A Local Invariant Feature Extraction and Description Method for Microscopic Image of Bacteria

  • Zhanshen Feng
Keywords: Microscopic Image, Distribution Features, Local Invariant Features Points

Abstract

In microscopic imaging of microorganisms, a wide range of factors can result in unclear images for
detection of microorganisms and difficulty in efficient target signal extraction in image segmentation.
There are various bacteria and many of them are non-rigid. They have irregular shapes and forms. Apart
from the common shapes of circle and oval, there are also other complicated shapes, including strip and
concavo-convex polygons. Their sizes also vary greatly. Besides, the difference between the edge of
bacteria and the background accompanied by the motionandchange in form of bacteria is slightly different
from that in the gray distribution between the bacteria and the background in the distribution graph of
bacteria.This paper combines the reasons that affect the microscopic imaging detection of microorganism
and clear image observation, studies the theories of local feature descriptors and scale invariance in
computer vision technology, and proposes a local invariant feature extraction and description algorithm for
microscopic image of microorganism. As the nearest-neighbor feature points are searched in image
matching, this method searches feature points through the random K-D tree and in order to avoid the
influence of mismatching on image registration, it uses RANSAC algorithm in the purification of matching
points and achieves high efficient extraction of target signal of microorganism. As shown in the
experiment test and analysis, it is proven that the algorithm presented in this paper has higher accuracy in
target signal extraction, and that it can obtain microorganism image with clear distribution of feature
points, which is very good for the observation and follow-up study of microorganism distribution image.

Published
2020-04-01