We have already reported a PQS (Picture Quality Scale) for still image. In this paper, we will start to define a PQSmotion for moving image. Fundamentally, in the similar case of still image, a PQSmotion is defined as follows ;(1) a linear combination of essential distortion factors by princial component analysis method and multiple regression analysis method. (2) a combination of essential distortion factors by nural network. We defined distortion factors generated in a course of image codings considering visual and perceptual properties for moving image. We have made some experiment about distotion factors discribed above. Some defined essential distortion factors approximat well the mean opinion score.

View full abstract

The improvement of motion area detection method of images is discussed, which is necessary to convert interlaced pictures to progressive scanning. In motion-adaptive deinterlacing, miss-detection of motion causes a class of the conversion artifacts. To improve this artifact, a method which adaptively utilize field-difference in motion detection with conventional frame-difference is proposed. the proposed method can eliminate miss-detection in horizontally periodical images like vertical stripes without great increase of over-detection. The method is simple and easy for implementation. The effect of the method is confirmed in experiments and the converted picture quality is improved.

View full abstract

It is obvious that the multi-viewpoint image set has much redundancy. However, its redundancy has not been fully examined from the viewpoint of the relationship between the redundancy and the nature of the object, such as the shape or the characteristic of its surface. In this paper, we intend to answer the question by proposing a new scheme for representing the information of 3D space using "ray parameters". First, we give the definition of the ray parameters. Then, we present a model for generation and observation process of rays assuming the Lambertian surface model and pin-hole camera model. Based on this model, we examine the relationship among the nature of object, rays, and view images, in terms of ray parameter space.

View full abstract

In fractal image coding, an original image is encoded by searching for self-similarity in the image. An original image is divided into fixed size blocks and they are encoded independently. But natural images have self-similarity at various size regions, so using fixed size blocks in searching for self-similarity is not always efficient. In this paper, we propose a new fractal image coding scheme using blocks of variable shapes and examine its efficiency.

View full abstract

We investigate a moving object extraction technique from image sequences for image synthesis. We propose two stage approach. First, Assuming that the motion field of the object is representedby Affine parametric motion model, the region of the object which is initially pointed in the first frame can be extended by integrating image sequence. Second, the boundary accuracy is improved by the synthesized background, which integrates motion compensated frames. We examined the performance of this algorithm by computer simulations.

View full abstract

Here we investigate the unique advantages of our proposed Virtual Space Teleconferencing System (VST), with emphasis to facial emotion transmission and recognition. Specially we show that this concept can be used in a unique way of communication in which the emotions of the local participant are transmitted to the remote party with higher recognition rate by enhancing the emotions using some intelligence processing in between the local and the remote participants. In other words we can show that this kind of emotion enhanced teleconferencing systems can supersede face to face meetings, by effectively alleviating the barriers in recognizing emotions between different nations. Also in this paper we state about a concept known as virtual person, which is a better alternative to blurred or mosaiced facial images that one can find in some television interviews with people who are not willing to be exposed to the public due to personal reasons.

View full abstract

This paper proposes a new scheme for the image coding using wavelet transform. We treat transformed image with 2 resolution levels. Except the lowest frequency subimage, we classify subimages into 3 groups by their directions, then unify two groupes with smaller mean vector powers into a single one. The resulted two groupes are further classified into 3 classes by their vector powers, respectively. Then each group-class is vector quantized by using its own code book. The lowest subimage is coded by 2D-DPCM. This scheme gives superior coding characteristic than that of 3 directional-power classified coding in the range of rate near 0.5 bpp.

View full abstract

By operating on a binary black and white image with on orderded dither method, it appears, to our visual senses, to become clearer because of a pseudo halftane. This is because when we don't notice of all the details of the images, instead we see from a more global standpaint.we investigated our one-dimensional visual information processing model. We presented a theory and experimented results for a layered model which extended xeternal world, retina and brain. This paper propose our two dimensional visual infomation processing model. Applyed an aquivalent approximate method and represent an estimation of the model because of to set the restoration images of character 「犬」 and GIRL.

View full abstract

For restoring linearly blurred image with additive noise. DFT or DCT domain Wiener filtering techniques have been reported. However, each technique requires such an unrealistic assumption as the periodicity of image signal for the forner or the even symmetry at the image boundary for the latter. This paper proposes a DFT domain Wiener filtering technique which does not depend on the above assumptions, but insted incorporates linear prediction to get the information outside the image boundary naturally assuming the image being generated by AR(1) model. Image restoration experiments based on this technique show better results than by the existing DFT technique.

View full abstract

This paper addresses a problem of extraction of the structural motion information for compact image representation. In order to extract a meaningful scene structure from image sequence, global motion and region shape of moving objects are taken into consideration. Intraframe segmentation and moving regions are firstly detected with edges and optical flow information respectively. Secondly, moving regions are roughly separated into moving objects based on clustering the orientation and the amplitude of optical flow for each region of the intraframe segmenation. Motion segmentation can finally be obtained by iterated estimatiou of affine motion parameters and region reassignment according to a criteriou using Gauss-Newton estimation algorithm.

View full abstract

This paper proposes a new coding technique, termed "Region Support DCT", for an arbitrary shape coding. The technique is based on the conventional two-dimensional DCT, and, therefore, the decoding process is achieved by the conventional one as it is. The technique outperforms other techniques for arbitrary shape coding, such as SA-DCT.

View full abstract

Radiographic images obtained by the screen-film system have various noise called quantum mottle, structure mottle, and film granularity. It is well known that these noise affect the signal detectability. It is useful to analyze the statistical properties of radiographic noise in constructing of high performance x-ray imaging system. Authors present a random-dot model with triply Poisson process that can explain statistical properties of radiographic noise. Rigorous expressions of the mean transmittance and autocorrelation function of transmittance are obtained. Numerical examples show this model can explain the characteristics of experimental results in a practical screen-film system.

View full abstract

A method used for identification of individuals from the handwritings has been reported. Three parameters such as ρ, θ, and the accumulated value of which characterize one's way of writing are extracted from handwritings under test by Hough transform. Although our special interest is in the inclination of linear elements, all these parameters are analyzed to distinguish the true his handwritings from the others. The final decision of the writer is made by the computed normalized cross-correlation coefficients between the corresponding two character elements, one of which was written by the subject. To discuss effectiveness of the method, the probability of getting correct answers has been measured with the coefficient in the experiment using three kinds of handwritings, that is, (1) subject's handwritings which are tested to be, (2) handwritings by the other persons which were written without being displayed his handwritings, (3) handwritings by the other persons who were displayed his handwritings. Based on the measure, 93.2% correctness for data (1). 76.4% for data (3), and 100% for data (2) have been obtained in this experiment for identification of individuals from handwritings. The proposed method would provide a measure that is useful for identification of individuals from the handwritings.

View full abstract