A filter was already proposed to filter interferences on impedance cardiographic signals, named scaled Fourier linear combiner (SFLC). At a first look at the filter power spectrum, one may believe that this filter can eliminate also the bias. In this work, we will study the effect of bias and spectral overlap on the SFLC, and its implication on filtering impedance cardiographic signals. We also have carried out some experiments to study these problems. We found that the bias alters the weights behavior, and therefore the filter output. We also discuss solutions to the problem of spectral overlap.

View full abstract

We discuss the uniqueness of 3-D shape recovery of a polyhedron from a single shading image, and propose an approach to uniquely determine the concave shape solution by using interreflections as a constraint. We show that if interreflection distribution is not considered, multiple convex (or concave) shape solutions usually exist for a pyramid with three or more visible facets. However, if interreflection distribution is used as a constraint to limit the shape of polyhedron (for a concave polyhedron), polyhedral shape can be uniquely determined. Interreflections, which were considered to be deleterious in conventional approaches, are an important constraint for shape-from-shading.

View full abstract

We describe a computational model of the precedence effect based on echo-avoidance. We propose that the precedence effect is caused by the inhibition of sound localisation, which depends on the estimated sound-to-echo ratio. A pattern of the aftereffect of an impulsive sound is adopted based on the results of previous psychoacoustic experiments and used for echo estimation. Using this model, the results of psychoacoustic experiments, e.g. equal-level and unequal-level paired click tests, Haas's tests, Franssen's tests and others, can be interpreted consistently. This model can also give critera for an available onset and an explanation of why a transient onset is required for the precedence effect.

View full abstract

A method called simulated breeding is applied to the problem of learning criteria for similarity-based image retrieval. Usually traditional genetic algorithms without any modifications are used in the simulated breeding methods ; however we propose a new model "pipeline-type genetic algorithms" that is more suitable for the simulated breeding methods, and also discuss genetic operators that are special to this model. We then show that this model can eliminate both time constraints and the restriction of population size, which are the disadvantage of the simulated breeding methods, and also demonstrate this model's powerful search capability and effective parallel processing. Finally we describe the application of this model to similarity-based image retrieval as a good man-machine interface.

View full abstract

Color perception has been studied in the psychology and phisiology, mainly with approaches of qualitive investigetion. On the other hand, quantitative study is seemed to be difficult because color perception is always dominated by subjective impresion. In this paper, we propose an artificial neural network model and attempt to realize the color illusions, e.g. color simultaneous contrast illusion and color appearance illusion. Based on this model, the geometry of color space can be investigated quantitively through the lateral inhibitory neural networks.

View full abstract

The visual sensitivity function for binocular disparity modulation, like the contrast sensitivity function, has bandpass characteristics. However, very little is known about its underlying mechanism. We show that surface reconstruction through regularization followed by detection of disparity gradient can account for characteristics of the disparity modulation sensitivity function. A neural mechanism for surface reconstruction based on regularization is discussed.

View full abstract

In this paper we will present a fuzzy face detector, KAOF. It is based on learning fuzzy faces by the method of Fuzzy Categorization and Classification (FCC). A set of images were used as examples for the definition of a fuzzy face. FCC will try to recognize faces in a new image by collecting evidence from these examples. KAOF demonstrates that FCC can be used in different applications by recursively defining concepts using examples and classify a new image by collecting evidence from these examples. Result of Kaof will also be given in this paper.

View full abstract

In this paper, we propose a grouping based method to solve the stereo matching problem. We employ intensity segments on scan lines of stereo images as primitives for matching, and formulate the problem by introducing a multi-stage grouping process of the intensity segments. A relation that should be satisfied between a matching pair of intensity segments and Some constraints, including the self-occlusion constraint, the uniqueness constraint, the disparity constraint, the minimum risk criterion ard the length ratio criterion, are established. A multi-stage grouping process is involved to make final decision.

