In this paper, we propose a technique for automatic generation of pixel art from character images based on color-difference tolerance. Pixel art is one of image expressions for digital images and is based on the pixel level. Character images that are expressed by pixel art are composed of edge lines and a few colors. The proposed technique is for automatically generating pixel art from a photograph where there was only a single object. However, in the conventional method, it is difficult to determine several parameters of edge detection and decrease color for automatic generation of pixel art. In this paper, to create the optimal decreased color image for each target image, we propose a clustering method using a maximum distance algorithm (MDA) based on the human color-difference tolerance. Moreover, the complete automation of making pixel art is achieved by using the result obtained by clustering using the MDA for the threshold decision of the Canny edge detector. As a result, it is shown that several optimum parameters for pixel art are obtained by using the proposed method.

抄録全体を表示

We developed invisible code that is suitable for embedding into a natural image. Our code consists of a background layer, color adjustment layer and code layer. The code layer is composed of edgeless high-frequency blocks and edgeless low-frequency blocks. Two key technologies were devised to implement this idea. One is the color adjustment layer that adjusts the color of the background image for high-frequency blocks using an original blur filter. The other is the edgeless high-frequency block that increases the transparency of the periphery of the high-frequency block. To verify the effectiveness of our invisible code, we estimated the invisibility and reading accuracy for code embedded in four common types of images. The results demonstrate that the invisibility has been improved about 1.34 times and the reading accuracy has been improved about 1.38 times compared with the conventional invisible code.

抄録全体を表示

This paper proposes a new method of correspondence point search for intermediate viewpoint image generation using Lagrangian interpolation from multi-view images obtained from the circular camera arrangement. Correspondence point search is not performed on the basis of the real camera image but is performed on the basis of interpolation domain. In the conventional method, if the camera interval exceeds 4°, this causes disturbance in the image. However, the proposed method was able to produce an image with fewer disturbances to the intermediate viewpoint with a camera interval up to 6°. In addition, it was able to suppress a significant increase in processing costs by varying the block size to be used for correspondence point search by focusing on the amount of edge that is included in the search range.

抄録全体を表示

The purpose of this research is to propose a new super-resolution method using DCT. The proposed method enlarges a small block in an input image to a large block through DCT. A high-definition image is created by applying this method to the entire low-resolution input image. The enlargement of small blocks is carried out by expanding the DCT coefficient's frequency domain to the same size as that of the large blocks. First, the high-frequency component that is lacking in a small block is searched for from a database. Then, it is added to a small block and restores the high-frequency component. The high frequency component is searched for using a DCT sign index. The database is trained using multiple natural images beforehand. The proposed method made a high-speed search possible by using DCT sign index. From many simulation results, it was found that the proposed method was more effective than the traditional one.

抄録全体を表示

This paper describes a method of producing scenes for TV program, using “dynamic 3D models”. Dynamic 3D models are 3D models generated from images shot by multiple cameras surrounding an actor for each frame using 3D-reconstruction and texture-mapping techniques. They can provide highly realistic images as well as natural motion without using a motion-capture system. In addition, they allow flexible production, so the positions of the models can be changed and the models can be duplicated. Dynamic 3D models are therefore suitable for producing crowd scenes with numerous people. Although techniques of 3D reconstruction and texture mapping have been discussed and several applications have been proposed, they have not been widely used into TV or movie production because of the size of data and insufficient textural quality. By establishing an efficient production flow with recent powerful processors, we succeeded in producing scenes with sufficient quality for a TV program, which was aired in December, 2009. The scenes involving one thousand soldiers could be produced by shooting several sequences with only two actors. We describe the production flow that we established and present some scenes that we produced using dynamic 3D models for the drama.

抄録全体を表示

In this paper, we propose a refinement filter for depth maps. The filter convolutes an image and a depth map with a cross-computed kernel. We call the filter a weighted cross bilateral filter. The main advantages of the proposed method are the filter fits outlines of objects in the depth map to silhouettes in the image, while the filter reduces Gaussian noise in other areas. Additionally, its computational cost is independent of depth ranges. Thus, we can obtain accurate depth maps at the lower cost than the conventional approaches, which require Markov random field-based optimization methods. Experimental results show that the accuracy of the depth map in edge areas goes up and its running time is low.

抄録全体を表示

News programs often have sudden brightness changes caused by electronic flashes from still cameras. This paper proposes a correction method for the video deterioration caused by flashes using the frame interpolation technique. The proposed method creates an interpolated frame from neighboring frames to replace the frame area altered by the flash luminescence. To estimate accurate motion vectors used for frame interpolation, the proposed method adopts a novel block-based cost function that takes into account not only the frames before and after the flash but the flash frame itself. In addition, the hierarchical motion estimation and a new vector refinement filter are used to improve the interpolation quality. Experimental results for broadcasting video show that the proposed method is superior to the conventional interpolation method.

抄録全体を表示

A human-like active binocular vision system, inspired by binocular eye movements in animals, would help robots with automatic fast target switching, smooth target pursuing and efficient visual stabilization. In this paper, a control model that integrates saccadic eye movement, smooth pursuit eye movement, vestibulo-ocular reflex and optokinetic response is proposed. The control interface of the model has been simplified to one external saccadic command input. By Using this target selection command, target switching, target pursuing and visual stabilization of camera would run automatically. To implement the system with parallel processing, like the one used in neural network, the control model and multi-motor control are implemented in a FPGA chip. Finally, the proposed model was tested Using an image processing PC and a binocular robot head and the results show high efficiency of this control model.

抄録全体を表示

Reading as many technical documents as possible is important to improve our research. When we read these documents, we can understand their contents more easily by referring to related resources such as images, audio clips, and videos. Videos contain a variety of helpful information, and facilitate our understanding of technical documents. We propose a method to define video scenes and document elements, and to annotate them with additional information such as relationships. Based on these annotations and relationships, we developed a support system that uses videos and helps readers understand technical documents. We performed some experiments to confirm whether the system was usable.

抄録全体を表示

We report the statistical simulation results of the success probability of optical FDM/WDM Add-Drop-Mux (ADM) for CATV systems transferring terrestrial broadcast digital TV signals. Statistical simulation results show that the required signal crosstalk attenuation value of the optical notch filter can be reduced by inserting blank TV signal channels. The success probability value of the optical FDM/WDM Add-Drop-Mux can consequently be increased.

抄録全体を表示

A method for generating oblique caricatures by computers is presented. The inputs are facial images captured from both frontal and several oblique directions. The proposed method can generate caricatures not only for the directions where the input images were provided but also for arbitrary directions by using the results of the principle component analysis of the data on facial features.

抄録全体を表示