A New Method Based on Local Binary Gaussian Pattern for Classification of Rat Estrous Cycle Stages Using Smear Images

Abstract

In this study, a unique dataset was created by classifying the images of vaginal smears taken from rats under a microscope for 4 different cycles. Classifying a new case image with the help of this dataset is a computer vision problem. In this study, to improve the weaknesses of the LBP algorithm, a new feature extraction method called Local Binary Gaussian Pattern (LBGP) is developed based on the Gaussian matrix, which helps to remove noise in images. Local Binary Gaussian Pattern proposes a Gaussian-like filter inspired by the Gaussian matrix. After converting the smearing image to the gray histogram, the image features obtained with the help of the Local Binary Pattern (LBP) and our proposed Local Binary Gaussian Pattern (LBGP) feature extractor are combined to obtain features that we call hybrid features. From these features, the ones above a certain threshold value are selected with the help of Neighborhood Component Analysis (NCA), and a Hybrid + Neighborhood Component Analysis (NCA) approach is presented. All hybrid features and hybrid features reduced by Neighborhood Component Analysis were trained with Support Vector Machine (SVM), Decision Trees (DT), Naive Bayes (NB), and k-nearest Neighbors (k-NN) classifiers. According to the classification results, it is seen that the Support Vector Machine (SVM) is effectively classified with the trained classifier. With the Support Vector Machine (SVM) classifier, a success rate of over 90 % (90.25 %) was achieved. Considering the difficulty of classifying smearing images, this result is promising for the future stages of this study.