We measured the visibility of several type of chromatic text presented on a chromatic background on a liquid crystal (LC) display and extended the visibility index function (VIF) as a function of character luminance, background luminance, and character size to explain the current experimental results with a higher degree of accuracy. In addition to character luminance, background luminance, and character size, we analyzed the effect of different chromatic text and background colors under several saturation and hue conditions. The results show that the original VIF cannot predict the experimental data well under small luminance difference and large character conditions. Finally, a method was developed that showed that our improved VIF can precisely estimate the actual visibility of chromatic text presented under various conditions on a chromatic background.

View full abstract

A new real-time closed-captioning system for Japanese TV news broadcasts is described. The system is based on a hybrid automatic speech recognition system that switches its speech input between the original program sound and speech rephrased by a “re-speaker”. By improving our speech recognizer and its error correction system, we decreased the number of correction operators needed from four in our previous news system to one or two, depending on the difficulty of the speech recognition. Experimental results show that the system could be used at broadcasting stations for captioning typical news programs more easily and by fewer staff.

View full abstract

National Oceanic and Atmospheric Administration (NOAA) and Advanced Very High Resolution Radiometer (AVHRR) data are available on a daily basis and have been frequently used for global observation. The ground image can be resolved 1.1 km immediately below the satellite on a horizontal scale. Both pure and mixed pixels (mixels) can be used to accurately classify land-, sea-, and cloud- cover conditions. We propose the use of a classification algorithm for the NOAA-AVHRR data. The algorithm has four steps. First, multispectral bands are used to estimate elements of three classes (sea, land, and cloud) as supervised data for pre-classification. Second, pure pixels of the three classes are extracted on the basis of the multispectral bands and the Normalized Difference Vegetation Index (NDVI) of the same pixel. Third, we determine pure pixels and mixels by using fuzzy reasoning for the remaining pixels mentioned above, with the exception of the "land and sea" class. Finally, the edge information facilitates the retrieval of the "land and sea" mixel. Our experimental results suggest that the proposed approach provides results suitable for classifying various conditions.

View full abstract

This paper describes two experiments that were conducted to evaluate the required luminance of the black level on liquid crystal displays (LCDs) as a function of the average luminance of a displayed image, the screen illuminance and angular screen size. In the first experiment, 18 subjects adjusted the luminance level of the black region of various still video images, which had different average luminance levels, until they perceived them as black. The results revealed that the required luminance of the black level of an LCD as a function of the average luminance of a displayed image corresponds to the following formula: BL1= 0.137×AL^0.46. BL1 is the luminance of the level perceived as black and AL is the average luminance of the displayed image. In the second experiment, 22 subjects adjusted the luminance of a black region embedded in the center of a screen with uniform luminance, under different screen illuminance levels and angular screen sizes, until they perceived it as black. The results revealed that the required luminance of the black level of an LCD as a function of the screen illuminance and background luminance of the black region corresponds to the following formula: log BL2=0.44 log Ei+0.62 log L_bg −1.8. BL2 is the luminance of the level perceived as black, Ei is the screen illuminance level and L_bg is the background luminance of the black region. The angular size of the screen did not influence the required luminance of the black level of the LCDs.

View full abstract

Displaying clouds is one of the important factors in creating images of outdoor scenes by using computer graphics. Methods based on fluid simulations are often used since they can simulate realistic cloud movements and shapes. However, although this approach can create realistic clouds, the resulting shapes and motion depend on many simulation parameters and their initial status. Therefore, it is very difficult to create the clouds of desired shapes. In this paper, we propose a method for generating clouds with shapes that users have specified. The user specifies the multiple shapes of the clouds at several key frames. Then our method automatically adjusts parameters during the simulation in order to generate clouds forming the specified shapes. Using the proposed method, clouds that change into specified shapes can be animated.

View full abstract

This paper proposes a novel method to improve the performance of digital watermarking on the basis of fractal image coding using Quadtree Partition. We apply the Quadtree Partition technique to search for the best matching domain blocks and to improve the compression rate of images and robustness of the technique against attacks. Experiments have been performed in order to measure the robustness of the technique against JPEG compression and such attacks as Gaussian noise, scaling and smoothing.

View full abstract

The aim of our study is to find impressional similarities between color combinations and music. In this paper, we analyze the correlation between musical and color impressions on the basis of a set of music words, the affective value scale of music (AVSM), and a set of color combination words, the color image scale. In the subjective assessments using the color image scale, 230 subjects answered questionnaires to choose the words that evoked the same impression as corresponding words in the AVSM. As a result, each word in the AVSM has been mapped onto the color image scale.

View full abstract

This paper describes a fast deformation method for blood vessels, in which there are blood streams. It takes a long time to analyze liquid behavior on the basis of the Navier-Stokes equation, so our method uses Boyle's law, which defines the relationship between volume and pressure for gas. Blood is incompressible, while gas is compressible. We then consider quasi-pressure when a blood vessel is shrunk by an external force, and restore the volume of the blood vessel by using the quasi-pressure. As a result, the volume of blood vessels could be kept constant and our method satisfies the incompressible condition of blood. Finally, the proposed method has enabled us to deform blood vessels very fast compared with the fluid dynamics method based on the Navior-Stokes equation.

View full abstract

We present a food-logging system that distinguishes food images from other ones, analyzes the balance of food, and visualizes the processed log on the basis of image analysis. The image processing is based on features of color histograms, image frequencies in the DCT domain, detected image patterns and so forth in conjunction with learning algorithms of Support Vector Machine (SVM) and of Random Forest. Experimental results show that the images are extracted with a level of accuracy of over 85% and the food balance is estimated with a level of over 70%.

View full abstract