Texture mapping is one of the key techniques for creating realistic images of 3D scenes. Mapping a real texture image to a broad surface in a 3D scene, we must expand the real texture image. To expand a texture image, one may repeatedly paste the copies of it or its folded versions. On the expanded image by the above method, however, the seam lines among the copies are conspicuous. We propose a novel method for expanding texture images. The method is cutting out a patch from the original texture image in optimum shape at optimum position, and pasting the copies of the patch repeatedly. Cutting out the optimum patch, dynamic programming (DP) is used so that the seam lines among the copies of the patch are inconspicuous (the differences of gray values between both sides of the seams are minimal). Experimental results show that the method proposed here is available for natural texture images with week periodicity.
When we recognize the shape of an object from an image to store in a computer, we use the pattern matching. The typical pattern matching technique is detection of a similarity of the shape of images. But, these techniques can not analyze a difference in the shape. This paper show the features of the shape by using numerical values, and take notice of morphology. The features of the shape by morphology is technique to describe pattern spectrum. This technique is more conscious to the shape of the image than other pattern matching technique, when extracting a different area from the texture image. At first, we express the features of the shape by pattern spectrum to use binary morphology. This pattern spectrum value has disposition that depends on the resolution, the shape of structuring element, and turning of an image. Next, we extend a gray scale image, and we consider its pattern spectrum to discriminate different areas of a shadow mask image. Analyzing different areas from a shadow mask we can detect a defective part of a repeating pattern on a grey scal image.
In the image understanding process, observed physical values are transformed into high-level concepts. The knowledge used in this transformation is vague and context dependent. Therefore, it is difficult to capture and represent this knowledge in a computer based image understanding system. The same problem exists in an image creation system such as a computer graphics system, which transforms high-level concepts into numerical values. Conceptual Fuzzy Sets (CFS) implemented on a multi-layered bi-directional associative memories system, have been proposed to process such concepts and to derive a knowledge building method capable of addressing context depending issues. The basic mechanism of this method is inductive learning. On this paper we illustrate this method on the problem of modeling facial expressions. Computer simulation results show the suitability of this method for (1) recognition and (2) automatic generation of facial expressions. Furthermore, with respect to (2) a learning method from linguistic instruction, based on CFS is proposed. Its effect is a better modeling of the user's mental image. Computer simulation results showed that the learning is simpler than conventional linguistic learning, and can refine the knowledge to create desirable facial expressions.
In this paper, the authors describe a method for the identification of human faces. In this method, the fiber grating vision sensor which has been developed by the authors is employed for the three dimensional shape of the faces. Before the identification of the face using the three dimensional shape of the face, it is necessary to calibrate the position and direction of the facial data. In this method, a set of the directions of normal vectors at data points in the facial surface is obtained, and calibrations are carried out in accordance with the extend of errors in the sets. To identify human faces, a multi-layered neural network is used in which the inputs are two component values of normal vector at data spot in the facial surface. The experiments using the experimental system in performed to demonstrate the efficacy of this method and the experimental results are shown.
This paper proposes a new framework of automatic interpretation for equipment drawings, which consist of lines, symbols, and characters, such as electric drawing distribution diagrams. There are three stages. First stage is segmentation and recognition for primitive components. Second stage is line interpretation using probabilistic relaxation. The last stage is elimination of ambiguity using fuzzy integral. When interpreting lines, there are two difficulties. First, symbols and characters, which are important information, cannot be recognized with complete accuracy. Second, a meaning of a line must be decided, not only by analyzing neighboring objects, but also by analyzing global relationship. Therefore, it is difficult to interpret each line in a deterministic way. In order to solve these difficulties, an interpretation problem was newly formulated for lines as a probabilistic relaxation labeling problem. Only relaxation method has a limit to accuracy of line identification. It is difficult for the relaxation method to deal with local structures. Ambiguous labels of lines are determined by fuzzy integral based on analyzing local structural features. An interpretation system has been developed and applied to several electric distribution diagram. Experimental results showed that over 85% of the lines were labeled correctly.
A problem in realizing a vehicle which autonomously travels along a guide line on the floor in an in-plant environment by its vision is the gloss of the floor which obstructs the recognition of the line. The gloss is caused by interface reflection from object surfaces. This paper proposes a method to obtain images of an interface reflection component and a body reflection component when the shape of object surfacs is known. This method is based on the Dichromatic Reflection Model and utilizes the difference in Fresnel reflection coefficient between parallel and perpendicular components of polarization. No knowledge on the number and disposition of light sources is required. A vision system based on this method has been developed using a liquid crystal cell and its performance was examined. By applying this system to the vision of a vehicle, the guide line was found to be reliably recognized from the image of the body reflection component even if the floor was glossy due to illuminations. Thus, the effectiveness of the proposed method wan confirmed.
Volume measurement of automotive engine parts must be carried out with high accuracy. For quick measurement, these parts must be measured without covering the hole, because they have depressions with holes. A method for measureing the volume of these parts without covering the hole has been developed. The developed method has the following three features: (1) The volume is measured by scanning the figure of the object using a 3-D vision sensor with structured light. (2) Based on the figure obtained from the 3-D vision sensor, a virtual lid is placed on the hole. (3) For accurate volume measurement, the position of the virtual lid is determined from the design CAD data. In order to measure the volume of automotive engine parts by this method, a new 3-D vision sensor has been designed. With the new sensor, we have constructed a new volume measurement system for the automotive engine parts. Using this system, a measurement accuracy of ±0.1% and a mesurement time of 5 seconds are obtained, which indicates sufficient performance for practice application.
