Recently, many researchers have paid attention to a technique called non-photorealistic rendering (NPR). In this paper, we propose a method for painterly rendering of water surface. The painterly images of water surface are efficiently generated by making use of techniques used in the field of three-dimensional computer graphics. Our method can create not only static images but also painterly animations of water such as a river. Moreover, we develop an interactive system using our method. This enables the user to reflect the user's mind on the resulting images.

View full abstract

This paper describes statistical analyses of colour information in paintings applying information entropy and Zipf's law. We found statistical characteristics are tend to be appeared in each trends, especially, Impressionist paintings have complexity and at the same time, regularity (follow Zipf's law) in their colour information.

View full abstract

In this paper, we generate a realistic drawing stroke that is highly affected by paper irregularity as one of the interactive NPR techniques. We focus on the similarity between textures appearing on illuminated paper surfaces and textures those are made by pigments depositing on paper surface by a drawing stroke. First, we capture 12 paper textures with shading from a paper surface. The paper surfece is illuminated from 12 directions by turning the paper surface each 30 degrees and photographed. These textures are corrected and defined as a Height Field. Next, a stroke is generated by the control points that is input by a pressure-sensitive tablet. By using paper textures along to the drawing direction and using Height Field values along to the pen pressures, we can simulate pigment depositions like charcoals, pastels, and crayons in real-time.

View full abstract

Tearing an elastic object on computer graphics with physical simulation is difficult to get a desired output because of the difficulty of expecting the part where has stress concentration. On the other hand, setting up key frames to all the points on the object, which is completely under control, is complicated and time-consuming work for creators. In this paper, we propose a method to represent a tearing of an elastic object with physical simulation and with a desired output. A creator sets a “tearing surface" where the elastic object is torn along. This simulation works on real-time so that creator can vary parameters such as external force, which allow them a trial and error process. We use Matthias Müller's elastic analysis model for a real-time elastic simulation.

View full abstract

In this paper we describe an animation system that can map a person's facial motion to a wide selection of realistic face models in real-time. The motion is obtained from a motion capture system that measures 3D positions of infra-red LED markers placed on a subject's face. Using a 3D laser scanner, we also scan nine predefined postures specific to speech production for the same subject. Target faces are generated from 3D mesh points also measured with a laser scanner. Transformation between the motion and static postures is computed based on linear mapping and PCA (Principal Component Analysis). With this method only a small number of parameters are required to generate facial animation: three parameters corresponding to the dominant principal components to control face motion, and six parameters to control rigid head motion. By reducing the parameter space and distributing the processing between two networked computers, motion capture processing, parameter transformation, and high-quality realistic facial animation synthesis is made possible in real-time.

View full abstract

This paper describes a method to bring 2D photographs of faces to life, addressing the problems of creating realistic 3D models from 2D data and making credible deformations of the models for use in animation. First, a realistic 3D model is created from photographs using a statistical approach based on analysis of faces in a large 3D face database. This method has the advantage of incorporating 3D structure information from similar faces found in the database into the resulting face. The database includes over 200 people with 9 different 3D face postures, analyzed by principal component analysis (PCA). A small set of feature points from the face and the profile silhouette line extracted from front and side view photographs are used to create the new 3D face from the PCA results by linear estimation. Any new neutral 3D face can be easily represented by fifty eigen vectors obtained by this PCA. Next, speech-related face expressions are created for this model where the deformation characteristics can be estimated from faces in the database that are close to the input face in the eigen vector space. The estimated facial expressions are quite natural. As we show here the same method can also be applied to the faces of statues or dolls, or other human-like representations. To create an animation, results of additional principal component analysis of the estimated faces are used.

View full abstract

In this paper, we propose a shape similarity comparison method for a polygon soup 3D model that employs a multiresolution representation of the 3D model. The method first produces a set of multiresolution representation of the shape by using the 3D alpha shapes algorithm of Edelsbrunner. One of many single resolution shape features is applied to the model at each resolution level to produce a set of multiresolution shape feature for the shape similarity comparison. This study experimentally evaluated the efficacy of the multiresolution shape comparison approach by using four different single-resolution shape features and two different databases. Experiments showed that (1) the multiresolution approach does improve the performance of all the four shape features tested, and (2) the multiresolution version of the Surflet Pair Relation Histogram performed the best among the combinations tested.

