propagation Network (CPN) which is a kind of artificial neural network for classification method. In the CPN, the SOM part plays a role of a feature extractor and the perceptron plays a classifier role. First, CPN handles clustering by the self-organizing map, and the similar input data are assigned to the near nodes. The label is put up by using perceptronfor this expression.
Computer-aided Diagnosis (CAD) systems which analyze medical images automatically and detect the lesions such as tumors, are developed in medical field. Many of CAD systems take time for such processing. Therefore, PC only for CAD system is often set up, and used. We developed the CAD system to use Web. As a result, the client can be used in case of the terminal with a Web browser. Because the data communication with the DICOM server was automated, it is possible to use it from the portable terminal.
Visual simulation of natural phenomena has become one of the most important research topics in computer graphics. Examples include water, fire, smoke, clouds, and so on. Many methods for the simulation of these phenomena utilize techniques developed in computational fluid dynamics. Various natural phenomena can be simulated basically by solving Navier-Stokes (NS) equations. This paper first describes the NS equations briefly. Then, our approaches using the NS equations for simulations of smoke, clouds, and aerodynamic sound, are explained.
In this paper, we propose a method for rendering simple splashes of water generating on object's surface. Based on the result of having analyzed movies which recorded experiments of dropping waterdrops, we create three types of models of splash of the water. The proposed method has the following features; (1) Generation of the splash of the water which reflects material and wet condition on the waterdrop-landing surface is possible. (2) Calculation of the orbit of the splash of the water considering with the influence of gravity and air resistance is possible. (3) Generation of the splash of the water which reflects the incident direction of fall waterdrop is possible. We confirmed the usefulness of the proposed method through generating animations.
The paper proposes a method for generating a sequence of images with smooth transition of illumination from two input images with different lighting conditions. Our relighting approach is image?based, such as the light field rendering. We store the luminances (pixel RGB values) into “ lighting functions ” consisting of a couple of parameters related on normal vectors. Images with different light positions are rendered by interpolating the luminances retrieved from the lighting functions. The proposed method is a promising technique for many applications requiring a scene with variety of lighting effects, such as movies, TV games, and so on.
Humans have the ability of motion sharpening. Blurred edges look sharper when they are moving than when stationary. The sharpness-enhancement methods for digital image processing are generally a kind of filter that amplifies high-frequency components of image data. In this study, we propose pixel-based methods for image enhancement using a motion sharpening model of human vision. The proposed methods were tested using three image sequences, and the results were compared with the results of a conventional unsharp masking filter. The results show that the proposed methods have useful characteristics, which the conventional unsharp masking does not have.
We found a visual illusion; the angle of corner at the vanishing point drawn by parallel white lines of road images was observed smaller than the correct one. In this study, we investigated a relationship between depth cues in the images and this visual illusion. Our results showed the evidence of this visual illusion, and the depth cues were important factor. In addition, the direction of presentation images also was one of the main effects for the visual illusion.
Recently, Sleep Apnea Syndrome is attracting a great deal of public attention. The polysomnogram is used in a present diagnosis. However, to have to install a variety of kinds of apparatuses, those who diagnose it multiply the load. It is an important problem to decrease the diagnosis person's load and is done various researches. Then, in this experiments, whether breath could be read from the electrocardiogram vector angle was examined.
This research measured the electromyogram of extension-palmar flextion radial-ulnar deviation and rotation of wrist joint .The muscle which took part in the wrist joint motion was identified from the electromyogram.
We propose a system which extracts important data from criminal occurrence information distributed as an e-mail, accumulates the data as a database and visualizes the content of the database compositing with the electronic map. The proposed system allows us to accumulate the criminal information occurred sporadically and it supports us in grasping the occurrence situation and/or tendency visually and intuitively. In order to construct a criminal occurrence information database from distributed e-mails, we developed a technique of applying natural language processing for extracting required information from the e-mail. We developed the prototype of the proposed system and checked the usefulness of it through applying the system to the actual criminal information e-mail system.
This paper proposes intelligent media analysis and its application which are to greatly change the common style of viewing video contents such as TV programs. The proposal contains some essential technologies: selective collection of peripheral information on the Internet deploying crucial word abstraction based on morphological analysis, reformed visual content creation based on media selection and synchronization by audio/image analysis which enables the easy viewing on the mobile phone even in the outdoor, and contents recommendation which prioritizes the contents in the order of viewers' interest and favor owing to analysis of viewing histories. It also introduces a prototype system.