Image Compression Regeneration and Depth Extraction by Silicon Retina

本报告描述了细胞神经网络（cnn）在信息编码和解码中的应用，特别是在运动图像的情况下。用广义CNN状态方程描述了编码（C-）和解码（D-） CNN的动态。C-CNN通过分辨率压缩和半色调对图像进行编码。D-CNN通过重构过程对接收到的数据进行解码，从而几乎识别出C-CNN的原始输入。在C-CNN中进行动态量化，从相邻的值中确定每个像素的二值。为了减小原始灰度图像与重建的半色调图像之间的误差，从能量最小化的角度解决模板合成问题。从拓扑学和正则化理论的角度讨论了分辨率压缩模板综合问题。通过动态电流分布在D-CNN中重新生成结构压缩图像。C-和d - cnn嵌入的通信系统由一个差分发射器和一个反馈回路中的内部接收器模型组成。此外，本文还描述了CNN对双目立体视觉信息的动态深度提取。对传输的漏斗半调图像重构的模拟图像进行模式识别，可以解决对应问题。使用竞争性的CNN。最后，本文描述了可分辨的细胞神经网络，通过该网络可以基于时空动态进行广泛的图像应用，为任何分辨率的全彩色输入生成三重[红绿蓝（RGB）]组合二次色。利用每个像素内细胞的局部动态生成副色的面积强度，利用全局CNN动态对副色输出和全彩色输入进行彩色图像处理。

This report describes the use of Cellular Neural Networks (CNNs) for information coding and decoding-especially for the case of moving images. The dynamics of the coding (C-) and decoding (D-) CNNs are described by generalized CNN state equations. The C-CNN encodes the image by resolution compression and halftoning. The D-CNN decodes the received data through a reconstruction process so as to almost recognize the original input to the C-CNN.A dynamic quantization is performed in the C-CNN to decide the binary value of each pixel from the neighboring values. In order to reduce the error between the original gray image and reconstructed halftone image, the template synthesis problem is addressed from the viewpoint of energy minimization. The resolution compression template synthesis problem is discussed from the viewpoints of topological and regularization theories. The structurally compressed image is regenerated in the D-CNN by a dynamic current distribution.The communication system in which the C- and D-CNNs are embedded consists of a differential transmitter with an internal receiver model in the feedback loop.Also, this report describes dynamic depth extraction for binocular stereo visual information by CNN.The correspondence problem can be solved by pattern recognition for analog images reconstructed from the transmitted funneling halftoning images. The competitive CNN is used.Finnaly, this paper describes resolutionable cellular neural network by which a wide range of image applications can be done based on spatio-temporal dynamics to generate triplet [Red Green Blue (RGB)] combinational secondary color for a full color input with any resolution. Area intensity of the secondary color can be generated by using local dynamics of inner cells in each pixel and color image processing can be done by using global CNN dynamics for the secondary color outputs and full color inputs.