Comparison of Haralick Texture Features and Gray Level Run Length Matrix Features for Analyzing Textural Variation in Cotton Leaves to Identify Spot Disease

The aim of this study is to analyse the textural loss in diseased cotton leaves using Haralick texture features and compare with Gray Level Run Length Matrix (GLRLM) texture features. Materials and Methods: The dataset is taken from the kaggle, having healthy and diseased leaf images. From which 180 healthy leaf images and 178 diseased leaf images are obtained, using G power tool sample size is calculated. Novel texture features extracted from Haralick and GLRLM are classified using SVM, KNN, Random Forest and Logistic Regression classifiers. Results: In this study, Haralick extracts better texture features than GLRLM with an accuracy (95.8%) and F1 (95.8%) using Haralick texture features which is significantly better than the accuracy (66%) and F1 (65%) achieved using GLRLM from SVM classifier. Gpower was used to calculate the sample size and set the effect size (d) to 0.3, the standard error (a) to 0.05, minimum power to 0.8 Conclusion: In this study, it was observed that novel features of Haralick perform significantly better than the GLRLM in detecting spot diseases in cotton leaves.