RESEARCH ON SEPARATION OF ARTIFICIAL AND NATURAL IMAGES

如果我们能够将人工图像和自然图像分开，即使场景可能因落叶等而随季节变化，机器人也可以轻松导航，因为那些随季节变化的图像是自然图像，而那些不随季节变化的是人工图像。在本课题中，我们注意到自然图像的分形维数是非整数的，而人工图像的分形维数是整数的，因此我们提出了ASIC的方法和概念设计，我们所做的工作和取得的成果是：（1）利用盒计数法计算分形维数; (2)以下两种方法是针对二值图像开发的。[静态分离方法]将目标图像划分为网格。分形维数是逐段计算的，这是一组网格。这些局部段按分形维数的值分组在一起。[动态分离法]将dx向x方向移动，dy向y方向移动，计算分形维数。通过局部段的大小以及dx和dy的大小，有时会发生重叠。如果发生这种重叠，则以像素计算分形维数，并使用这些值的平均值。(3)本文将上述方法推广到彩色图像，计算彩色图像的HSL（Hue，Saturation，Lightness）直方图，并在直方图的基础上进行峰值分割。对于每个颜色段应用与二值图像相同的过程。(4)我们开发了ASIC的概念设计，上述方法也可以应用于人工和自然声音之间的分离问题。

If we can separate artificial and natural images, a robot can easily navigate even if the scene may change from season to season because of the falling leaves, etc. , since those images which change from season to season are natural ones and those which do not are artificial ones. And we can easily detect cracks in a very complex structure without any prior knowledge because cracks are natural in nature.In this work, we noticed that natural images have non-integer Fractal dimensions and artificial ones have integer dimensions and developed the methodology and conceptual design for ASIC.What we have done and achieved in this project are(1) We calculate Fractal dimension using the box counting method.(2) The following two methods are developed for binary images.[Static separation method]Divide the target image into meshes. The Fractal dimension is calculated segment by segment, which is a group of meshes. These local segments are grouped together by the value of Fractal dimension.[Dynamic separation method]The Fractal dimension is calculated moving dx in the x direction and dy in the y direction sequentially. By the size of the local segment and the size of dx and dy, overlap sometimes occurs. If such overlap occurs, Fractal dimension is calculated in terms of pixels and the average value of these valuse will be used.(3) We extended the above method to color images.We calculate the histogram for HSL (Hue, Saturation and Lightness) of the color image.Then it is divided by peak based on the histogram. The same procedure as the binary image is applied for each color segments.(4) We developed conceptual design for ASICThe above methodology can also be applied to separation problems between artificial and natural sounds.