论文部分内容阅读
细胞神经网络具有局部互联结构和高速并行处理的能力,被广泛应用于图像处理.然而,一方面,现有处理方法大多采用固定模板,在处理实际复杂图像时难以得到较好的效果.另一方面,因传统CMOS工艺发展瓶颈,利用其实现大规模细胞神经网络变得不切实际.本文首先以人眼感知原理为基础,考虑图像中各个像素空间分布的影响,提出了一种新型阈值自适应算法,其克服了传统边缘提取算法的局限性.利用具有独特开关转换机制、非易失性和纳米级尺寸等优点的新型非线性两端电路元件忆阻器来解决CNN的硬件实现难题.将自旋忆阻器与细胞神经网络相结合形成忆阻细胞神经网络作为算法的硬件.最后,对彩色图像的边缘提取进行数值仿真,抗噪性检测,与传统边缘提取算法对比分析,并计算各算法边缘提取结果的FOM(figure of merit)值和峰值信噪比(PSNR),验证了基于像素空间分布的阈值自适应忆阻细胞神经网络在彩色图像边缘提取中的有效性.
However, on the one hand, most of existing processing methods use fixed templates, so it is difficult to get good results when dealing with actual complicated images. Another , Because of the bottleneck of the development of traditional CMOS technology, it becomes impractical to use it to realize large-scale cellular neural network.Firstly, based on the human eye perception principle and considering the influence of the spatial distribution of each pixel in the image, a new threshold self Adaptive algorithm that overcomes the limitations of the traditional edge extraction algorithms and solves the CNN hardware implementation problem by using a novel non-linear two-terminal memristor with unique switching mechanism, nonvolatile and nano-scale dimensions. Combining the spin-memorizer and the cellular neural network to form the memristor neural network as the hardware of the algorithm.Finally, the numerical simulation of the edge extraction of the color image, the detection of anti-noise and the comparison with the traditional edge extraction algorithm are calculated and calculated FOM (figure of merit) and peak signal-to-noise ratio (PSNR) of the edge extraction results of each algorithm verify that based on pixel space Adaptive threshold memristive effectiveness Cellular Neural Networks in the color image edge extraction.