Recently,some studies on perception of transparency have been reported;the conditions of perception of tranparency using a physical model in an image, the conditions of the mechanical detection of transparent objects and the effect of transparency on pattern recognition and stereopsis. In this study, we consider not a monochromatic images, but a color image based on these stduies. First,we present that light intensity is not enough to precisely perceive transparency. Next we theoretically explain such a case that one of transparency conditions is not satisfied. We also discuss the tendency of human perception of transparency under the situation,not satisfying transparency condition,based on our experimental results.

View full abstract

The relationship between the position of the fixation point and the position of recognized letters was analyzed during several letters were briefly exposed. While the fixation point is fixed and no eye movement is planned letters around the fixation point are correctly recognized. While the fixation point is fixed but a saccade is planned, letters around the next fixation point are recognized. After a saccade, the recognition of letters around the new fixation point is continued in the visual system.

View full abstract

Two-image photometric stereo method have been developed to recover shape and reflectance of a smooth convex object, and we have extended the method to work on a convex polyhedron. However, most scenes consist of concave surfaces and/or multiple objects. Therefore, interreflections maybe exist between surfaces. Shape and reflectance recovered from intensity of those images can not be relied on, since they are influenced by interreflections. In this paper, we extend our two-image photometric stereo method to treat a concave polyhedron, and present an iterative algorithm to remove the influence of interreflections. We deal with a concave polyhedron with perfectly diffuse(Lambertian) and unknown constant reflectance. Correct results were obtained by erasing interreflections between object surfaces iteratively. Both simulation and experimental results show the feasibility and accuracy of the method.

View full abstract

A method to identify and to detect facial inclination is proposed, and the results of experiments are presented. We use pattern matching as an identification method. In these experiments,70 × n facial images were taken as sample data, where n is the number of inclination. Among these sample data,20 × n were selected for identification, and correctness is more than 90%. In the approach of facial inclination detection With triangular model, the features of eyes and mouth are used instead of the marker. Via experiment of mannequins, the effectiveness of our approach is shown.

View full abstract

The popular vehicle detector with ultrasonic waves are conventionally used to detect quickly the oar existence on the road, but it cannot acquire the more detail information about car size or car type. To detect the more detail traffic information about the moving vehicles, we have developed the image processing system based on personal computer(PC/AT compatible) combined with CCD color camera and image board. The developed algorithm can detect two cars moving counter-direction on the lane by color image information and also can estimate the car size and type. The experimental results showed that the correct estimation rate about car size and car type were 75% and 75% respectivly on upward traffic lane, and also were 80% and 83% respectivly on downward traffic lane.

View full abstract

In order to measure the snowfall rate every minute, weight of snowflakes in a bucket was automatically recorded using a electric balance and computer system. This system was synchronized with the TV camera for analyzing the image of snowflakes. Temporal density of snowflakes was analyzed from the images and weight of snowflakes. Moreover, the snowfall rate was calculated using the image data and it was consistent with the directly measured one.

View full abstract