Conversion of Resolution and Intensity using Cellular Neural Networks

使用细胞神经网络转换分辨率和强度

基本信息

批准号：
10650379
负责人：
TANAKA Mamoru
金额：
$ 1.41万
依托单位：
SOPHIA UNIVERSITY
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
1998
资助国家：
日本
起止时间：
1998 至 2000
项目状态：
已结题

来源：
https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-10650379/
关键词：
Cellular Neural Network Retina Information Image Quantization Energy Minimization Conversion of Intensity

项目摘要

The spatio-temporal dynamics by CNN(Cellular Neural Network) must generate nonlinear interpolative effects which have not been generated by a conventional image processing by sequential machine. In this research, a high quality secondary color image can be generated by using the CNN spatio-temporal dynamics for any resolution for both of a dense gradation image and an area intensity image. The dense gradation image is got from the area intensity image generated from the spatio-temporal dynamics of the CNN.Each cell of the CNN controls on-off state of the corresponding area element in pixel and then the intensity of each pixel can be controlled by changing the number of area elements per each pixel. That is, multi-value intensity can be generated by only using a set of 1-bit cells in the universal CNN.The number of bits per pixel for color density should be controlled efficiently by the number of pixels for a given input resolution because high density is required for low resolution and low density is enough for high resolution. This auto-resolution is very useful from practical viewpoints.The CNN can be used efficiently for the auto-resolotion because the number of inner cells in each pixel is controlled automatically, even if the number of quantized levels of inner cell is only 2 (binary). The problem is how the color density per pixel can be generated by using spatio-temporal dynamics by a set of inner cells in each pixel. The problem can be solved based on a conversion from spatial area intensity process to density process in each pixel. The CNN for auto-resolotion has hierarchical structure in which a global CNN consists of cells corresponding to the number of pixels in an image and a local CNN consists of real inner cells. The discrete time CNN has been designed as a chip by using Parthenon high level language developed by NTT.

CNN（细胞神经网络）的时空动力学必须产生非线性插值效应，这些效应尚未通过顺序机器的常规图像处理产生。在这项研究中，可以通过使用CNN时空动力学来生成高质量的辅助颜色图像，以实现密集分级图像和面积强度图像的任何分辨率。密集的渐变图像是从CNN的时空动力学产生的区域强度图像。CNN的每个单元格在像素中控制相应区域元素的On-Off态状态，然后每个像素的强度可以通过每个像素来更改面积元素的数量来控制。也就是说，多值强度只能通过使用通用CNN中的一组1位细胞来生成。颜色密度的每个像素的位数应通过给定输入分辨率的像素的数量有效控制，因为高分辨率需要高密度，低密度需要低密度，而低密度足以实现高分辨率。从实际的角度来看，该自动分辨率非常有用。CNN可以有效地用于自动蛋白，因为即使每个像素中的内部细胞的数量自动控制，即使内部细胞的量化水平仅为2（二进制）。问题在于如何通过每个像素中的一组内部细胞使用时空动力学来生成每个像素的颜色密度。可以根据从空间区域强度过程到每个像素中的密度过程的转换来解决问题。自动蛋白溶质的CNN具有层次结构，其中全局CNN由与图像中像素数量相对应的细胞组成，局部CNN由真实的内部细胞组成。通过使用NTT开发的Parthenon高级语言，CNN的离散时间已被设计为芯片。