F. Javier Toledo

A retinomorphic architecture based on discrete-time cellular neural networks using reconfigurable computing

This paper describes a novel architecture for the hardware implementation of non-linear multi-layer cellular neural networks (CNNs). This makes it feasible to design CNNs with millions of neurons accommodated in low price FPGA devices, being able to process standard video in real time. This architecture has been used to build a CNN-based model of the synapsis I of the fovea region, with the aim of implementing the basic spatial processing of the retina in reconfigurable hardware. The model is based on the receptive fields of the bipolar cells and mimics the retinal architecture achieving its processing capabilities.

Implementation of a discrete cellular neuron model (DT-CNN) architecture on FPGA

This paper explores different alternatives to carry out a model of digital CNN from the point of view of its implementation on FPGAs. It shows the developments of four different DT-CNN models obtained from different transformations made to the original continuous model of CNN. Next, each discrete approach is simulated and compared with the rest of approaches and the continuous models. The objective of this study is to find the approach which best emulates the continuous neuron model at minimum computational cost. The simulations and temporal analysis of the discrete models have been made both in feedback and open system in order to verify their functionality. Finally, the architecture of the best model is implementated on an FPGA obtaining very interesting results.

Study of the contrast processing in the early visual system using a neuromorphic retinal architecture

In this paper we propose a retinal architecture that incorporates the neural circuits found in the different retinal regions. It is implemented in a reconfigurable system for observing in real time the contrast processing capabilities of each retinal region over the provided stimuli. The retina model is based on a discrete-time cellular neural network (DTCNN) that will be implemented on reconfigurable architecture (FPGA) with a time multiplexing approach. This architecture is able to incorporate 50 million neurons in its structure for processing video in real time. It has been observed that the contrast detection and the detail resolution are influenced by the convergence factor of neurons and by the lateral inhibition, specific characteristics of each neural circuit.

Model and hardware emulation of the first synapse of the retina using Discrete-Time Cellular Neural Networks

A retinal model and its implementation on reconfigurable hardware are proposed in this paper. The model incorporates the neural circuits found in the different regions of the first synapse of the retina. The model is based on a discrete-time cellular neural network (DTCNN) approach. The implementation on reconfigurable hardware makes it possible to carry out in real time the processing tasks implied in the model execution. Like in the first synapse of the retina, it has been observed that contrast detection and detail resolution are influenced by the convergence factor of neurons and by the lateral inhibition, which are specific parameters of each neural circuit.

An augmented reality visual prothesis for people affected by tunneling vision

An augmented reality based visual prothesis for people affected by tunnel vision is described in this paper. Using a head mounted display, useful information of the environment is superimposed on the own patients view, to enhance the users knowledge of the environment. That information is extracted by processing the images obtained from a camera. In order to provide versatility to the system and to accomplish the requirement of high performance, the image processing algorithm is performed with a cellular neural network, implemented on an FPGA device. The discrete CNN model proposed is derived from the continuous model, and details of its hardware implementation and the overall system architecture are reported.

Image processing with CNN in a FPGA-Based augmented reality system for visually impaired people

A cellular neural network is proposed as the main processing core in a novel FPGA-based augmented reality system. The described application is focused on visually impaired people aid. The aim is to enhance the users knowledge of the environment with useful information extracted by image processing. A CNN architecture oriented to hardware implementation on FPGA is presented, and used as the image processor in a fully FPGA-based system. So, CNNs and FPGAs are combined in a system which makes the most of their characteristics to achieve high performance and versatility.

Acceleration of a DWT-Based Algorithm for Short Exposure Stellar Images Processing on a HPRC Platform

Our objective is to provide an enhanced algorithm for the FASTCAM instrument, developed by the Instituto de Astrofísica de Canarias in collaboration with the Universidad Politécnica de Cartagena. In this paper we propose an algorithm for the detection of astronomical objects and its implementation on a High Performance Reconfigurable Computer. Our algorithm introduces wavelet based preprocessing and post-processing stages that considerably enhance the image quality when compared to the initial algorithm.

HANNA: A Tool for Hardware Prototyping and Benchmarking of ANNs

The continuous advances in VLSI technologies, computer architecture and software development tools make it difficult to find the adequate implementation platform of an ANN for a given application. This paper describes HANNA, a software tool designed to automate the generation of hardware prototypes of MLP and MLP-like neural networks over FPGA devices. Coupled with traditional Matlab®/Simulink®environments the generated model can be synthesized, downloaded to the FPGA and co-simulated with the software version to trade off area, speed and precision requirements. The tool and our design methodology are validated through two examples.