A New Bag-of-Features Method using Biogeography-based Optimization for Categorization of Histology Images
Raju Pal, Mukesh Saraswat · SSRN Electronic Journal · 2018
The automatic categorization of histology images for medical diagnosis is a prime application area. The bag-of-features method is a popular method for the automatic image categorization. However, due to the complex background structures of histology images, it is a tedious task to quantify such images using the bag-of-features method. Therefore, a new bag-of- features method is introduced to categorize the histology images into their respective classes. In the proposed method, firstly scale invariant feature transform is used to extract the texture features from the images. Then these features are clustered by applying biogeography-based optimization to obtain optimal visual words. Further, the obtained visual words are used to generate histograms of visual words occurrences in the images and these histograms are fed to the classifier for training. To analyze and validate the efficacy of the proposed method two breast cancer histology image datasets have been considered, namely UCSB Bio-Segmentation Benchmark dataset and ICIAR Grand Challenge 2018 dataset. The efficacy of the proposed method is evaluated in terms of overall average accuracy, precision, F1-measure, and recall. The simulation results show the efficacy of the BBO based BOF method over other considered image classification methods.