We have developed a bright LCD projector based on a new concept using three chrominance panels and one luminance panel. These panels are driven by a newly devised control circuit which generates a wide-band frequency luminance signal for the luminance panel and narrow-band frequency chrominance signals for the three chrominance panels. This is based on the characteristic of human vision in which the spatial resolution for chrominance is low compared with luminance. This paper explains the mechanisms of the distribution of the light energy for the luminance panel and the chrominance panels by using two models. One is the distribution model of the light for the same resolution of panels. The other is the digital filter model using the difference of resolutions between the luminance panel and the three chrominance panels. The results show that the original images are precisely displayed by first model. On the other hand, results of the second model show that the filter is effective for the representation of the original images, and would yield sufficient picture quality on our developed LCD projector based on this concept.
A monocular 2D hand-written figure is often used as a simple and useful medium to communicate a rough 3D shape in mind among humans in our daily life. In this article, we discuss 3D shape recovery to realize such communication between a human and a computer. In the previous work on shape recovery for object recognition, what to be recovered from an image is a detailed quantitative 3D shape that is completely consistent, in the geometric sense, to the image. On the other hand, when we assume shape recovery for man-machine communication, what to be recovered is the information about the 3D shape that is originally implied by a human. Considering this issue, we propose to recover the 3D shape from its hand-written figure interactively based on some qualitative features that would be important to describe 3D shapes in the mind of humans when they choose a monocular hand-written figure as the medium for communication. We also propose to infer the 3D shape that is not drawn in the figure from the shape recovered from the drawn part. Procedures for these two processes are also presented.
This paper describes a method of detecting intruders by video image processing. There are some studies on detecting intruders in video image by analizing the difference-binary images and motion vectors. These methods are, because of using simple features such as size and motion vectors of changing regions, not satisfactory to reduce erroneous detections caused by other factors, such as small animals, sways of trees, and changing in brightness. We propose a method to detect intuders by time series data by projection pattern of silhouette. In this method, changing regions in video images are extracted by interframe differential, and are transformed into projection patterns. Evaluation scores (a measure of similarity to human's silhouette) are caluculated by flexible matching between observed patterns and the reference pattern. Recognition processing is carried out using time series score by tracking a target of each video frame. We have made experiments with a prototype and confirmed the effectiveness of this method.
This paper describes a study of the detection of human eyes without employing complicated procedures. Template matching is widely utilized in supervised algorithms, however, most of these methods employ a single searching window in a limited scope. The authors take advantage of symmetry with respect to the center of the facial image and propose a pair of windows as a template. This is useful to the symmetric structure of the image data, and helpful in handling a pair of objects at once. Three window-pair templates (circular, elliptic, and square) and a single elliptic window for comparison are considered in this study. Up to six parameters (in the case of elliptic and square window-pairs) are employed to adjust the window-pair. Genetic Algorithm, based on the mechanics of probabilistic evolution, is employed to meet the multi-point search requirements without resorting to complex procedures. The algorithm, based on an unsupervised searching approach, is applied in evaluating the results from the differentiation of the image data. The differential image is used since the contours of the eye have a higher score in differentiation compared with other parts of the facial image. Some experimental simulations are made to confirm the proposed system's performance. The system is then adapted to investigate the window models listed above. The suitability of the different models in eye-detection was investigated by application to a set of images under identical Genetic Algorithm parameter settings. The detection capability of this method proved to be fairly reasonable without employing any deterministic approach, and the provided parameters in this system are easily handled regardless of the shapes of the faces in a sense.
Recently, the method which apply eye motion to human interface has become more popular studies. Generally, we are perceiving the outside by the five senses. Most part of the perception is depending on the sense of sight. If the eye motion can be detected and applied to human interface, we can not only make use to communicate to old and handicapped persons but also apply to study and medical treatment. Thus, we constructed the system using image processing that detects the gaze-point from relation between the eye rotation center and the iris center. The iris center is the center of approximate ellipse that is calculated by two iris edge points and ellipse's long and short radiuses in image, and the eye rotation center is the eyeball center that is approximated to a sphere. The line from eye rotation center to iris center is gaze-line, and the point of intersection to the screen and the gaze-line is gaze-point. In this paper, we describe a detection method of the gaze-point and the gaze-point input system that using wink.