View full abstract

In medicine, detailed observations of a three-dimensional fundus shape is necessary for diagnosing eye diseases accurately and checking the treatment process. In the paper, we propose a method for reconstructing fundus volume data accurately taking into account a fundus shape from cross-sectional images scanned radially with an optical coherence tomography (OCT). The method determines the position of cross-sectional images with a genetic algorithm and interpolates volume data from the cross-sectional images with displacement vectors. The fundus shape and its internal structures can be easily observed using the proposed method. The method is a promising technique for the diagnosis of glaucoma.

View full abstract

We propose an video coding techniques which enables efficient long-term archiving of surveillance video. MPEG-4 shape coding is one of the way which can achieve significant increase in storage capacity, although, automatic segmentation is very difficult. False positive error reduces the compression rate and false positive spoils the image quality. In this paper, we improve the intruder detection and segmentation system in, reducing low-false alarm, repairing the shape of the segment and interpolating the segment in temporal-manner. Experimental result showed that our object coding achieved in high compression rate 15% compared with standard MPEG-4 MC+DCT and low false negative rate in less than 1%.

View full abstract

Higher order local autocorrelation (HLAC) features are basic features which are used in various applications. They have been restricted up to the second order and are represented by 25 mask patterns. We increase their orders up to eight and extract the extended HLAC features using 223 mask patterns. Furthermore, we create large mask patterns to support large displacement regions. We use masks of different sizes together and construct multi-resolution features. In texture classification, the proposed method outperformed other methods such as Gaussian Markov random fields, Gabor features, and local binary pattern operator. For example, in the classification of small 32 × 32 texture images, the proposed method achieved a 97.5% recognition rate compared to 93.4% using Gabor features. Good performance was also shown for classification of scaled images and 90-degree rotated images, and for classification of numerous classes, over 300, using only a few training samples. The proposed method performed well in face recognition as well. It achieved a 98.4% recognition rate, as opposed to 96.0% using Gabor features. This result indicates its applicability to objects with shape.

View full abstract

In this paper, we propose a method for lossless Hadamard transform. Firstly, it is shown that 2nd Least Significant Bit (LSB) plane coincides with the binarized pattern of transform basis. Secondly, we make clear the characteristic that when the LSB plane and the 2nd LSB plane of Hadamard transformed coefficients are inversely transformed, each bit plane is converted to integer value.
By using of these characteristics, we propose a new lossless Hadamard transform named a partial retransform method. In this method, Hadamard linear transformed coefficients are divided into an integer part and a decimal part. And, only decimal part is retransformed. Then, the retransformed result of the decimal part is added to the integer part. Processing the lossless transformed coefficients with same method as above, original data are reconstructed.
By applying the partial retransform method to 2D 4×4 points Hadamard transform, we construct a new lossless 2D 4×4 points Hadamard transform.

View full abstract

A method to estimate the relative position and orientation of a camera to a circle having unknown radius is presented. In landing approach of a helicopter, the human pilot estimate his position and orientation relative to the landing site by observing the mark drawn on the heliport. The proposed method is formulated on the framework of the recursive estimation process using an image sequence of a circle that is a part of the heliport marking. The experiments were made both with the synthetic data and with the real image sequence taken in the laboratory for investigating the accuracy of the proposed method.

View full abstract

Scene understanding of photographs is very important for photographic applications, such as optimal photo finishing, contents-based retrieval and categorization, and so on. It includes technologies, such as recognizing what the main subjects are, inferring in what situation the photographs were taken, and so on. The first step to do this is to recognize what objects are on a photograph and where they are. In this paper, we introduce a new method to recognize typical objects on an outdoor photograph. Experimental results show that our system is effective to recognize these objects.

View full abstract