In order to recognize stamped characters, conventional methods measure them as binarized gray-level image using camera. Then, it is difficult to get exact image of stamped characters, because gray-level noise is generated on the image by uneveness of stamped surface. This paper presents a new approach to measure stamped characters as range image using range finder system for the recognition system. We analyze the characteristics by comparing gray-level image with range image, and discuss the effectiveness using actual sample data.
A recent trend of printed circuit boards (PCB) shows high density of electric devices and mechanical devices. And, solder joint fatigue and electrical contact failure on the PCB are caused by shock and vibration to PCB. Therefore, as the reliability evaluation of PCB, the vibration measurement in become one of the most important subjects. The authors developed a new measuring system (HPMS) for thermal deformation measurement of PCBs, which is composed of both techniques of holography and graphic image processing. In this study, the HPMS was applied to the vibration measurement of PCB due to operation of mounted electromagnetic relay. From the experimental results, the relationship between vibration pattern and fixed condition of PCB was made clear by the 3-D graphic image. And, the stroboscopic holography was proved to be a good measurement method for time division vibration pattern measurement of PCB.
A visualization and a quantification technique of a latent electric charge image on a dielectric material have been developed by using Pockels effect and a computer image processing technique. The Pockels cell of a BSO (Bi12SiO20) single crystal was used as a transducer to transfer the surface charge density to retardation of transmitted light beam. In order to reduce a spatial noise of light beam, an optical phase modulation technique and a computer image lock-in amplifier processing were used. The measurement system has several improved performances; the sensitivity of surface charge of 1nC/cm2, the spatial resolution of 200μm and discrimination of the charge polarity. The basic principle of advanced electrooptic surface charge measurement is described and the typical surface charge distribution of positive discharge streamer produced by an impulse high voltage (5kVpeak) is demonstrated. The measured surface charge images are shown on a color computer monitor as pseudo color in three-dimensional beautiful figures.
In a current experimental automatic navigation system, an Automatic Radar Plotting Aids (ARPA for short.) is only a device to acquire navigation information about ships in a radar visibility. We are developing image processing techniques for calculating the distance to a target ship based on processing scenery pictures, which has originally been obtained from the visual images and used as navigation information by ship operators. This scheme is realized by measuring a minute angle between a horizon and a waterline in a scenery image and computing the distance to a target. In this paper, we describe (1) a derivation of theoretical equations for calculating the distance to a target by means of the computed minute angle, (2) a verification of the efficiency of Mouse pointer ranging system, and (3) development of semiautomatic system assisted by image processing techniques, that is named Semiautomatic image ranging system. We have tested correctness of the equations and the accuracy of their calculations by applying 148 pictures and 214 images to the systems.
We developed a low noise detection system using an InGaAs-PIN photodiode for near infrared spectroscopic measurement. Since no practical photomultipliers are available in near infrared region, a Ge PIN photodiode cooled at 77K with a transimpedance (TIA) circuit and lock-in amplifier is usually used. InGaAs PIN photodiodes are more suitable than Ge PIN photodiodes for detecting low level light in the view of dark current and quantum efficiency. The detection system consists of an InGaAs PIN photodiode with a charge integrating amplifier operated at 77K. The minimum detectable power of 10-16 W was achieved at the wavelength of 1.28μm. The integration time was 10s. The system performance was tested by the measurement of the spectrum of singlet oxygen.
In the previous research, the authors proposed a method to generate high quality function-fonts for brush-written characters automatically. This method extracts sharp edges of boundary lines, so-called joint points, and approximates the boundary line between each adjacent joint points using piecewise polynomials. As the next step, we applyed the previous method for printed characters and figures. We find that they have another rules of quality and that the previous method cannot be adopted to these fonts. This paper aims to make clear how joint points should be determined for many kinds of fonts and to present a multi-stage algorithm to extract them exactly. The performance of the present method is verified by some experiments.
A three-dimensional vision sensor for inspection of painted automotive body dimensions has been developed. This sensor measures the flushness and gap between adjacent automotive body parts, such as the door and the fender, using the slit light method. In order to restrain the influence of body vibration, a pulsed slit light is projected onto the automotive body. Furthermore, both of the projection strength and width of the pulsed slit light are controlled to obtain constant bright images independently of body colors. Moreover, the optimum thickness of the slit light is selected to improve the quality of the slit light image of the black body with dusts and stains. The sensor measure the flushness and gap with accuracy of within ±0.1mm in measurement experiments using white and black automotive bodies and therefore proved effective for the inspection of painted automotive body dimensions.
Backpropagation algorithm (BP) having only a gradient term converges very slowly, because of the oscillation of weights occurring in regions where the error surface forms a ravine. In order to reduce the oscillation of weights, the momentum term was introduced. However, it has not worked well to reduce the oscillation because the gradient includes the component across a ravine which causes further oscillations. To overcome this problem, we should focus near the bottom of a ravine where the steepest descent direction is the same as the downward direction along a ravine. We described a method to correct the position of weights near the bottom of the ravine and proposed a new accelerated learning algorithm based on Jacobs' algorithm with it. The proposed algorithm reduces the oscillation of weights quickly and converges about 18 times faster than the standard BP in the problem of a sine function approximation.