View full abstract

In this paper, we have analyzed emotions perceived from speech by psychological experiments. In the experiments, subjects were asked to listen randomly presented speech vocalized with emotions (Neutral, Anger, Sadness, Joy, Surprise, Disgust), and select perceived one. The rate at which the emotion intended by speaker coincided with that perceived by listener was 51.5 %. Especially, The coincidence rate of 'Sadness' was high, but that of 'Joy' was low. As the result of analizing speeches for which more than half of listeners could perceive speaker's emotion, the rate which the speaker's emotions agreed with listener's ones was 73%. Then this value would be target of an artificial system that estimates emotions from speech. We have found characteristics dependent on emotions : it was difficult to distinguish between 'Anger' and 'Disgust' ; it was easy to identify emotions about 'Anger' or 'Joy' from speech of word whose meaning is very similar to the emotion. Furthermore, humans' judgement on emotion seems to be less dependent on experiment date.

View full abstract

The authors have used a retrieval method to investigate properties related to how humans process facial images. In the retrieval method, subjects retrieve a target image, such as a favorite or a happy face, from a database by making a compromise between what they want and what they are presented with. Although important features can be derived without the subject being aware of them, the subject must retrieve many faces and his/her awareness of features can not be discriminated. Using this method, emotionally expressed faces, such as a favorite face and a fierce face, were examined to see how clearly the human could retain and recognize them. Comparing the results of two retrieval experiments for the same task, the retrieved faces were similar for each one. We can conclude that ambiguous images corrcsponded with certain abstract images even if the image was not clear like a picture.

View full abstract

In judging the gender of a face, humans use color information of the face more than shape information of the face. We have additionally found that there are differences in accurate gender judgments between male faces and female faces, under different degrees of rotation in death. Subjects made many mistakes in judging the gender of male faces in the left profile. With female faces, in contrast, they made many mistakes in the right profile. We therefore think that viewpoint-dependence for gender judgments varies with respect to gender of the face and the rotation of the face.

View full abstract

When we use multimedia communicaton like video conference, the movement of user is restricted in front of the camera. So we propose an interface to track the face of user walking around a room. To track face image, we must detect face area. Face detection algorithm for multimedia communication must be fast and robust enough to track it in realtime. To realize such system, we combine informations of skin color and defference from background.

View full abstract

We propose the feature extracted from the direction edge images using Sobel filters for gesture recognition in this paper. We can recognize from connectivity of direction of movement of gesture. Direction edge feature is feature vector that elements are total number of pixcel satisfied with condition in 8 direction edge images. We can recognize a gesture from fluctuation of special feature vector. In this paper, we describe about direction edge feature and report some gesture recognition experiment results.

View full abstract

Subjective Contours are phenomena to perceive illusory lines or figures brightly. The neurons which respond to subjective contours were found in the visual cortex area V1 and V2. The existing physiological models use only the direction selectivity of the V1 and V2. This paper proposes a model considering the positions and the directions of illusory lines by the responses from the circular receptive fields of the layer 4C_β on the V1. The model explains the brightness differences as well as the subjective contours of Kanizsa's triangle.

View full abstract

We present a new recognition engine concept for PDA. The concept processes input in three ways according to the characteristics of handwriting and human memory ; (i) character recognition for careful writing, (ii) candidate selection for running stroke, rewritten, split, and mismemorized characters, (iii) character recognition with learning for cursive script, personal, and abbreviated characters. The stages are combined in a hierarchical system that uses both OCR and stroke order assisted recognition processing. We constructed a PDA simulator, and confirmed that the proposed concept offers satisfactory performance for many applications.

View full abstract

Our researcher's ultimate objective is to do development for an bi-directional translation system between JSL and Japanese. The observer of sign language will receive the information of visual language. There is not enough communication means for its language that deaf people can utilize currently. For the translation system support, we construct computerized dictionary. And this dictionary contains the function of automatically generation animation from the phonetic notation to synthesis by rule. In this report, as application of this synthesis by rule function, we made system to can do communication by animation with only data sending of several bytes on Internet.

View full abstract