Kazutoshi Koga

A handwritten character recognition system using hierarchical displacement extraction algorithm

A handwritten character recognition system using the hierarchical algorithm to extract displacement between a template pattern and an input pattern is proposed. In the proposed system, the displacement can be computed by Gauss-Seidel iteration derived from Euler-Lagrange equations of the energy functional, which consists of a correspondence error between patterns and a smoothness constraint of the extracted displacement. To extract both global and local deformations included in input patterns, the hierarchical structure is introduced. In computer experiments, the recognition performance is clarified. In addition, the relation between the ability of displacement extraction and the recognition performance when the correspondence error is used as the distance is discussed. Finally, we show that it is possible to improve the recognition performance by using both the correspondence error and the smoothness of the extracted displacement as the distance.

Determining motion fields under non-uniform illumination

Abstract Two realistic methods based on a generalized conservation equation of image brightness are proposed for the accurate determination of motion fields under non-uniform illumination. The conservation equation includes spatial and temporal derivatives of image brightness, a motion vector and a rate of brightness generation. The term for the brightness generation rate can represent non-stationarity of a brightness pattern during its motion under non-uniform illumination. By introducing an optical model and prior knowledge of non-uniform illumination the conservation equation can be solved. One of the two proposed methods assumes a stationary motion field under spatially non-uniform illumination. The other assumes local constancy of a motion field under temporally non-uniform illumination. The application to artificial and real image sequences confirms the usefulness of the proposed methods.

Measuring Surface Shape from Specular Reflection Image Sequence –Quantitative Evaluation of Surface Defects of Plastic Moldings–

A sensitive method for surface shape perception based on image sequence processing is developed. Utilizing a moving slit light source, defects on a plane plastic injection molding are visualized as deformation of the slit image through specular reflection. By considering the geometric configuration of the light source, a plastic test piece and a video camera, the distributions of surface inclination and surface undulation are evaluated quantitatively.

Extracting depth‐information from dynamic image by spatio‐temporal correlation analysis

Optical flow-field obtained by the analysis of a sequential scene derived by a relative motion between object and camera includes three-dimensional information (depth and/or shape). This paper develops a spatio-temporal correlation method for determining optical flow and attempts to extract the depth information on the basis of the analyzed optical flow field and the principle of the “motion stereo.” The method herein is improved in two critical points: (1) analysis of the plane wave propagation; and (2) algorithms for rapid calculation. It is demonstrated that the improved method and the theory of “motion stereo” are available for extracting depth information from actual scenes.

Spatial Filtering Velocimetry by Dynamic Image Processing

A flexible velocimetry based on dynamic image processing is proposed. Through an analogy of spatial filter technique and laser Doppler velocimetry, a raw dynamic scene is transformed to a scene superposed with a sinusoidal spatial pattern of gray-levels. Using temporal change of accumulated gray value of each frame, we can evaluate velocity information of moving particles by spectral analysis. The validity and usefulness of the velocimetry are confirmed using artificial images created by computer simulation. The effectiveness of the maximum entropy method in the spectral analysis is also emphasized.

Dissipative Structures is Nematic (MBBA)-Cholesteric (CN) Mixed-Liquid Crystal

Systematic studies have been carried out on the dissipative structures caused by the electrohydrodynamic instability in mixtures of nematic MBBA and cholesteric CN . Newly-observed structures have been classified into two types according to composition of the mixtures. The frequency vs. voltage phase diagrams of the structures have been obtained by detailed microscopical observations.

Velocity-Field Measurement of a Vortex by Dynamic Image Processing

A new dynamic image processing technique is applied to the measurement of the velocity of a vortex flow. Velocity-field of the vortex (i.e., the spatial distribution of speed and that of moving direction) is measured and visualized by pixel-based analysis calculating the mutual-correlation functions among temporal brightness changes at each pixel. It is shown that the dependence of the measured local speed V on the distance from the center of the vortex r is approximately V\varpropto(1/r), consistent with the results of pioneers in the field.

Detecting a velocity field from sequential images under time-varying illumination

A generalized approach based on the gradient method is developed. Two algorithms are proposed to detect velocity fields from sequential images udner non-uniform (spatial or temporal) illumination. According to preliminary knowledges of the illumination characteristics and of the velocity fields, practical and accurate methods are tested by utilizing simulation images and real image sequences. The usefulness and reliability of the methods are confirmed.

Determining optical flow from sequential images

An exact method determining the optical flow is presented by means of pixel-based mutual-correlation analysis of dynamic images. The mutual-correlation function is calculated between the temporal brightness change of the target pixel and that of its neighboring 16 (or 8) pixels. The local velocity of the target pixel is determined exactly through logical considerations using a reliable lag time estimated by the quadratic interpolation technique. The validity of the proposed method is confirmed by computer simulation of various dynamic images.

Optical Flow Analysis Based on Spatio-Temporal Correlation of Dynamic Image

We propose several methods for analyzing the velocity of moving particles in dynamic images. The methods are based on the analysis of temporal mutual-correlation between the time series of gray level changes at a target pixel and that of its neighboring pixels on an image plane. In this paper, we clarify that the temporal mutual-correlation functions can be understood as the intersections of a kind of spatial autocorrelation of an image irradiance function which is assumed to be rigid. According to this new understanding and assuming a Gaussian-like decrease in the spatial autocorrelation, we clarify a relation between the correlation-value and the lag-time of the mutual-correlation and velocity of a moving object. This relation results in an improved method for determining the optical flow. Our previous methods are useful only for the analysis of spherical particle motion. However, the improved one can be applied to more general objects. From the analysis of the artificial image sequences, it is suggested that the accuracy of the obtained result is less than 0.01 ±0.01 p/f in speed and 1.0 ±2.0 deg in direction. These results indicate that the improved method has ten times the accuracy of the typical method called the “gradient method.” The improved method is also evaluated by analyzing an actual dynamic scene and confirmed to have the capability to measure the two-dimensional velocity field (i.e., optical flow).