US 5,521,857 A · Process and device for the analog convolution of images

Abstract

A process and device for the analog convolution of images, especially two-dimensional convolution operations of Gaussian type on a set of image pixels represented by voltages. The voltages are applied to a first set of capacitors. For each capacitor, one terminal is connected to a reference voltage; the other terminal is part of a network of nodes, each of which is insulated from the others. The nodes which are adjacent are connected together for a predetermined duration via a second set of uncharged capacitors. The nodes of the network are then insulated from each other again and the second set of capacitors are discharged. The steps of connecting the nodes together, insulating them from each other, and discharging the second set are repeated. Finally, the new voltages on the nodes of the network, which represent the pixels of a convolved image, are read out.

Claims8

Dependent claims

Process for performing recursive especially two-dimensional convolution operations of Gaussian type on a set of pixels of an image, each pixel being represented by an electrical voltage, where the process comprises the following steps:

a) a plurality of electrical voltages representing image pixels are applied to the terminals of a plurality of first capacitors (Cd), first terminals of which are connected to a reference potential and second terminals of which constitute a plurality of nodes (N(i,j)) of a network and are electrically insulated from one another,
b) the nodes which are immediately adjacent in the direction or directions of the network are connected together for a predetermined duration, via second, uncharged, capacitors (Cb),
c) the nodes of the network are insulated from one another anew and the second capacitors are discharged by short-circuiting together the terminals of each second capacitor,
d) steps b) and c) are repeated, and
e) the new electrical voltages which are present at the nodes of the network and are representative of the pixels of a convolved image are read out.

2 more elements

Device for performing recursive convolution of images, especially two-dimensional convolution operations of Gaussian type on a set of image pixels represented by voltages, the device comprising:

a network of first capacitors (Cd), a first terminal of which is connected to a common reference potential and a second terminal of which is able to receive an electrical voltage representative of an image pixel,
a set of second capacitors (Cb) respectively associated with each adjacent pair, according to the direction or directions of the network, of first capacitors,
controlled interrupter means (Q1, Q2;
Q1, Q1', Q2, Q2') capable, for a first predetermined period, of joining up the terminals of each second capacitor (Cb) to the second terminals of the associated pair of first capacitors (Cd) and, for a second predetermined period following on from the first, of insulating the said second terminals from the associated pair of first capacitors and of short-circuiting together the terminals of each second capacitor (Cb).

1 more element

Dependencies are inferred from the claim text.

Process and device for the analog convolution of images

Life of this patent

Process and device for the analog convolution of images

Life of this patent

Abstract

Claims8