IMPLEMENTATION OF AN EFFICIENT SEPARABLE VECTOR DIGITAL FILTER ON FPGA

Abstract

In modern video surveillance systems, in which the use of computer vision technology is widespread,
the most important information in the image is data on the contours of objects and the highlighting of small
details. The systems are subject to stringent requirements, such as: high speed of processing information
from a large number of cameras simultaneously, operation in conditions of poor lighting of the object and
under the influence of external noise (electromagnetic fields, short interference from high-voltage transmission
lines). Therefore, improving image processing methods using parallel computing devices and building a
multi-threaded system is an urgent task. In this work, a 3x3 anisotropic high-pass filter is designed and simulated
for image processing on an FPGA. An algorithm for its construction in the form of a separable vector
representation is described. A detailed description is given of the development of an effective separable twodimensional
digital filter for sharpening and highlighting the boundaries of objects in RGB images. The filter
is based on the synthesis of the proposed 3x3 anisotropic high-pass filter and the Sobel gradient filter.
The corresponding block diagram of the filter has been designed. Based on the results of processing the distorted
image, we can conclude that the developed filter has the property of more uniform detailing and high lighting of objects in the image and is less susceptible to Gaussian noise compared to the Sobel gradient filter
and the Laplace high-pass filter. A filter pipeline circuit has been developed on an FPGA for processing one
plane of an RGB image. Due to the use of separable filters, the proposed implementation is almost 2 times
more optimal in terms of the number of addition/subtraction operations performed than the direct implementation
of a 3x3 Sobel gradient filter and a 3x3 anisotropic high-pass filter.