To overcome lack of biased dataset, the evaluation process was done by shuffling the dataset for ten times in both RGB images and grayscale images and observe the model accuracy.