On Image Keywords for Picture Database

本文提出了一种图像关键字替代文本关键字的算法。与人的五种感官相对应的图像有很多种。人们提出了许多图像压缩方法并对其进行了标准化，如JPEG等。但其目的是高保真地复制，它们并不直接满足我们的使用。重要的是能够扮演类似指南针或向导的角色。自由曲线可以用来表示图像图像。自由曲线的最佳逼近，即控制点的自动提取，是研究的首要课题。处理步骤为：降噪、边缘检测、细化、预处理后的自由曲线逼近线段。从设定适当的初始值开始，根据爬坡法对控制点进行振动，使其接近最合适的分辨率。因此，只需记忆控制点就足够了，所需的数据也比位图数据少得多。此外，它的优点是可以容忍全局转换和局部转换，例如仿射转换和扭曲。换句话说，这意味着我们可以提供高质量的线图像，即使屏幕尺寸大，尺寸小或有些扭曲。该方法已应用于浮世绘人物识别与分类系统中。该系统基于图像数据库文件中人脸的视觉特征。在本研究中，从图像中提取了直线轮廓，并用贝塞尔曲线进行了近似。通过计算机仿真验证了该方法的有效性。该方法可扩展为一种三维形状分析工具。

An algorithm for image key-words in place of usual text key-words has been developed here. There are many kinds of images corresponding to human five senses. Many methods for picture compression have been proposed and standardized, for example, JPEG and so on. But the purpose has been reproduction with high-fidelity, and they don't meet directly our uses. Important thing is to be able to play a similar role to a compass or a guide.Free curves can be used to represent picture images. The best approximation by free curves, that is, the automatic extraction of control points was the first main subject. The processing step is as follows : noise reduction, edge detection, thinning, approximation of line segments by free curves after the preprocessing. Starting with initial values set appropriately, the control points are vibrated and getting near to the most suitable resolution according to the hill-climbing method. As a result, it is sufficient to memorize the control points only, and the necessary data are much less than the case of bit-map data. Furthermore the advantage is in tolerance for global transformations and local ones, such as Affine transformations and distortions, respectively. In other words, it means that we can offer line images with good quality, even if the screen is big in size, small in size or distorted somewhat.The proposed method has been applied to a person identification and classification system of the actors drawn in Ukiyoe pictures. The system is based on visual features of the face from the image database files. Line profile from the picture has been extracted in this investigation and has been approximated by Bezier curves. The effectiveness of our method has been confirmed through computer simulation. The method developed here can be expanded to one of three dimensional shape analyzing tools.