An object recognition system is presented that generates object models from image examples. It also searches for and recognizes objects in images, using the object models. An object model is described by segmented regions and the hierarchical relations between them. The system generates one object model from one image example and compares the two to generate a more general object model. This paper describes a method for describing an object model and generating it from a series of image examples, and searching for an object in an image. Also shown are results of an experiment to verify proper system operation.
This paper proposes a network model that learns multiple views of 3D objects and adaptively recognizes their transformed images. The model achieves view invariance by training bidirectional networks using examples of object views. When novel, transformed images of the objects are then given to the networks, the model performs a dynamic recognition process that iterates between estimating the view using the trained networks and aligning the estimated view with the input view. Computer experiments using gray-level images of 3D objects demonstrate the flexibility of the proposed model.
This paper proposes viewpoint estimation method intended to be used as the basic device to estimate user's intension in a visual environment. Our method is based on moment feature which represents non-local feature of image, such as the gravity center or the breadth, and has less computational cost and considerable robustness to noise. We also propose the form and use of marker suitable for our purpose. That is, putting an oval marker on the top of a user's head. Experimental result shows that our method has good performance.
This paper proposes a new real-time method for estimating human body postures from thermal images acquired by an infrared camera, regardless of the background and lighting conditions. Distance transformation is performed for the human body area extracted from the thresholded thermal image, in order to calculate the center of gravity. After the orientation of the upper half of the body is obtained by calculating the moment of inertia, significant points such as the top of the head and the ends of the hands and feet are heuristically located. In addition, the elbow and knee positions are estimated from the detected (significant) points, using a genetic-algorithm-based learning procedure. The experimental results demonstrate the robustness of the proposed algorithm and real-time performance (faster than 20 frames per second).
This paper describes a method for automatically approximating an unknown image transformation from an original image to its target image by the use of a sequence of several known image filters, where the target image is an ideally processed image made manually. This automatic construction problem is regarded as a combinatorial optimization problem for the ordering of image filters and a genetic algorithm is employed to solve it. A population of individuals whose chromosomes represent sequences of image filters is evolved. In generation iterations, each individual is evaluated based on the difference between the target image and the image processed according to the filter sequence defined by its chromosome. This method can deal with arbitrary image transformations, and it can be applied to expert systems for image processing.
In the field of computer graphics (CG) it is necessary to design many kinds of models and develop distinctive behaviors for artificial creatures. This paper suggests a new method for easily creating many different shapes of artificial creatures in plastic art. We mainly propose a novel kind of Evolutional Modeling based on a conventional genetic algorithm. In evolutional modeling, plentiful descendants are produced using multiplication, crossover, and mutation operations. The final goal of plastic art can be achieved choosing some of these descendants based on the merit of balance. Evolutional modeling can automaticallygenerate novel and different ways of creating art.
Higher-quality visual contents are required for use in broader-band networks and higher-quality visual media, and information structures characterizing the “amenity” of visual contents have become a significant research theme. We developed a method for assessing the amenity of synthetic texture by referring to the results of studies on audio-visual quality. The method combines an a-EEG indicated physiological assessment with a psychological assessment by means of Sheffe's paired comparison method. Experiments in which three synthetic textures were evaluated showed that α-EEG potentials are significantly higher for textures rated higher in the psychological assessment. It is suggested that the assessment method is effective for evaluating the amenity of synthetic texture, and that we can use the texture called a “fractal pattern” as a sample for calibration.
An alternating voltage driving method (AVDM) for gray scale drive technology is proposed that utilizes the low-pass filter characteristics of the TFT-LCD to eliminate periodical components from alternating voltage and applies DC voltage which is represented as an average value to a pixel. This paper describes the structure of the AVDM six-bit driver and the logical design techniques used to create its output circuit, which is a voltage interpolating circuit. From evaluation of the low-pass filter effect, it is proved that most of the effect is produced by the time constant of the TFT and the pixel itself. It is confirmed that voltage deviations due to the line resistances in the driver die are a principal cause of the output deviations along with the low-pass filter effect. Design criteria for the voltage deviations are described. Last, the multiple AVDM is proposed for non-linear interpolation. A logical combinational method is described for applying it to a six-bit driver that needs 5 gray scale voltages.
In order to consider the effects of spatial attention on human saccadic eye movements, we analyzed peak velocity, duration, latency and amplitude of saccades during Kanji-recognition tasks. The results suggest that the choice of region allocated for spatial attention has no effect on the characteristics of saccades. The distributions of saccadic latency showed 1 to 3 peaks, and their peak values were consistent with the results of earlier studies. Because of these facts, we consider that there are basic phases for making saccades regardless of the complexity of the tasks. The relation between saccadic latency and amplitude shows that a longer latency leads to a smaller amplitude. This suggests that the state of spatial attention before execution of a saccade affects its amplitude. Based on these results, a schematic model of the brain system for generating of saccades is proposed, and the process of generating saccades is hypothesized using this conceptual model.
This paper presents a method for constructing object models from a single sketch image and optimizing their shapes. The sketch images used correspond to style sketches and all figures are drawn in three-point perspective, in which all lines ultimately converge on a common vanishing point. It is assumed that these images are drawn on paper by industrial designers. Basic models represented by Bezier curves are constructed based on straight homogeneous generalized cylinders (SHGC), using a contour image generated from the sketch image. Next, the shapes of the basic models are optimized by moving the control points of the Bezier curves with the use of genetic algorithms (GA). As a preliminary experiment, four methods of shape optimization are applied, using the same generation size. It is shown that the best of the preliminary methods may also be used for optimization of another kind of object model.
It was investigated whether the existence of a fixation point and saccade target influences the dynamics (peak velocity and duration) of voluntary saccades. In addition, the difference in dynamics between visually guided reflexive saccades, which reflexively occur oriented forward a suddenly appearing target, and visually guided voluntary saccades, which were voluntarily elicited toward the target, was investigated. Five experiments were conducted on young subjects in darkness. The results gave new findings as follows. (1) Independent of the fixation point the existence of the saccade target increases the peak velocity and reduces the duration of the voluntary saccades. (2) Independent of the existence of the saccade target the existence of the fixation point does not influence the dynamics of the voluntary saccades. (3) The visually guided reflexive saccade is larger in peak velocity and shorter in duration than the visually guided voluntary saccade.