Visual simulation of fluid phenomena such as flame and water flow is an interesting and challenging research topic that has been attracting researchers for a long time in the history of CG. In recent years, physically-based simulation is a major fundamental approach in the area. In this paper, we propose a noise-based morphing technique for animating flame phenomena that is more efficient compared to the physically-based approaches. We obtain flame animations from a still image of candle by applying morphing technique continuously to the still image. The continuous morphing of the still image is handled by the control lines swaying according to 1/ f β noises representing the fluctuation of flame caused by air flow.
N-Sided hole filling is being studied by a lot of people from of old. However, in the case of long and slender shape, the distorted and discontinuous shape is generated, and the other method, a manual input by the user is necessary. Each of methods has its merits and demerits. It is still difficult to solve these three problems at same time. In this article, we have studied a method for solving these three problems. First, we calculate divided edge and parameter, and generate the edge that the area is divided. And then, we apply the curved surface to each area.
Colorization is a rapidly advanced computerized process that adds color to a monochrome image and a black and white movie efficiently. Since the colorization is an ill-posed problem, most of the conventional colorization techniques have been realized by giving partial color information. However, the colorized results depend on the color information given by a user subjectively, and training becomes necessary to obtain better results. In this paper, the author tries to colorize automatically by using the Bayesian inference.
We propose a real-time generation method of a snowfall by computer graphics. On snowfall phenomena, Shiina et al. observed real motion of each snow particle under no wind and no confliction among snow particles, and they introduced falling speed of snow particle from its diameter. They also classified its falling pattern from a shape of snow particle into vertical, slant and spiral falling. Then, in order to generate three kinds of falling pattern of a snow particle, we analyze observation data on snowfall motion and formulated horizontal direction and distance of a snow particle from its shape and size. Based on these results and using spherical polygon as a model of snow particle, total motion of snowfall is expressed by varying size and shape of snow particle and by combining three types of falling pattern.
We propose a generation method of human behavior in three-dimensional virtual space. In our method, each human model has its vision and behaves variously and autonomously based on the information obtained from vision. Visual information is obtained by rendering the environment CG data at the view point of each human model. As a result, each human model can get the color and distance information of objects. In addition to the information obtained from vision and the parameter that human model has, fuzzy theory enables us to generate various human behavior that is very similar to the human being. By applying this method to the human models that are passing the crosswalk that has traffic lights, we could generate various kind of human behavior, which takes account of the signal color and distance and changes according to individual character.
This paper presents several techniques usable for developing sketch interface to create line drawing image-like vector images. Recently, many researchers are studying on the design of user interface using pen-input-devices. We implemented experimental user interface by realizing the following functions: refining given rough strokes into lines considering the feature of hand drawing, generating plane graph presenting the planer structure of the lines by introducing new vertices to the intersection points of the lines, and erasing useless lines based on the plane graph efficiently. This approach taking into account the feature of analog devices and data structure of vector graphics enables us to create vector images efficiently.
Excavated cultural assets are reconstructed manually. Therefore, it takes time to reconstruct earthenware if the number of fractured objects are too much. In Addition, technical knowledge becomes necessary. In this article, we have studied a matching method of fractured objects which are measured by a 3D scanner. In our method, feature lines of fractured objects are extracted. After that, an adjacent fractured object are explored with matching geometrically these feature lines.
Efficient rearrangement of office equipment can decrease the total amount of moving distance of workers in the office. We study an algorithm that enables to detect workers in an office and to trace their trajectories. Implementing this algorithm, we analyse behavioral patterns of the workers and propose an efficient rearrangement of office equipment.
A method for object region extraction for partial image retrieval is proposed. In partial image retrieval field, it is important problem to extract object region to use it as query. We improved a method using distribution map of existing probability and brightness histogram in this report. In proposed method, extraction result of object region is recursively learned by using distribution map of existing probability and color histogram from rough object region to extract an accurate object region.
The purpose of this report is to improve retrieval accuracy of existing method by considering the positions of the local features. It is assumed that the interesting points concentrating on the partial of the image is a problem in the image retrieval. Loupias suggested the salient point extraction algorithm that does not concentrate points in the part of the image, and applied it to similar image retrieval. In order to further improve retrieval accuracy, we extracted the local features and compared each of them by considering their positions. Our method retrieved similar images with Geometric Hashing to compare the local features and their positions. Then we compared retrieval accuracy with the existing method.
