Image Classifier based on Histogram Matching and Outlier Detection using Hellinger distance

Abstract

In this paper, we developed a prediction model based on histogram matching of Chest X-ray images. Hellinger distance metric is used to match two histograms. The chest x-ray images are pre-processed and converted to histograms. A benchmark histogram is obtained by finding the average of all pixel intensity values. Then outlier images are detected by comparing the histogram of an image with the benchmark histogram using the hellinger metric. Finally, a prediction method is proposed which matches the histogram of unseen images to histograms of nearest neighbor images.  Hypertuning of input parameters to the proposed prediction method is performed to get the best set of parameters. The proposed model gives an accuracy of 92.3 % and F1 score of 94.6 % on the training set, accuracy of 86.2% and F1 score of 89.6% on the test set.