The image capturing conditions are not always enough for the monitor of the trespasser. Therefore, the image processing to improve the distinguishing ability of the trespasser is effective for monitor duties. However, the conventional methods often use the motion of the parson. So, these methods are not effective in the situation that a parson stands still. This paper proposes a method for the extraction regions of human and the contrast improvement of the extracted human region. The method uses the density slope vector as the spatial feature of input frames. The experimental results show that the resultant images with good contrast in human regions were generated by the proposed method in comparison with by the conventional methods.

View full abstract

As an outline of an object in images, there are two kinds of edges, the one is an edge of the natural image that changes smoothly and another one is a steep edge of the character and the CG image. However, since the previous enlargement method processed regardless of the type of edges, the noises, such as jaggy, dotage, ringing, had happened in the enlargement image. In this paper, an enlargement method to extract the edges of a character and CG image among two types of images is proposed. Here, the concentration difference of 4 pixels near the interpolation point is used. In a smooth concentration domain with a less concentration difference, it interpolates smoothly with linear interpolation. Conversely, in the domain with a remarkable concentration difference, the position and direction of edge were presumed from the concentration difference, and the concentration value was computed from the distance of interpolation and edge. Moreover, the presumed edge was pursued to the neighboring domain, and it connected smoothly by changing inclination. The proposal method is compared with the previous method using visualevaluation and numerical evaluation, and as a result, validity of the proposal method is clarified.

View full abstract

In this paper, we identify a focus of epileptic seizure by plane-based spatio-temporal correlation (PSTC) method. A spike wave of epileptic seizure breaks out from focus and spreads to neighborhoods of brain. As the feature of the focus, the focus potential goes down when a spasm occurs (invigoration). First, we demonstrate that the presented PSTC is effective to estimate a moving object with divergences/disappearances. Next, we implement/make images of observed brain data and pursuit the spike waves' propagation by PSTC. We identify the focus by capturing the invigoration feature.

View full abstract

We propose two methods to apply calculation of the stability of hierarchical clustering results by adding a temporary element method (ATEM) to high dimensional data. Using ATEM, we can calculate the stability of hierarchical clustering without statistical processing. However, ATEM needs fast calculation algorithm of super volume of high dimensional geometry. In this paper, we propose a sampling method calculating stability of every dimension from specific dimension in case the clustering with Euclidean distance and centroid method. In addition, we propose an acceleration method with a lookup table of stability and its interpolation. Combining these two methods, we can calculate the stability of every dimension in practical time. As results of comparing with Ben-Hur method that is one of the representative methods, we can calculate stability more than the equal and between 1,000 and 100,000 times faster than Ben-Hur method.

View full abstract

“HYCLASS”, a new hybrid classification method for multi-dimensional images is proposed. This method consists of two procedures, textural edge detection and texture classification. In the textural edge detection, the maximum likelihood classification (MLH) method is employed to obtain the “color edges”, and morphological filtering technique is employed to convert the color edges into the “textural edges” by sharpening the opened parts of the color edges. In the texture classification, the supervised texture classification method based on normalized Zernike moment vector that the authors have already proposed. An experiment using some simulated texture images is conducted to evaluate the classification accuracy of the HYCLASS. The experimental results show that the HYCLASS can provide reasonable classification results in comparison with those by the conventional classification methods that employ either color feature or textural feature only.

View full abstract

In statistical pattern recognition problems, it is important to consider distribution of patterns. In this paper, we propose an algorithm to construct a multi-template dictionary for precise and efficient pattern recognition by considering the distributions. First, by considering relation between distributions of different categories, the possibility that a pattern may be misclassified is investigated. Then an algorithm of dividing a category into subclasses to reduce the possibility of misclassification by decreasing region overlaps is proposed. The proposed algorithm is applied to Japanese character recognition problem which requires high dimensional feature vectors. Experimental results show the effectiveness of the proposed algorithm.

View full abstract

In this paper, a statistical recognition method is proposed for surrounding scenes captured from an on-board camera. Particularly towards the recognition capability which has the special familiarity with in-vehicle record and communication, firstly, feature vectors are formulated on every block of pixels from the encoded features included in MPEG images and related situation information. Next, confidence vectors regarding object indices are estimated from these feature vectors and an index which has the maximum confidence value is determined as the recognized result. The multi-variate linear regression method is adopted for the estimation, from the viewpoint of applications that need sequential update and distributed cooperation on the basis of learned results. In the experiments, various cross-learning tasks are performed among representative scenes which have been captured at five places in two years. And the recognition rates on ten major objects are calculated so as to investigate particularly their relations to the learning data, place, time, season, and weather, according to the temporary characteristics.

View full abstract

Navigation services on GPS enabled cellular phones have been in widespread use in recent years. However, in places like airports, train stations, underground shopping centers, etc., where GPS positioning is difficult, providing navigation services remains a challenge. For example, when continuous positioning is not possible, we need to identify the primary factors to generate timing and content for effectively guiding the navigation. Researchers have studied the shape, complexity, and the Navigation Demand Model for crossing on routes to determine the timing and content of guidance. We have defined “close-set crossing”, which is a crossing with two or more passages that can be recognized as a destination for passengers. We hypothesize that the Navigation Demand Model for close-set crossing is different from the model of a regular crossing. This hypothesis has been verified through questionnaires. An expression of the model for close-set crossing has been derived by evaluating the relationships of direction of forward movement, crossing angles, and width of each passage at close-set crossing. We have also shown how to apply proposed model to determine the timing of guidance for pedestrians.

View full abstract