When we can possess images of enormous quantity, it is a very important theme that high speed and accuracy can search an objective image. Then this paper proposes a method of the image search that uses the Hilbert scan for the method of scan the Wavelet transformation. When the image is Wavelet transformed time becomes the problem for the same pixel to handle it twice. Then processing time has been shortened by using the Hilbert scan and converting. The image was enabled to be retrieved at high speed.
In this paper, we describe a system for off-line recognition of Khmer characters. Khmer language is a phonogram and its character consists of consonant letters and vowel letters. Moreover, these letters contain many horizontal strokes and verticals strokes. These characteristics can be utilized to recognize Khmer-letter image which is taken by built-in digital camera of mobile phone. And, we aim to apply our system to be an automatic Khmer-Japanese translation service, that is, if we send a picture of Khmer language to the server we will receive its meaning in Japanese language via email.
Watermarking methods need the robustness against the various image processings. We are examining the tolerance of the watermarking to a geometrical approach. Since the synchronization information in the watermarked image attacked by the geometrical approach has been usually lost, it is difficult to extract the embedded watermark from the image. In this paper, we present a new approach for tolerant watermarking method against to geometrical attack. In this method, the synchronization of the position of watermark information is taken by inverse affine transformation to the attacked image. This method uses each of four corner point of two images of before attacks and after attacks to calculate the parameter of inverse affine transform.
In codec of motion pictures a good measure for evaluating the image quality including the color reproduction is strongly desired to determine a proper compression rate. In this article, we focus on the image quality of motion pictures compressed by H.264/AVC codec which is receiving increased attention. At first, S-CIELAB color difference calculation was applied to the frames of motion pictures after H.264/AVC codec. The performance was better than CIELAB color difference but not satisfactory. We then limited the region of calculation of CIELAB color difference to the smooth regions where compression error tends to attract attention of observers. Experimental results showing that the modification is promising were obtained.
In the fields of electronic commerce and the electronic art museum, the correct prediction of color reproduction of the subject is required and the estimation of the spectral reflectance of the object has been studied. However the color reproduction of fluorescent objects is difficult. To solve such problems, a method for modeling of the Donaldson radiance factor representing the spectral reflectance and the fluorescence characteristics and estimation of the model parameters was proposed in our previous study. In this study, we investigated the precision of the proposed method by a simulation experiment with Donaldson radiance factor database of fluorescent objects.
When we generally quantify the difference of the color, CIELAB color difference, determined by CIE, has been used. However, as CIELAB color difference is an objective evaluation method for the spatially uniform color patches, it is not suitable for the evaluation of the color still image with spatial distribution. The establishment of the objective evaluation method highly correlated with subjective evaluation is therefore desired for picture quality evaluation. In this paper, we propose an objective evaluation method considering human visual characteristic, such as contrast sensitivity function and the masking effect. The proposed method showed a good performance in the evaluation experiment.
In this paper, we proposed the noise reduction and color restoration method that takes into account the color and/or texture reproduction characteristics as a priority issue. Color difference between a target pixel and every one of surrounding pixels and spatial correlations of pixel values with in 5x5 pixel mask are affected for reproduction of pixel value of the target. The proposed method was evaluated on the points of noise detection accuracy and image restoration, which was applying to the images impaired by salt and pepper noise. As the results, the proposed method shows that it highly reduces noise and it keeps color and/or texture reproduction.
For practical 3-D computer-generated holograms (CGH), it is important to display shaded surface model images like computer graphics (CG). Therefore, we have developed a simple process that is used to obtain 3-D data for CGH from two CG images. The first image is a shaded CG image. The second image is a depth-cued image, i.e., image intensity is modulated by the depth of the object. In order to render these two images, we have implemented our proposed method on OpenGL. OpenGL includes not only support for HSRM but also texture mapping, and alpha-blend. Therefore, in this report, we show the effectiveness of the proposed method for texture mapping and alpha-blend.
We have developed Volume Extractor Ver.3.0 (VE), which has interactive functions of manipulation, image processing, segmentation, visualization, 3D model re-construction, and image measure for 3D images. In this paper, we show the details of 3D model re-construction functions from 3D images. We also describe the several applications and the effectiveness of VE.