Page segmentation is the task of extracting components of a document such as columns, textlines, figures and tables from a page image. This paper presents a method of page segmentation by analyzing background (white areas) of a page image. For page images without skew, it is known that white areas can be represented as white rectangles each of which maximally circumscribes white pixels. In general, however, extracted white rectangles include gaps between characters, words and those in figures which cause erroneous segmentation. Thus white rectangles need to be selected for correct segmentation. The characteristic point of our method is that the selection is based on the simple measure called effective area which is to estimate the effectiveness of a white rectangle as a delimiter of components by taking account of proximity with surrounding black areas. Our method correctly segmented 92.7% of components in 154 images of Japanese and English pages with various layout and resolution.
We have proposed an online navigation system for the visually impaired. In our previous system, we used a pedometer and a terrestrial magnetic sensor. This system has the problems that there are cumulative errors in the measurement of walking distance and errors in the detection of walking direction by the lean of the terrestrial magnetic sensor. In this paper, we develop another navigation system that uses a guide stick attaching a small wheel at the top of the stick instead of the pedometer and a device for keeping the terrestrial magnetic sensor horizontal in order to improve the problems of the previous system. Thus, we can obtain a more exact walking locus. To navigate the visually impaired, we propose a method using auditory signals that consist of voice sound and buzzer. Further, we attach an answering machine to the guide stick to provide the required information whenever requested. To evaluate the navigation system, we make some walking experiments by the blindfold and the visually impaired. This system shows that it can guide then to the destination of about 620m distance, and it is useful as a navigation system for the visually impaired.
Biological sensory information processing system mainly consists of the sensory organs and the brain. Modeling of this makes a new intelligent sensing system. In this paper the establishment of a new instrumental technique is discussed from this point of view. One of the major features of biological systems is parallel distributed processing. Multi-functional instrumentation (MFI) which spends great efforts on signal processing is one of the candidates of this technique. Its basic idea is that it can detect some kinds of information simultaneously and discriminate them effectively. But its weak point is how to acquire the suitable information processing. Then adaptivity which a neural network, a model of the brain, possesses is used. According to some experiments such as instrumentation of temperature and wind velocity it is clear that the estimated values by this method are almost the same as those by the conventional one. So it is said that this proposal system is useful.
Inductive learning algorithms acquire knowledge from a training example set. The algorithms basically deal with discrete attributes. To deal with numerical attributes, the algorithms must have the procedure which transforms numerical values to discrete values. We proposed a fuzzy inductive learning algorithm IDF which learned a fuzzy decision tree from a training example set. IDF also has the procedure which transforms an attribute range to fuzzy ranges and can deal with numerical and fuzzy values. However, this procedure tends to decompose the attribute range for the upper attribute in the fuzzy decision tree to many fuzzy ranges. Thus we propose new procedure which composes redundant fuzzy ranges using Akaike's information criterion and improve the IDF. We made numerical experiments for some real training example sets and verified the efficiency of new IDF. We also compare new IDF with C4.5 which is a representative inductive learning algorithm and show that IDF is more efficient than C4.5 in terms of classification accuracy.
To clarify the relationship between earthquakes and electric phenomena the earth potential difference (EPD) has been observed continuously at Aobayama in Sendai and Tsukidate, 60km north of Sendai. A new measuring method of the EPD is used to reduce the unnecessary signals induced by artificial and natural earth current sources. We have noticed just the EPD variational signals induced by seismic waves. On all earthquakes of more than 1 of JMA (Japan Meteorological Agency) intensity scale we can clearly observe EPD signals. The P-wave arraival times agree quite well with ones obtained by the seismograph at Meteorological Agency Seismograph. The EPD signals of Kobe earthquake, though less than 1 scale, have been clearly recorded at the both observatories points. The EPD signals are also observed at three different positions under the ground. We have found interesting electric field characteristics caused by seismic waves as follows; (1) the electrical field in the vertical direction is higher than in the horizontal direction by one order (2) this field increases rapidly coming up to the ground surface.