Research And Optimization Of Rotation-invariant Vorticella Detection Based On Histograms Of Oriented Gradients

Microorganism detection through computer vision is of great essence in many scientific and industrial fields.Vorticella detection is always a challenging and focused topic among them.In this paper,by means of Histograms of Oriented Gradients(HOG)and Local Binary Pattern(LBP),and utilizing discrete Fourier transform(DFT),an effective rotation-invariant descriptor for Vorticella detection is designed and implemented.The main work and contributions of this paper are as follows:1)The basic principle and features extraction process of HOG and LBP are deeply discussed and summarized.Their defects are analyzed to find the possibility of improvement,and quantities of comparison experiments are performed to pursue more effective capacity of detection.2)The basic principle and manifestation of image rotation transformation are comprehended and analyzed.From the angle of preprocessing of vorticella images,through experimental comparison of expansion to the training set of samples through rotation of original vorticella images,more efficient performance were obtained,which increased the completeness of machine learning.3)By combining the two methods,HOG and LBP,this paper presents an efficient method to build rotation-invariant descriptors.The idea is dividing whole image into spatial blocks to obtain local relative distribution and then utilizing the discrete Fourier transform(DFT)on cycle-shift patterns in histogram to acquire rotation-invariant features.This paper presents a new direction for research of descriptive features extraction in image processing and microorganism detection and provides a useful exploration and experience on analyzing and solving the problem of rotation transformation in pattern recognition.